I entered the stock earlier this year because I do think that BILL may end up being acquired. After all, there’s been a long history of M&A in this space, and even though margins are expanding nicely now, it may take a combination with another party to structurally improve GAAP profitability.

The recovery in the stock over the past few months has been shared across the software and business services space - that rally from an oversold base was mostly a realization that AI is not necessarily going to eat the world of software. As such there’s still a prize ahead if BILL can convince investors that there’s a terminal value worth underwriting - part of this would come from M&A speculation no doubt, as acquirers would also feel more comfortable about BILL’s long term prospects with some additional proof points. My hunch is that the stock will grind up more here as more investors look at what’s going on and consider how things might look like in a year or so.

My bottom line thought for BILL stock here at $49 or so is that it’s worth holding onto, but I won’t be adding for now - I have stocks such as Flywire that I have higher conviction in whereas BILL I would put in the riskier bucket because of its exposure to volatile SMB activity. Obviously, not everything is perfect at the company, but the important thing is that at least in my opinion, a lot of the heavy lifting has been done and so I would be willing to bet that growth starts to be augmented by the actions they have taken. Not immediately but within the year.

As mentioned, you have to take on trust that management’s enthusiasm about their prospects will translate into results - that part is a judgment call each of us has to make. Underlying growth is currently ‘OK’, it’s roughly what Visa is expected to grow at, and frankly, at this point in its lifecycle I would forego higher growth for reasonable growth and higher margins and cash flow. BILL could grow faster if they hired more rather than doing a series of reductions in force, or they could do Superbowl ads and the like - but BILL is maturing, it is focusing on profitability as well as growth and there’s nothing wrong with that. Regardless, the key question is whether the stock is mispriced or not? I think the stock should be higher, in large part because I think there should be more priced into the valuation to account for the probability of a sale.

I think investors who dig deep will likely start to appreciate that good things are happening at BILL. Again, you have to form a judgement as to whether it all makes sense - the reduction in force, the focus on AI enablement, the folding of product silos into more of a broad platform delivery, the pivot up market, the ongoing build out of adjacent products, often AI centric. If you choose to focus on the temporary (and modest) drags in FY27 that don’t appear related to competitive actions as far as I can tell, you may be missing the bigger picture which is that the narrative seems to be changing for the better. Bill is maturing, and seems to be heading in the right direction.

Meanwhile, there is no mistaking what management thinks of the current stock price. The company bought back 14% of the company’s stock since May, including over eight million in the mid-$30s during the June quarter, with more to come.

My May 5th write up “BILL Below $40: Buy the Narrative Normalization” pushed back on the prevailing perception that companies like Bill are under threat from AI. I see more evidence from these fiscal 4Q results that Bill is on the front foot with respect to using AI both internally and in products that help their clients adopt the technology. While revenues have yet to experience any apparent benefit from AI, I think that in combination with their new go to market focus, BILL has a prominent seat at the table for providing a broad backbone of SMB back-office infrastructure. They seem intent on building out their product stack even more, and becoming as AI native as they can get - if that started to work, investors would likely bid up the valuation.

In terms of that valuation, BILL trades at a 35-45% discount on an EV/Sales or EV/gross profit v’s that mid-2025 deal price where closest peer AvidXchange (AVDX) was acquired (this using BILL’s new definition of revenues which excludes rewards gross up, starting in 1Q27). BILL is faster growing than AVDX was at the time of its sale, its margins are certainly higher. Further support could be that AVDX seems to be doing well within CorPay, and that since the AVDX deal Capital One bought Brex for a much higher multiple (Brex was smaller and faster growth, but as we saw with BILL and others, growth always slows). I would say that these developments have if anything increased the strategic value of a company like BILL, and that’s without any assumption that they will achieve the status of a go-to, AI native platform company for the SMB back office.

One ‘administrative point’ worth highlighting - current consensus revenue growth from Koyfin (who get their data from the same place most others get it from - S&P) shows an average fiscal 2027 grow rate that is 2.5% below what the median is. This is because one or more analysts are already using the net of rewards top line which won’t become official until the 1Q27 report. Revenue revisions therefore appear weaker than they really are, as well as consensus growth for 2027. It’s also worth pointing out that the 15.7% ‘core’ (ie excluding float income) revenue growth reported for the year to June 2026 would be reported as about 12.5% when you back out the rewards from revenues.

Why skew positive when the consensus seems to be more mixed?

In no particular order:

1. Management conviction #1. They are buying the stock very aggressively in recent times - I think the prices paid (eg the tranche in 4Q of $300m at $35.31) may turn out to be very additive to shareholder value and will make them look good. But is also signals strong confidence in the outlook. The CFO is even being conservative with respect to the share count per her guide - she includes the ‘as converted’ tranche of near 12m shares from the late 2030 convert that GAAP mandates be included. But these convert at $119 - ie it’s straight debt and won’t be converted and should not be in the count for valuation purposes - and if they are that’s a high-quality problem. The CFO’s diluted share count guide is 104m for fiscal year to June 2027 but the reality is that in economic terms the diluted count will be closer to 86m (and sub 83m in 4Q27) if you treat the convert as debt only and give credit for post June ‘26 buybacks and also assume they execute on the rest of their buyback plan which is my read of what they said on the call (but they did not factor into their share count guide, clearly).

   It’s worth bring up here that these converts of $1.4bn due in late 2030 are effectively interest free debt and when BILL refinances or pays them off there will be an incremental cost. The company earned 3.85% blended rate on their corporate cash last year so it would cost around $54m starting in 2031. That’s almost 10% of 2030 adjusted income - as such it is helpful to be buying back 20% of the stock at recent prices.

2. Management conviction #2. The enthusiasm the CEO has been displaying comes after a period of change at the company, with material RIFs and a reorganization, a new CFO, a new head of sales. Sure, there’s a chance he’s wrong, but at the same time maybe it’s just that BILL has come through a period of uncertainty and headwinds and the path ahead is becoming clearer to him because of the feedback he is getting from clients. SMBs still have the need to modernize their back office, and with BILL being a trusted brand and with an expanded product set, maybe they have a lot to offer, at a time when there’s likely a thrust among clients to improve systems to accommodate the use of AI.

3. Execution and guide philosophy. The company has been executing well lately, with consistent beats even as they are still transitioning internally. Investors seem underwhelmed with the 2027 top line guide but there is 2-3% of specific one time headwinds for 2027 which I’m willing to give them a break on, partly as it may be low margin, and so underlying guide would still be within their low DD - mid teens longer term framework they just articulated. Bear in mind that the median company in the S&P500 grows at around 6-7% over time, so doing low double to mid-teens is premium growth especially if it comes on at expanding margins.

4. I like the pivot up market, including to mid-market and enterprise clients, and I like the go-to-market focus on offering a platform rather than separate products. Even though larger clients take time to win over, they are stickier and reduce cyclicality. It makes sense to me that these changes could result in an increase in growth and revenue stability. It also makes sense that the company would be prudent in guiding in the coming year as these changes take effect and AI capabilities lead to an eventual pickup in demand. Clients likely want to consolidate as much of their workflows onto a single provider as possible, and with AI making it easier to develop software capabilities, BILL is well positioned to take on this role. It may take more than a quarter or two to gain momentum in enterprise and so this has to be looked at as a multi-year thrust.

5. I like the reduction in stock based compensation. There’s likely a handful of reasons behind the fall from 17% of total revenue in 2024 down to 13% in 2026 and 10% in 2027. There’s even a little bit of shift to putting some SBC into capitalized software but it’s still going down nicely. A good part of it is getting past the amortization of grants made at very high stock prices in the Covid era bubble in the stock, but some is also the 30% RIF as those grants are no longer amortized, they are gone.

6. In case investors have not noticed, this is a consolidating space. Consider AvidXchange, Melio, Brex among others. A stock like BILL ought to be priced as if there is a meaningful probability that they will be acquired and that has not been the case lately. While not the only reason that I am in the stock, M&A is a constant in this space.

   If BILL can demonstrate that AI is helping it win new clients and improve monetization of existing ones, buyers will take notice. Potential buyers being mentioned include Intuit, Amex, and Coupa which is owned by Toma Bravo at the moment, or even Corpay which acquired peer AvidXchange in 2025.

   I have done the math around BILL’s M&A setup but won’t reproduce here - after all its quite speculative and also quite subjective, but it’s easy to do should you feel inclined.

   Using the valuation that AVDX was acquired for (and using revenues net of rewards for BILL), I get a potential takeout in the low $70s. That’s a reasonable place to start discussions if you are BILL, as they have higher margins than Avid, and Avid sold at a time when growth was flatlined, perhaps temporarily but still. BILL has been executing well lately and it has more breadth than Avid and that’s worth something.

   When you factor in some top and bottom line synergies that might come along with a more strategic type deal rather than a pure PE acquisition, you can get to $80-85 and that’s today. If BILL starts to demonstrate growth solidly in the mid teens as a result of its efforts to build out AI capabilities while broadening its platform even more, then you could be more aggressive and perhaps get closer to $90-100 - but that would also include an additional year or so of growth on top of the $70 which is more based on current fiscal year numbers.

At the end of the day, you can crunch numbers ‘till the cows come home but what really matters for BILL is whether it gets to a place where:

• There’s no questioning its terminal value due to threats from AI etc

• They can broaden their product set and successfully pivot up market to make themselves less susceptible to the vagaries of the small business cycle.

• They can show innovation in product design to match the likes of Brex, Ramp etc which are valued at massive multiples compared to BILL (faster growth but off lower bases for the most part)

• They can achieve reasonable growth ambitions while continued to expand margins and also keep stock based comp under control.

BILL is guiding for 10-14% core revenue growth net of rewards this year inclusive of a total 3% headwind - if that is indeed temporary without much if any follow through in guide for next year fisc al through (June 2028) AND if the new go-to-market approach is successful or certainly not a failure, AND if they can point to incremental growth from their AI efforts, then the stock could do very well. I say that because then it would be more firmly into the rule of 40 category as a standalone investment. Of course, they still have to contend with anything the economy throws at them as this is still a stock that’s more exposed to SMBs than most. That’s partly why I think it might be better off within a larger entity - investors might be inclined to continue to fret over the macro side of things, and even though BILL will become more balanced over time, it’ll be helpful to get there sooner.

Standalone valuations for BILL and others (RELY, FLYW, DLO, in fintech are very attractive. Looking into calendar 2027 and 2028, the adjusted EV/EBITDA are sub 10x and indeed for 2028 they are all between 6-8x. This is at a time when stock based comp add back is at 10% of revenues for BILL for example but declining each year. It’s at a time when free cash flow is ramping nicely such that these companies can buy back stock at single digit multiples, retiring shares even after stock grants. It’s at a time when most of them have accumulated a history of beating estimates. And it’s at a time when they are all growing much faster than the typical market growth. And it’s at a time when M&A activity is picking up. This is why I like this space - it’s fundamentally sound and investors are way behind the curve when it comes to assigning appropriate valuations for them.

Flywire: The Current Visa Noise Amounts to a ‘Net Nothing’ Impact on Deeper Analysis. The TAM Is Worth Billions.

Short interest is 11.5% on soft early visa data. A census of 600+ universities across the Big Four says the shorts are fighting the last war — the real story is how much share is still on the table.

The visa scare driving Flywire’s recent underperformance nets out, on my math, to roughly a £48,000 swing in 2026 UK revenue — the cost of a new car. Meanwhile my university-by-university census of the Big Four markets shows Flywire named on 95% of top-of-market US international fee volume, a near-monopoly in the UK, and a domestic US opportunity roughly 4x the size of the cross-border base it has already won. That asymmetry is the note.

Visas first, briefly: the math says stand down

Student visa applications have started 2026 weak in the UK and the headlines write themselves. But not all visas are created equal, and the weakness is concentrated where Flywire’s revenue isn’t.

The UK shortfall sits almost entirely in one-year postgraduate courses drawn from Pakistan, Nigeria and Bangladesh — roughly 17.7% of last year’s visas — after a handful of tier-three colleges paused recruitment to cure high refusal rates. Those students pay roughly £15,500 in tuition. High-fee undergraduate applications are up about 5% overall, with Chinese students (40% of tier-one undergrads) up 12% YTD, at an average ticket near £32,000. Stress-test it: assume the affected cohort falls 50% (extreme) and the undergrad gain holds, apply a 1.5% take rate, and the net TPV swing costs Flywire about £48,000 of revenue. Flywire is also over-indexed to the high-fee institutions where the weakness isn’t. Full workings are in the appendix of the data file below.

The other markets are quicker still. Canada is 4% of revenues, down from ~14% at the 2023 peak, and largely washed out; even admissions down 20% instead of 10% moves 2026 revenue by about $1m on a $750m+ base. Australia (6% of revenues) opened the year guided down 30% and the intake came in up low teens, driven by wins. US cross-border (10–11% of revenues) was guided assuming visas down 30%; ApplyBoard has undergrad applications down only 9% through March, and the missing students skew to lower-ticket regional programs.

Then there’s the offset most bears miss entirely: the 14% of revenue from non-Big Four education grew 40% y/y in 1Q26 and is taking 60% of new client wins. Students turned away from Canada and the UK are showing up in Germany, Spain and Italy — often paying through Flywire, which in Germany’s case processes the mandatory full-year living-expense remittance before a visa is even granted. I model ROW at 30% growth for the full year. Aggregate global student mobility remains at all-time highs; this is a mix shift, not a secular decline. We’ll know more on August 4th.

The census: quantifying what management only asserts

Flywire has always claimed lots of headroom without quantifying it. So I quantified it — a detailed provider census of the top ~300 universities and colleges across the US, UK, Canada and Australia: Flywire penetration, product attribution, and the incumbent provider for both international and domestic payments at each institution.

Flywire Exposure to Top Global Universities

297KB ∙ XLSX file

Download

College by college data showing mix of students, tuition, who servicing cross border and domestic, where the penetration lies, what the share opportunities are.

Download

The share-gain mechanism, in the CEO’s words on the Cornell win: clients are consolidating fragmented financial workflows onto a single platform, and Cornell is moving billing, payments, payment plans, refunds and collections onto one system “that only Flywire can provide.” Keep that consolidation logic in mind — it recurs in every market below.

US cross-border: breadth won, depth is the battleground

Across the top 157 US institutions, Flywire is named on 95.3% of international fee volume and present at 133 of 166 institutions. The land grab at the top of the US market is over — Flywire won it.

Depth is thinner than breadth. Flywire is sole or lead provider on only 38.2% of volume ($9.6B) and exclusive on 25.8% ($6.5B). Roughly 60% of the volume where Flywire is present is shared shelf space. New-logo whitespace is narrow (~20% of institutions, skewing smaller), so US cross-border growth from here is a share-of-checkout fight inside existing logos: converting the 26 primary-but-shared relationships to exclusive and taking wallet share in the 37 shared accounts.

US domestic: the 4x prize, and the incumbent is the school itself

The census counts $96.8B of domestic tuition volume against ~$25B international — roughly 4x. Flywire’s domestic footprint today is 19.2% of that volume, and the confirmed full dual-mandate piece is only 5.4%. Management’s own ~10% SFS penetration claim checks out against my data. This is the whitespace; cross-border is the beachhead.

The economics of conversion are the point. Flywire has said a full domestic student-financial relationship adds 3–5x the revenue and 2–3x the gross profit of the cross-border relationship alone. Penn State is the live example: Morgan Stanley pegs the new SFS relationship at ~$12m of revenue on top of $4m of pre-existing cross-border — call it ~$7.5m of added gross profit, much of it at high incremental EBITDA margin. Directionally right, though accounts this size likely price below what a smaller institution would.

Here’s the finding that reframes the domestic thesis: the incumbent to displace usually isn’t a competitor. In-house, self-run portals control 54.9% of domestic volume ($48.0B across 73 colleges) where Flywire already owns the international relationship. This is a build-vs-buy conversion story before it’s a competitive takeaway story. Among named vendors, the displacement targets rank Nelnet ($14.7B), Transact/CASHNet ($7.1B), TouchNet ($6.5B) and Workday ($3.6B) — and Flywire SFS has already converted 19 such relationships worth $12.9B. Penn State, Cornell and Calgary within a few quarters of each other suggest the enterprise conversion pace just stepped up, and from what I see in the data, the competition is not prepared for it.

UK: a near-monopoly that still has room

The UK is ~22% of my 2026 revenue estimate, roughly 80% of it cross-border. Flywire is named at 99 of 137 UK universities covering 79.8% of international fee volume (£10.6B of £13.3B) — volume penetration running ahead of logo penetration, meaning Flywire skews to the bigger, higher-fee institutions. Over half the census has Flywire as sole or lead provider. Only 12 accounts are 90%+ penetrated, per management, so most of the base still has share to give.

The attackable whitespace is real: 25 universities with a competitor and no Flywire, worth £2.5B of volume — and the count share matching the volume share tells you these are average-sized schools, not scraps. Most of it is Convera’s to lose.

Two accelerants. First, ERP integration work just completed (Unit4/Agresso live, Oracle Fusion certified) opens roughly a third of the market that was technically closed to Flywire; Unit4 alone is a 50+ pipeline, and the four beta clients going live should start getting named on upcoming calls. Second, the FT reports UK international tuition fees rising 29% over four years from 2025 — inflation that drops almost straight through Flywire’s take rate at very high incremental margin. I’m not fully factoring either in.

The 2025–26 win roster (Greenwich, Bristol end-to-end conversion, Heriot-Watt, Royal Holloway, Birmingham City, Huddersfield, Kent, Oxbridge college additions) is in the data file.

Canada and Australia: Calgary is the template

Canada is only 4% of revenues, but it’s the cleanest read on what happens next everywhere. Flywire has sub-20% penetration of Tier 1 Canadian institutions — a share of Toronto and nothing at the next nine — because it arrived as the newcomer, over-indexed to the immigration-driven colleges that got decimated. The Calgary win, a full international-plus-domestic mandate taken from the incumbent Convera/CIBC duopoly, is the same consolidation motion as Penn State and Cornell running in a market where Flywire’s starting share is low.

Australia (6% of revenues) has the same structure: Convera dominant, little dual-mandating, and headline visa volatility concentrated in vocational colleges that don’t appear in my census of core universities. The core institutions are high-ticket and stable. The underlying reason share keeps moving: Convera’s front end is fine, but the back end is legacy tech, while Flywire is cloud-native, modular, and integrates into the ERP and the recruitment workflow. With Penn State as the reference transition, switching-cost objections get harder to sustain. I would not be surprised to see Flywire land an Australian whale or two in the next year.

Everything else is going right

Travel (the second-largest segment) has no Middle East exposure, an inelastic luxury end-client, and Sertifi synergies that only started executing in 1Q26 and don’t appear to be in consensus. Healthcare and B2B (mid-single-digits of revenue each) are both having strong years that carry into 2027. My January travel bridge stands — it’s one reason I’ve been ahead of consensus.

Margins: the investment is already spent

Incremental EBITDA margins are running 35–40%, and that’s with the platform build-out, Sertifi integration spend, domestic SFS product development and the healthcare ramp already inside the number. Adjusted EBITDA margin rose ~350bps to 20% in 2025 and the CFO has pointed to ~25% for 2027. Run the arithmetic forward: a 20% grower adding new revenue at 40% incremental margin adds ~3 points to consolidated margin in 2028 alone — against a consensus 2028 margin of only 26%. Both the top line and the margin line in consensus look low to me. Mid-40s incremental margins in due course would not surprise me.

SBC is ~10% of 2026 revenue — not great, not SaaS-egregious — diluted shares fell last year on buybacks, and I peg SBC at 6–7% of revenue by 2030.

Valuation: $38 fair value, $30 year-end target

The stock sits roughly where it was before the February 2025 guide-down and Sertifi selloff, while the market is up 25–30% and Flywire has outgrown it: 17% organic in 2025 through an adverse macro backdrop, 30% organic in 1Q26, margins up ~400bps since the start of 2025.

My framework: grow the business to more mature margins (mid-30s adjusted EBITDA, and that’s not full maturity), discount at 10.5%, apply 14x EV/EBITDA — the multiple of a mature, non-consumer payments compounder like Corpay — on earnings including stock comp and a diluted share count grown to ~149m, i.e., giving no credit for cheap buybacks. That yields $38 today. My official year-end target is $30, conceding something to group sentiment.

The company is also financially unrisky in a way the multiple ignores: I have cash building toward $1.8B by 2032 absent M&A and buybacks, retention is exceptionally high, and the regulated-network, deep-integration moat is not the kind AI disruption attacks.

The Roper wrinkle

One speculative closer (tagged as such). Penn State and Cornell were competitive takeaways from Roper’s education subsidiary — Roper has spent ~$7B on its education vertical, including Transact/CASHNet, now Illumia. My census shows exactly these incumbents sitting on universities that need to consolidate and upgrade, with no real cross-border capability of their own. International students are the hardest and most lucrative piece; whoever owns that flow wins the consolidation. A Roper paying 14x-to-high-teens post-synergy EV/EBITDA for slower-growing education assets can do the math on what Flywire will cost it in lost business. I don’t think Flywire sells — management can see the upside in its own stock — but the market pricing this possibility at zero says something about the market, not the asset.

What to watch

August 4th earnings: UK peak-season commentary, whether the -30% US visa guide gets walked up, ROW growth against my 30% full-year model, and naming of the Unit4/Oracle beta clients. A likely investor day this fall, plausibly timed to declare the visa headwind over. And the enterprise win cadence — another Penn State-class SFS mandate or an Australian Tier 1 would confirm the consolidation thesis; two quiet quarters on enterprise wins would be the first real crack in it.

This is my largest position and a multi-year holding. The debate resolves not on visa counts but on whether the Calgary-Cornell-Penn State pattern is a trend — and August 4th is the next data point.

International Student Volumes in 2026

My work on visa topic says that at this stage the visa volatility (in the UK mainly) is not material. The math shows why - the weakness is in one year post grad visas that would carry tuition that is often one third that of an undergrad at a prestigious university. High tuition paying Chinese undergrad applications are up 12% YTD in the UK, this offsets lower volumes from Pakistan and a couple of other countries to a good degree (even accounting for UG being 1/3 of student admissions in the UK). And regardless, I will show in the appendix just how much tailwind the company has from new UK wins.

Canada no longer moves the needle at 4% of revenues (and has some other drivers this year), Australia is already past its peak intake season, and the US is tracking better than the company’s -30% (conservative) initial guide (ApplyBoard UG applications down 9% through March, and anecdotal evidence suggesting post grads would be down a more, maybe 15%+ but coming from lower ticket regional colleges so not as impactful on TPV).

Another dynamic is the 40% growth in 1Q26 from ROW countries, which at 14% of 2025 is material. I model 30% for the full year. All the moves above cancel each other out at a minimum - visa count is not a reason to be short.

However what’s important to realize is the momentum in client wins, the product ramps, and even tuition inflation which is a material tailwind especially in the UK (not that I fully factor it in). I add an appendix for those looking for more detail on this visa topic - but as I said, we’ll know on August 4th.

I will finish with one observation about international student volumes in general. There is no secular decline going on. Rather, there is a shift away from the Big Four, and it’s coming from lower-ticket, price-sensitive, employment-dependent students from select countries. The number of students globally has not declined versus the peak of 2023, and Flywire is engaging with more and more of them in every country. Finally, one could argue that 2026 will mark the low point for even these types of students engaging with the Big 4 countries, simply because of how far that cohort has fallen in the mix.

Flywire’s structural advantages in education are becoming more clear

I have been able to do a lot more detailed work courtesy of Claude and its ability to extract information efficiently. This has enabled me to quantify Flywire’s education exposures in a granular level in all the top markets, showing who the competition is, what the penetration is, and it makes the growth avenues clear. In short, Flywire has a lot of share still to capture, they are on a roll in terms of client wins, and structurally I think that roll is going to continue based on what I see in the data, and because recent wins at marquee accounts tells its own story.

Before we dive into the detail of the data, here is what CEO Mike Massaro said on the 1Q26 call about the recent Cornell win, as it explains the share gain dynamic from the college’s standpoint.

“Let me start with vendor consolidation. Clients are choosing to consolidate fragmented financial workflows onto a single trusted platform. We are leveraging this dynamic across our verticals and the reason we win is that we are the only platform that can handle all the complex workflows they need.”

“Cornell is a large institution, tens of thousands of students, significant national enrollment, multiple funding sources, including sponsor billing and loan disbursements and a collections operation that touches separate debt types simultaneously. They are consolidating their billing, payments, payment plans, refunds and collection processes onto a unified global platform that only Flywire can provide. This reduces the complexity and cost of managing multiple fragmented vendors while giving Cornell a simpler, more automated and uniform view of their student financial activity.”

The Data

For those who like to dig deep, here is an Excel file with a lot of Flywire related data you will not have seen in one place before.

Flywire's exposure to top global universities

298KB ∙ XLSX file

Download

Detailed list of the top 300 or so universities and colleges in the Big Four with associated Flywire penetration and product suite attribution. Lists the incumbent payments provider for international, domestic, for each College. Analysis tabs show dollar penetration and white space.

Download

Summary of findings:

USA Cross Border - Flywire has already landed here, and now they are expanding.

Flywire has 1,000 cross-border clients in the U.S. I have focused on the top 157 universities and colleges, as these likely form the bulk of tuition and should be a fairly accurate reference for their share. US cross-border is about 10-11% of total revenues.

• Flywire is named on 95.3% of international fee volume in the census and at 133 of 166 institutions (80.1%) — at the top of the US market.

• Depth is thinner than breadth: Flywire is sole/lead on only 38.2% of volume ($9.6B) and exclusive on 25.8% ($6.5B). Roughly 60% of the volume where Flywire is present is shared shelf space with competing providers — the real battleground is share-of-checkout within existing logos, not new logo wins.

• Institution-count whitespace is narrow: 16 schools (9.6%) have an intl provider but no Flywire, plus 17 (10.2%) with no named provider/TBV. Combined, ~20% of institutions remain a realistic US new-logo pipeline, but they skew smaller by volume.

• Net read: US top-of-market growth for FLYW has to come from wallet share within the 37 shared accounts (15.7% of institutions) and converting the 26 primary-but-shared relationships to exclusive.

Domestic US - up for the taking.

Flywire has articulated how taking on the full domestic student-paying experience adds 3 to 5 times the revenue and 2 to 3 times the gross profit of a standalone cross-border relationship with the college. A recent example is the Penn State win. Morgan Stanley analysis suggests that the new SFS relationship with Penn State will add $12 million to revenues on top of the $4 million pre-existing cross-border revenue. So this likely adds about $7.5 million to gross profit, a lot of which will come on at high incremental EBITDA margin. I believe Morgan Stanley’s work is directionally correct, but it might be the case that larger accounts like this come on at somewhat lower margin versus a smaller institution.

U.S. domestic revenue is about 6% of total revenues, with a somewhat lower gross margin profile. It is growing rapidly.

Flywire tells us that they are around 10% penetrated in terms of their SFS product in the US. This data I have skews to the large 157 accounts, and it depends on how you define the domestic relationship as you see in the data. The bottom line is that there is a lot of upside.

• The domestic prize dwarfs the intl base: $96.8B of domestic tuition volume in the census vs. ~$25B intl — roughly 4x. Flywire’s total domestic footprint today is just 19.2% of volume ($18.6B) across 27 schools (16.3%), and the confirmed dual-mandate piece is only 5.4% ($5.2B at 10 colleges). Domestic is the whitespace; intl is the beachhead.

• Even that 19.2% overstates the win: more than half of it ($9.8B, 10.1%) is partial inroads — domestic wires, 529s, payment plans, collections — not full SFS mandates. The full dual-mandate number worth anchoring on is 5.4–9.2% of volume depending on how you treat the “likely/unconfirmed” $3.7B.

• The incumbent to displace isn’t a competitor — it’s the school itself. In-house/self-run portals control 54.9% of domestic volume ($48.0B at 73 colleges) where Flywire already owns intl. The domestic upsell thesis is a build-vs-buy conversion story before it’s a competitive takeaway story.

• Among named vendors, the displacement targets rank Nelnet ($14.7B, 22 schools), Transact/CASHNet ($7.1B, 15), TouchNet ($6.5B, 12), Workday ($3.6B, 10). Flywire SFS has already converted 19 of these relationships ($12.9B, 14.3%) — proof the intl-to-domestic land-and-expand motion works, but the pace of that conversion is a question.

• I think the answer can be found in the recent capture of Penn State, Cornell, and also the Calgary win in Canada. This suggests a major step up in convincing enterprise clients to consolidate vendors and upgrade their technology. These themes are very much in evidence and from what I can see, the competition is not prepared.

The UK - Flywire the Conqueror.

Partially through acquisitions done a few years ago, Flywire is a virtual monopoly in the UK in some respects. Although some accounts may be mature (they said only 12 accounts were 90% penetrated), I agree with management that most of them have upside share capture available. Secular drivers are putting this business into Flywire’s hands. By this, I mean the themes of vendor consolidation, systems modernization and integration, and exclusivity. The goal is to have most of the payments made through one vendor, which leads to less confusion, easier reconciliation, and a better all-round experience for students and employees.

UK revenue is around 22% or so of 26 revenues, per my numbers. And 80% of that is cross-border, with the rest domestic and SaaS fees, is my estimate.

• Flywire is named at 99 of 137 UK universities (72.3%) covering 79.8% of intl fee volume (£10.6B of £13.3B) — volume penetration outrunning logo penetration means Flywire skews toward the bigger, higher-fee institutions.

• Depth of position is strong where won: 29 exclusive (21.2%) plus 42 primary (30.7%) means over half the census has Flywire as sole or lead provider, driving the 50.4% sole/lead volume share (£6.7B). Only 25 schools (18.2%) are shared- the UK is much less of a multi-provider market than the US.

• Whitespace is symmetrical and material: 25 universities (18.2%) with a competitor and no Flywire, worth £2.5B (18.5%) of volume. That the count share and volume share match tells you the competitor-held schools are average-sized, not scraps, this is real, attackable volume, mostly Convera’s to lose.

• The tier structure frames where the money is: 26 Tier 1 high-fee unis (£28–43k) hold 247k intl heads — 41% of the 602k total heads at 2–3x the fee anchor of Tier 3. To the extent there is weakness in visa count in the UK, it is mainly a tier three phenomenon, such that the dollar impact is less than headline units.

In the UK Flywire operates what’s called a “Onedoor” model, which processes cross-border, domestic, including accommodation, and they get a SaaS fee. I peg cross-border at about 80% of revenues, SaaS about 5%. I haven’t broken out Flywire’s SFS penetration because they’ve only just started a year or so ago. I think this is a major opportunity to add high-quality revenue made up of several modules that lend themselves to increasing the penetration of all aspects of the client.

One other thing of significance in the UK is the recent completion of systems integration work that will free up a third of the market they were not able to penetrate. We should be hearing more about this on the next call or two, I would assume.

As you can see while there are pockets of the UK that are mature, the overall market has a lot of growth potential in terms of consolidating existing clients and adding new ones from the new integrations and the new products they’re rolling out.

A lot of the work has been done, and the company must be quite efficient in the UK, given their concentration. I would assume the incremental margins on the new revenues are going to be very high.

Canada - only 4% of revenues but…

Canada may have been a low teens revenue mix client at the 2023 peak, but it is now down to being about 4% of revenues. Canada is one of the countries where, alongside Australia, Flywire was the newcomer. As such, they were more exposed to the newer colleges, which is why the revenue hit there was so severe. They did outperform the market, probably by 20% each year on account of organic wins, and I guess the good news is that Canada looks pretty washed out right now. That said, you can often hear management talk about how much upside there is in terms of share gains, and the data supports that.

There’s more detail in the attached file, but the thing that stands out in the above is the sub-20% penetration of the Tier 1 Canadian institutions. The modest 24% overall penetration of Canada on the international cross-border side leaves quite a bit of room to grow.

What particularly stands out is they only have a share of the top college, Toronto, but nothing of the next nine in terms of importance. Calgary is number 10 and is a recent SFS (full service intl and domestic) win against the incumbent players in the market, Convera and CIBC. This, along with the likes of Penn State and Cornell in the US, could be the start of a trend, no?

Australia - Similar to Canada structurally, top accounts ripe for the picking

I peg Australia at 6% of total revenues in 2026.

Australia has some similarities to Canada in terms of its market structure. Convera is quite dominant there in terms of share of tuition, and it seems to be less dual-mandated there.

One thing to keep in mind with Australia is that headline visa volatility skews toward vocational colleges, which are not included above. These core universities above tend to be a lot more stable and also high-ticket, which makes the market more stable than it might appear from the data at times.

Flywire started out with a negative 30% guide for Canada for 2015 because they feared something similar to Canada was possible. In the event, it was up low teens. One of the reasons for that is increased wins, and I think management seemed very upbeat about prospects for continued share gains.

Again, going back to the core reason this share gain is happening: the likes of Convera may have a decent frontend, but they are legacy tech at the backend. Flywire is cloud-native, can easily add on modules, help with real-time reconciliations, and integrate with a frontend agent and other communication avenues that help in the recruitment process.

Flywire now has a documented track record showing just how much they can save colleges in time and actual dollars. Transitioning the likes of Penn State will be a great show piece if there is any pushback in terms of switching cost friction.

I think we’re going to see continued wins coming out of Australia, and I would not be surprised to see them land a whale or two in the next year.

Conclusion

Flywire has rarely quantified in detail what its penetration is in various countries, except to say that it believes it has a lot of headroom to grow. The nearest thing it told us is that in the UK, only 12 clients are at 90% penetration. If you take the time to look at the actual numbers that I’ve put together (I know it’s a lot), you can certainly see that they are being truthful about their opportunity.

What is exciting as someone who sees this stock as a multi year holding, potentially, is that the enterprise wins they are seeing send a strong message to the rest of the market that they are able to bring a best-in-class product and also integrate it safely. This is all made easier because Flywire seems to take a very thoughtful approach to building out its tech stack, both internally and for clients.

I don’t think I have to go out on a limb here and give an exact number for the revenue upside, except to say directionally there’s lots of it available over time. I have witnessed the company investing a lot in product development and integration work. One thing I can say is that when this revenue comes on stream, it’s going to come on at very attractive EBITA margins. Incremental EBITDA margins have hovered around 35% to 40%, but arguably that includes a lot of this investment. It wouldn’t surprise me to see that number in the mid-40s in due course. That’ll make a big difference, given that they were aspiring to mid-twenties adjusted EBITDA margins for next year.

Channel Checks

As far as the channel checks and the outlook - to summarize, I think the company will keep surprising on the upside this year. Some investors (short interest is up) are latching onto early data from particular markets that show a weaker start to certain student arrivals. While the company may acknowledge some added uncertainty on the margin in one or two end market, I think these investors are likely missing material mix components, missing the revenues from new wins, and likely misses Flywire’s expanded diversity. For example the 14% of revenues that is non-Big 4 education grew 40% y-y in 1Q26. These students increasingly attending non-Big 4 (“ROW”) colleges are often coming out of the Big 4, they are not coming from Mars colonies just yet. And so you can’t really look at one market in isolation anymore.

Recent background as regards student visas

Before going into Flywire international student exposure let’s put it in context - when you sum the education revenues and isolate the known and likely domestic component, the actual cross border part is around half or less of total revenues. The rest is travel, B2B, Healthcare and then the domestic sourced portion of education, chiefly in the US and UK.

It is increasingly diversified, so for example Canada is only about 4% of revenues, Australia is only 6% of revenues, the cross border portion of the US is only 10%-11% of revenues. The cross-border part of the UK is about 17-18%, and you can assume the 14% rest of the world is all cross-border.

Growth in this rest of world segment actually accelerated to 40% in 1Q26 and is accounting for the bulk (60%) of new client wins, so it will likely be at or above 15% of 2026 revs. I am modeling 30% growth in the rest of world segment for the full year.

The aggregate international student count across the globe continues to be at all time high levels despite the mix changes. There is no secular decline in student mobility, it is just shifting, and projections anticipate further growth in the years ahead.

The UK

Summary - the UK is a mixed bag in 2026, but overall growth should be fine. This may come as a surprise to those who have looked at early stage visa numbers (in isolation), but there’s some nuance to it, so I will show how little impact it ends up having. Before doing that, I will summarize some of the organic things happening in the UK.

On the 1Q26 call the CFO said “We also saw some strong domestic performance within the U.K. where we continue seeing strong growth.”

Here is a list of some new wins live or ramping in 2025

Go lives / ramping

• Greenwich — 2025 RFP win, full-suite

• Cumbria / UWE Bristol — full-suite, signed late 2025

• Bristol — software-only converted to end-to-end, live pre-Q3 2025 peak

• Heriot-Watt / Royal Holloway — Q3 2025, ~39k students / ~11k international

• Birmingham City, Huddersfield, Kent — Q1 2026 roster

• Oxford / Cambridge — added colleges, payables and refunds cross-sell

• Birkbeck / De Montfort — StudyLink, application fees and deposits

• US federal loan disbursement — 15 UK universities since launch, seven added Q3 2025

ERP integrations. The UK is roughly two-thirds SIS, one-third GL. The company has been building integrations into two or three ERPs that were outstanding so to speak, such as Unit4 and Oracle. Now that the four beta clients are going live, they should likely be named and add to revenue. Plus, it opens up one-third of the market that was closed to them. Unit4 alone is a 50+ pipeline.

• Unit4 / Agresso — real-time bi-directional GL live late 2025 with three as yet unnamed development partners

• Oracle Fusion — connector certified 2025, first UK go-lives FY26

• Tribal SITS — deeper workflow certification 2025, share gains at existing accounts

I will add this as a medium-term driver because I am not factoring it in fully. The Financial Times reported that UK international student tuition fees are set to rise by 29% over the four years from 2025. Flywire’s model is built to scale, and such inflation will largely drop to its bottom line. The margin outlook looks very strong regardless, but exposure to such inflation will result in even high incremental margins than they are currently reporting.

2026 visa volumes: how much of a risk is it?

This year, student visa volumes, as evidenced by visa applications, are after a week start. Versus an unusually strong first half of ‘25, they were down 30%, although relative to ‘24, they were down less than 17%. The June 26 number was down about 18%, so some recovery is going on as we approach peak season. First half applications normally account for only 21-22% of total year, and they have a history of being quite volatile, given the low numbers.

The issue is entirely related to material weakness in lower-ticket postgraduate one-year courses. This is a result of a drop in the availability of places for Pakistani, Nigerian and Bangladeshi origin students, because of historically high visa refusal rates, which colleges are now responsible for keeping under control starting this year.

This is because a handful of tier three colleges needed to pause recruitment from these markets in order to cure their historically lagging visa refusal rates. As such, they likely won’t be resumed until late this year, at best, and these markets will continue to face scrutiny.

That said, international undergraduate applications are up about 5% overall, with Chinese students (40% of top-tier undergrads in 2025) up 12% so far. As you’ll see, this is actually quite important, because your typical Chinese student (who are 40% of undergrads at the tier 1s) will pay about three times the tuition of a typical tier three student from one of those listen countries.

I can give you some fancy math on my spreadsheet here. The important column is on the right, which shows the modest impact on total Flywire revenues, even assuming an 11% decline in total headcount. Tier 3 and maybe a little bit of tier two are the ones that are affected. Undergraduate who tend to go to tier one: that cohort is up 7% in the first half in terms of applications, it’s up 12% among the Chinese students who tend to pay top dollar.

I can explain it in a more wordy way. Last year there were 427,000 new student admissions, about 35% of which were undergrads, and that’s up 7% so far this year. I would give you an extra 10,000 or 11,000 students.

The postgrads that are affected are really from three or so countries, which comprised 17.7% of visas last year. If you assume that those were down 50% (which is quite extreme, by the way), that would cost almost 25,000 students.

The tuition for the undergrads, particularly given the China weight, would be about £32,000 a year, versus about £15,500 for the cohort that has gone AWOL. One more factor is that Flywire is over-indexed to the high fee and slightly under-indexed to the low-fee colleges.

When I do the math on that net TPV variance at a 1.5% take rate, I get a negative £48,000 swing. As in the cost of a new car. It could be a bit more because maybe there’s an extra £6,000, or £8,000 of living expenses that Flywire may process. It depends on whether they live in the dorms or not. But it’s not gonna make much difference.

The fact is that if you look at what’s on tap for the UK in terms of new win related revenue, I’m pretty sure I’m underestimating what that is. In any case, let’s go back to that rest of the world 40% growth in 1Q26. That is where a portion of these people are going. Flywire may not capture every student at the same price point even compared to the low tier, but it’s enough to make a difference. And you certainly shouldn’t be looking at one market in isolation these days because it’s all fluid.

Here is my revenue bridge for the UK. As you see, I am assuming some headwind from new students, just in case. And the growth at 15% is a bit less than what I would put in at the start of the year. One positive aspect is that I’m fairly sure 2027 won’t have the same issues with those countries. Right now, it’s confusing for the students and the colleges that are affected, but things will probably normalize a little bit more next year.

The US

I’m going to keep this short because I’ve touched on it above. The company guided the year with visas down 30% in the US, and even after Q1, which was very strong, they maintained that guide. I think one of the reasons Q1 was strong was that a lot of deferred students arrived in the US a bit late, having secured their visas late in the year. They started in January.

There’s a lot of noise in the US, but students are still drawn to it, particularly at the high end. The US never had a severe problem as regards students trying to get immigration or asylum, and in fact, most of the Chinese go home, and there are only 85,000 H-1B visas. It’s very hard to work in the US as a professional without a visa.

Per ApplyBoard, undergraduate visas are down 9%. As with the common theme elsewhere in the Big Four, it’s not the high-ticket students who are missing. It’s the low-ticket ones.

For the one or two people still with me here, here is my 26 revenue bridge for us cross-border. I have it up 6% as I assume visas down 15%. If it was -30%, the headwind from new students would be negative $10 million rather than negative $5 million. But in the context of a company earning over $750 million in ‘26, I think people should chill out.

Canada

Investors should not worry too much about Canada or other mature markets losing some students here and there because Flywire is capturing them wherever they go. Canada is only 4% of revenues - does it even matter for the investment case?

Regardless, Canada serves as an example of what is happening in many territories, although it was by far the extreme case in terms of the drop in student numbers - most of them were there for immigration first and foremost.

An important aspect of any changes in student volume/visas is this - they are not created equal, and that statement, which I expand on below, is why the damage is not as severe as the unit count. On a dollar weighted basis, the student that has not applied, or has been refused a study visa in Canada, was likely paying 1/3 the tuition of what I would call the core. The core are the students at the established universities, who attend primarily for academic reasons. These are the students who don’t tend to get refused, and they are the ones who attend the colleges that add value to their resumes.

It just happens that countries such as Germany, Spain, Italy and many others are happy to accept many who would have gone to Canada (and the UK). There are some differences in economics but likely not as much as some may think because for example in German, before getting a visa, students are required to remit a full year of living expenses, often but exclusively, via Flywire, to a special account. For Flywire, it’s not exactly a decline, it’s a shift, certainly in 2026. How else can you account for a 40% increase in the 14% of revenues that is ROW (some is new logos of course).

Let’s touch on Canada’s modest numbers quickly - because some of the themes here carry over to other Big 4 countries.

In 2025, the company saw revenues down almost 30% against a backdrop of a decline in visas / arrivals of between 50% and 60%. They earned about US$26m in revenues, so 4% of total. I think Canada used to be 14% or so of total in 2023.

They are gaining share which I will expand on because there’s a good trend at hand here that applies across all markets. That’s why they outperformed the market every year in Canada. However Flywire in 2023 was quite exposed to the types of colleges that got decimated by the visa changes. They were the newcomer and were over-indexed to that type of client. Now ironically, from the standpoint of where we are today, that’s a good thing because as I will show, they have a lot of upside from share gains in the upper part of the market because that market seem ripe for it - the Calgary win speaks volumes as to what is about to happen in this space in Canada and elsewhere. Flywire’s cloud-native, modular offering for international and domestic resonates well with clients who have been stuck with adequate but outdated technology from the likes of Convera and CIBC. It remains to be seen how sticky those relationships will prove to be.

Back to the numbers. They won three good sized clients which onboarded either in 2H25 or in the case of the largest (Calgary), 1Q26. Along with anticipated wins I peg that share gain revenue at $5m for ‘26, so almost 20% growth alone from wins.

Then there’s a timing effect first mentioned on the call where they took numbers down in 1Q25. It impacted 2025 guide by about $5m+, but would be recaptured in 2026. I am using $2m to account for lower student numbers.

Then there is the drag when you consider both the prior cohort effects as students from 2023 graduate, and then new student numbers move around. Bear in mind that many students who came on board in 2023, 2024 have a local bank account now and do not use Flywire - although that said many colleges are either encouraging or requiring payments, both domestic and international to go through Flywire - because it more fees to be more easily reconciled into the ERP. The whole point of using Flywire’s technology is to automate as much as possible and save the college money.

But in Canada in particular, because of the aforementioned over-indexing to lower quality one year postgraduate courses, most of the cohort from 2023-24 are gone, there is not much cohort drag. So my math says its only a $1m drag from the cohort effect.

At then the new visas, I have those down -10%, in line with the company. Here’s the important thing most people will miss - when the likes of Canada see a 10% decline in student visas at this stage of the game, or 20% even, it’s not coming from the University of Toronto, or McGill, or Calgary, it’s coming from the likes of Conestoga College (Google what went on there as it tells the whole story). The former have fairly stable international student flows because those are real students - there’s a reason they are paying CAD60,000 tuition. Meanwhile at the likes of Conestoga, it’s $15-20k perhaps. That’s the lost revenue per head (TPV for Flywire), not the $60k. Like I said above, not all visas are created equal, not all visa impact Flywire’s TPV in the same way.

The numbers have started off weak, down 30% perhaps, but those are very early and the thing which seems set to improve this year is the approval rate. Both the colleges and the agents are getting more savvy, there’s plenty of room in the caps this year, and the Canadians seem to want to move that approval rate up. But say they don’t, or admissions are down 20% rather than 10%? The difference that makes to revenues this year would be…. around $1m, on a base of over $750m.

The math shows that Canada does not matter either way for 2026.

Travel

The next biggest segment these days is travel. Again, I cannot see any short thesis that could come about from this area.

They told us in recent months that they do not have a presence in the middle east - so no direct war impact. Plus in the legacy Flywire travel segment the end client is a higher net worth individual or family or group and let’s say it like it is - these people going on luxury safaris or heli-skiing of renting a yacht for a family cruise in the Med don’t care about a $200 fuel surcharge. The demand is fairly inelastic and these trips are often special occasions planned well in advance. Outside of (really) major wars this is a sustainable type of revenue.

In terms of the Sertifi acquisition (from 1Q25) - that part of travel is more hotel events, and again this is conferences and weddings etc, so not as discretionary as a middle income family trip. If you look at my prior notes esp from January I outlined how they have spelled out a load of synergies from this deal, and although these are multi-year in scope, they do not appear to be in consensus estimates because they have only recently started executing on them after making some investments in people and territories. This is a source of upside over the medium term is my thought, and it’s certainly not a wise thing to base a short thesis on as when the synergies start to ramp as we started to see in 1Q, they could surprise on the upside potentially.

Here’s my revenue bridge from January. I don’t think I have to update it. It’s one of the reasons I’ve been ahead of consensus by some way.

Healthcare and B2B

Both of these segments (mid single digits of revenue each) are having great years which will flow through into next year as well. I’m going to leave it at that.

Margins

Having exceptionally high client retention makes growth easier to exhibit as you don’t have to spend marketing dollars on replacing the churn. That’s not to say student volumes were not an issue for this stock, obviously. If you put it in the context of the massive ramp into 2023 post-COVID, and now this fairly substantial normalization in the Big Four, you can convince yourself that it’s just part of the cycle. Looking much more broadly globally, Flywire is capturing a lot of the flows, and that’s a good thing. The company said that they’re going to come out of this stronger, and that’s exactly what I see happening.

When you already have your global payments platform more or less built out.

When the CEO and CFO are championing AI leadership within the company, and creating the internal data architecture needed to get the best use out of it.

When you have already made the investments to enable the synergies from Sertifi to come through.

When you have already built out your next-generation platforms to tack on the domestic portion of enterprise clients within education.

When your healthcare build out is finally paying dividends from marquee wins.

…. it should not be a surprise if margins surprise on the upside at some point. Even though the trajectory is quite good already. I’m not saying this year, but at some point in the next two or three years, it’s quite possible.

Incremental margins are around 35-40% and that’s with some investment spend lately. Consolidated adjusted ebitda margins were up around 3.5% to 20% in 2025, and they are on track to be near 25% in 2027 per the CFO. I would assume that there’s several years of rising margins ahead under a steady state scenario.

I think FLYW will be a 20% grower over the medium term. 2027 margins are (initial, informally) guided at about 25%. If 20% of your revenues (the new revenues you book each year at 20% growth), come on at 40% incremental margin, then that adds 3% to the consolidated 2028 margin, and after that it’s 2.4% etc. But consensus margin for 2028 is ‘only’ 26%. I am just doing some basic math here not making a forecast - they may spend more to grow of course. But I think consensus top line as well as margin forecast remain too low.

Stock based comp

Like many smaller cap growth names there’s some stock comp add back here. However at around 10% of revenues in 2026 it’s not terrible, unlike some SaaS stocks - and lets face it a big part of the SaaS sell down was just that investors finally looked hard at the SBC at these companies. In any case they have seen diluted share count decline this past year given the buybacks at a cheap valuation. I expect stock comp to gradually decline over time as a proportion of revenues, and it shouldn’t be an issue for investors. I peg it at 6-7% of 2030 revenue fwiw.

Valuation

The stock is around the level it was at prior to the weak guidance and the expensive Sertifi deal that made the stock implode in late February 2025. The broad market is up 25-30% since that time and FLYW has materially outgrown it. Organic growth in 1Q was 30% y-y, after 17% in 2025 despite all the hand wringing, and margins have ramped maybe 400 bips since the start of 2025. In short there is still a fair amount of recovery left to be had in my opinion.

Again, given the high quality of their revenues (over the long term certainly), the low client churn, the low level of competition they experience, the high 17% organic growth in 2025 even through an adverse macro backdrop, this is a high-quality company and it ought to be valued as such. It is not at risk from AI disruption - the regulated payments network, the deep domain expertise, along with deep integrations into client workflows and systems of record, will ensure that. It’s not in an area that attracts too much competition.

Many payments names have been essentially given away in the past year or so, Flywire was certainly in that bucket a few months back when many large growth funds gave up on it. But after doing well since then, it is still meaningfully undervalued per my work. The very recent sell down v’s peers because of fears about student visas doesn’t stand up to analysis - I could be wrong, but I think the short interest at 11.5% here is unwarranted. Let me know if I am missing something.

How to look at valuation

It’s funny how the market can apply a 100x revenue multiple on certain large and exciting stocks, with key man risk and no way of knowing what long-term profitability will be because it’s so new. And yet, with some investments, it’s staring you in the face that it’s ridiculously cheap, and most people don’t find it exciting for whatever reason. Likely because it doesn’t have momentum - but that can change.

When looking at Flywire, I think you have to look ahead a few years. Look at incremental margins. Look at what they’re spending so those incremental margins could be even higher when that spend starts bringing in revenues. Grow the business, and you’ll see that margins can grow from this year’s 22.5% to the mid-30s. And that’s not even full maturity.

If you want very clean numbers, you can take 10% off this year’s margin based on stock-based comp. Maybe in four years you can take 6% off margin. Almost every small cap software company in the market, uses a lot of stock-based comp. It’s certainly not gonna be a big deal if someone acquires them. They would just tack on 6% extra shares to buy all the employees.

Flywire is quite cash-generative. I have cash growing to $1.8 billion by 2032. They’re likely to do M&A and buy back stock instead of accumulating this much cash, but the point is this is not a risky company, financially.

Diluted shares today are 128 million. If the valuation stays this low, it could go lower over time given the power of buybacks, but I’m assuming it rises to close to 150 million by 2032. I’m assuming the market will be rational at some point such that the company won’t be able to buy back stock so cheap.

You can do your own valuation work, I’m sure. When I look ahead, apply more mature margin, and then discount at 10.5%, and when I apply a reasonable multiple (for example, the same EV/EBITDA multiple as a mature, non-consumer-facing payments company such as Corpay, CPAY ticker), using the earnings including stock-based comp and using the diluted share count (of 149m) as if they didn’t buy back a bunch of stock.

I get a fair value today of $38.

My official target is $30 for this year-end. I’m not blind to sentiment across the group. Although that seems to be changing for the better.

This is a rough draft I did a handful of weeks ago showing how I get $38. Again, this multiple is just based on 14 times EBITDA, fairly modest for high cash generation, low capital intensive, compounder. This stock has been growing at the top quintile, if not the top decile, of all stocks in the market, and those revenues keep going higher regardless of what fate has thrown at it recently.

I suspect that Flywire will do an investor day very soon, possibly in the fall. It may coincide with when they are confident that they are completely past the headwinds in terms of student visas. But that student visa stuff is not the driver of this company anymore. It has been driven by share gains and expanding its TAM, diversifying and ramping profitability. That’s why it’s my largest position and why I think it’s a multi-year holding, with a potential multi-bagger outcome.

How I get to $38…

Final point: it’s just me introducing a little bit of speculation. If you look at who Flywire took Penn State and Cornell from in the competitive tenders, it was the education subsidiary of Roper (ROP). Roper spent about $7 billion on its education vertical in the past few years, some of it in the payment space, including Transact Campus, aka Cashnet (all now branded Illumia). If you look at the spreadsheet that I embedded, you’ll see that firms like this are incumbent players for universities that likely need to upgrade and consolidate their vendors. They don’t have an international presence for cross-border, and they just use a third-party, small-time player for that.

The hard bit for a college, and by the way, the most lucrative, is the international students. Flywire dominates there. When you’re consolidating, that’s the part you need. If that entity, doing the cross-border, happens to have a great domestic solution, then it’s very easy to get rid of your incumbent with all its technical debt and go with the new guy. In fact, it makes total sense to have only one provider because then it’s less confusing for students, and it’s certainly less confusing for the employees. It’s all automatically reconciled rather than having five different drivers of your payment.

Universities and colleges probably have a fiduciary duty to get on board with the likes of Flywire and save costs for themselves and ultimately for the students.

So if you’re a Roper who has paid between 14 and high teens post-synergy out years EV/EBITDA for education assets, some of which weren’t even growing as much as Flywire, you might consider adding in all the lost business that you’re going to suffer if you don’t acquire Flywire.

I don’t think Flywire will sell because they can do the math see that there is a lot of upside in their stock. But the fact that the market gives essentially zero chance of this happening is just a little bit indicative of what type of market we are in.

My 2027-28 top line is now 3% and 4% below the prior-day consensus. It’s pretty clear here that they’re focusing on profitability, and some might worry that cutting the sales force in Homes might lead to further weakness in net new bookings. My model says I need a 22% increase in bookings next year to meet my top line. That rate of increase is not unreasonable, as it’s happened before, and there’s definitely an argument that their new sales force needs to season.

My out years EBITDA is a little ahead of consensus that existed yesterday because I do think they’re intent on executing on getting more efficient. It’s a long way in the future, but for what it’s worth, my EBITDA estimate for 2030 is about 17% below what the company was aspiring to at the beginning of this year.

That said, the stock is massively below where it was at the beginning of this year. I’ve increased the pace of buybacks throughout this period, keeping leverage where it is. That reduces the share count by a net 15% over that period. In a way, EBITDA per share is not far off what they were thinking about.

I’m going to post my model here because I think it’s necessary to see what’s going on here. This model is an example of what you can do using AI these days. It takes a bit of iteration to build it up as good as thing, but it’s incredible what it can do.

You can even have it audit various lines that look a bit funny, and it’ll go and research what’s in the Q’s, the release and presentation, and everything else in the transcripts that you upload. It’ll come up with a reasonable number that fits in with the facts.

This is building with Claude Cowork and then iterating and adding tabs etc with Claude in Excel using Fable. I would not bet that it’s perfect but what model is.

CoStar Model Post 2q26

131KB ∙ XLSX file

Download

CoStar model through 2030 with 2026 quarters Shows a take private analysis, which supports a good IRR even in the 40s

Download

I don’t have any dispute with the fact that they are missing the bookings expectations. Homes.com continues to be a distraction, perhaps, and in general there’s a lot going on in terms of new investments to build out geographies etc. What is perceived as a fight with Zillow may be spilling over into the apartments side - might it be a bit of a war of attrition, perhaps they need a mediator. I get all that, but then again, so what? It’s just ridiculously mispriced versus a less emotional outlook and the belief that businesses go through ups and downs. This is a bit more of a downer right now but there is no evidence that it is permanent. For the most part, apart from some more volatile transactional elements, the core CRE side of the business remains healthy. It could improve quite a bit if the market picks up.

I have articulated how the implied value of homes is a multi-billion dollar negative. Here it is with updated numbers. I’m using 2.2x for resi, which is what Zillow is trading at on 27 numbers. Remember, prior to all this competition and perhaps the SaaS implosion, Zillow was trading at five times forward. If these two players were to stop having a go at each other, who knows? But regardless, right now, homes are costing CoStar $24 a share in missing valuation. By the way, the 7x multiple for the CRE business reflects what is typically the bottom for the whole company over time, and as such, that includes the resi. The point is that CRE probably bottomed higher than that if you could split it out.

Seeing the stock down double digits after market made me ask my pal Claude to come up with a take-private tab, which is in the model posted. You can play around with the assumptions, but for example, the 14 times exit multiple for the company on the EV to EBITDA in year five, it is not at all egregious compared to where the stock would normally trade. In short, if someone was to pay a 60% premium to the $26.50 where it was after hours, they stand to make a 29% IRR over five years.

Would they consider it? Of course they would. Who wouldn’t? My two cents says that management would make more money by going private, extracting their cut, fixing the business without the market on their backs, and coming public again whenever things are looking good. By the way, there’s a toggle which lets you either shut down or keep homes running. Isn’t Claude smart?

Full disclosure, I’m in the stock with an average of $28.50, and I sold some prior to earnings today.

However I have owned both of those stocks in some capacity over the years and so I do pay some attention. And I have to say, I am a little shocked when I hear that Stripe appears to be involved in a joint bid to acquire PayPal.

Most investors who know Stripe consider them among the more well read and thoughtful people in the payments space. Look at the breadth of interviews they publish on YouTube. And if this deal turns out to be true, it may make me feel somewhat inadequate once again - because you wonder what fantastic insight do they have that I may have missed, why didn’t I see what they see now?

But it almost reminds of how weird the rumor was that PayPal would acquire Pinterest a few years ago. To the extent that was true, I’m guessing that PayPal knew that they had to do something to broaden their lead generation, but since then it’s been all downhill for PayPal for various reasons.

One of the bear points I always had about PayPal was that it seemed to have a bunch of revenues that were ‘lazy’ on the part of users - in the sense that these were fat margin products that would one day go away as the user base churned. I’m not sure where that issue stands today - maybe the users did churn enough such that PayPal isn’t still over-earning on certain legacy products.

But still, a largely organically built company such as Stripe buying PayPal? I don’t get it at the moment. Aside from the valuation, which I’ll touch on in the context of the broader space. Going back to the time hog aspect of looking at a PayPal - wouldn’t the same issue come up for Stripe - I mean don’t they have better things to do with their time than buy and integrate PayPal? Think of the cultural issues alone.

At the same time you have to consider that Stripe has been doing some deals lately - particularly in stablecoin infrastructure and the broad wallet area. And they have been developing Tempo, the stablecoin settlement rails. And with the advent of agentic commerce, there’s generally a lot going on in this space and it’s moving fast. So what bottleneck might Stripe be coming up against that would make them consider buying PayPal?

Well, it’s likely brand, existing users across multiple areas of distribution who trust PayPal, and PayPal does have some good technology especially when it comes to authorization and consumer identity. The identity aspect is likely going to be more of an important aspect of payments than ever - agentic commerce needs robust infrastructure in that regard, and consumer might one day get very nervous should the bad guys learn to use AI to scam even more people globally. And so trust, brand, tried and tested infrastructure is perhaps something Stripe needs on the consumer side.

I would imagine their goal is to shift payments increasingly off the network rails - that’s no easy feat as there’s lot of inertia out there especially among the consumer with the money. But as use cases develop further for stablecoin and other more real time forms of payment, and as agentic commerce grows especially within B2B, you could see that the player with all the best technology also needs to have the trust and distribution of the consumer side of things.

But there’s another aspect to this as well. This is a market that does not care about valuations much - witness the analysts fawning over SpaceX in their initiations at 100x revenues, and how the company will change humanity for the better etc :-). What do I know however it’s been pretty clear to me that valuation in the broad payments space just became irrationally cheap, especially in relation to many parts of the market.

Consider that I picked up GPN stock at what was then a NTM PE of 20-25% of the market multiple (maybe 25-30% of the equal weighted market). Seriously a quarter of the market multiple for a player that it entrenched in the offline payment world as much as Stripe is entrenched in online?

Or the likes of Flywire or Bill.com trading at 2-3x gross profit, when mature payments companies with less growth trade at 7-8x. That’s assuming that the smaller companies won’t see their margins right sized overnight via being acquired. Or that their incremental margins are now very attractive after having invested in their infrastructure already. Payoneer got fed up waiting for the market to value them at a reasonable multiple and sold to Nuvei - likely bad timing on Payo’s part if you ask me, but what does the CEO of Nevie see that the broad market could not. He took his company private and kept most of his roughly $2bn holding - do you think he’s going to do a dumb deal and risk that - no, he likely thinks he stole it, not that most investors care as they are too busy chasing SpaceX at 100x revenues.

So yes, despite all the baggage and the time hog that PayPal might be, perhaps it just got too cheap, just like a lot of the payments space.

This is my current view of CoStar Group at a share price below $28 in mid-July 2026.

The company’s history and business mix are well documented elsewhere. The issue that matters for the stock today is much narrower: how much permanent value destruction should investors attribute to Homes.com, both directly and indirectly?

The market’s answer appears to be: a very large amount.

My answer is: probably much less than is currently embedded in the share price.

CoStar is trading at or near multi-decade valuation lows relative to the market, its peers, and its own history. Sentiment is extremely pessimistic. Yet EBITDA estimates for the overall company are rising and, in my view, have further to go. The core commercial real estate businesses remain highly attractive, and the company’s established marketplaces have not suddenly come under existential attack.

The problem is instead that CoStar is attacking a new market, residential property portals, in an expensive and arguably undisciplined manner. This is not primarily a case of a core franchise being disrupted. It is a case of management choosing to incur heavy losses in pursuit of a new opportunity. That said, there has been a reduction in the spend outlook, partly to appease activist shareholders, but for some reason the stock seems to be more upset than ever over it.

Importantly, because that spending is discretionary, the problem is at least theoretically fixable.

The bull case does not require Homes.com to beat Zillow, become the dominant residential portal, or achieve the ambitions that management once set out. Nor does it require CoStar to return to the premium valuation it commanded during the 2019–2024 period.

It requires only some combination of the following:

• confidence that there will be greater spending discipline in general;

• a credible path toward lower Homes.com losses sooner than later;

• evidence that the subscription model has a viable customer base;

• improved execution in the rest of the company (perhaps investors worry that Homes.com is taking focus away from the core);

• or a more conciliatory acknowledgement from management that shareholder returns matter alongside long-term strategic ambition.

Even a partial improvement could produce a meaningful recovery from the current valuation.

The principal risk is that CEO and founder Andy Florance remains so committed to beating Zillow that he allows Homes.com to consume capital for much longer than shareholders expect. He has historically been willing to invest through periods of substantial losses to create valuable marketplaces. That history is one reason not to dismiss his strategy. It is also the reason investors worry that he may continue spending regardless of the stock price.

The stock therefore rests on a judgement about both the assets and the allocator.

I lean bullish because the valuation appears to discount an outcome substantially worse than the one I consider most likely.

As always, investors should conduct their own due diligence. I have followed CoStar on and off for many years, including during a period when a prior employer was one of the company’s largest shareholders. That gives me some familiarity with the company and its management style, but nobody likely has great insight into where the stock will trade over the next quarter. I doubt anyone thought a year ago that it would go from $95 to sub $30 now. My argument is about the asymmetry over at least the next 12 months, not 2Q bookings.

What the current valuation appears to imply

At this price, the investment case is not primarily about whether second-quarter net bookings are slightly above or below consensus.

The more important questions are:

1. Does management demonstrate greater sensitivity to investor concerns about Homes.com? (I thought Andy sounded a bit too ‘business as usual’ at a time when many long term holders were still hurting, during the 1Q call)

2. Does the company establish credible limits around future residential investment? (They have a lot of projects to invest in, often a good thing, but core investors looking at the longer term cash generation are likely wondering when will they get paid)

3. Does the market recognize how extreme the implied valuation of Homes.com has become?

Using depressed multiples for the core CoStar businesses and the non-Homes.com residential assets, I estimate that justifying the current share price requires assigning approximately negative $10 billion of value to Homes.com.

Homes.com is currently expected to generate approximately $275 million of 2027 revenue, representing only around 6% of total company revenue.

On that basis, the current enterprise value appears to assign something close to a negative 35 times revenue multiple to the business.

To be clear, this should not be interpreted literally as the standalone value of Homes.com; it also captures the governance and capital-allocation discount applied to the entire enterprise. But still, Homes.com is why these are issues we have to consider.

The following takes the current stock price, applies very conservative multiples to 2027 revenues, and back into what the implied valuation of Homes.com is.

A note about the multiples I used above. Over the past 10 or so years the stock traded at a mean EV/Sales (NTM) multiple of 10.5x. Apartments.com was in that mix then, so you could say that the core high quality, monopoly-like CRE businesses would be even higher valued if isolated. But I use 7x because even though I do not think CSGP is at threat from AI, certain peers have seen valuations decline this year, so it’s prudent to adjust lower. Plus some of that mean was influenced by a period when the stock was priced at generous levels, like a lot of longer duration assets were in the 2021-22 era in particular.

The Resi ex-Homes multiple of 2.3x is just using Zillow’s current year EV/Sales multiple. Now you may have noticed that Zillow’s multiple has imploded this year as well mostly due to CoStar’s antics perhaps. The mean 10 year multiple is actually 5x - if we plugged that into our SOTP it would imply that CoStar’s Homes.com is valued at negative 56x 2027 revenues, or negative $15.4bn. Again I am using a discounted multiple of the core CoStar business and solving for the current stock price. And again, in case the reader is unaware, there’s been no significant change to the underlying margin and cash flow characteristics of the rest of CoStar (or 94% of the company).

An imperfect analogy would be offering $400,000 for a house worth $1 million because one bathroom requires renovation. The bathroom may indeed be a problem. It may cost more to repair than the owner originally expected. But at some price like in this example, the discount becomes disconnected from the cost of solving the problem.

That is where I believe CoStar may be today.

The caveat is that investors are not obligated to value the company on a sum-of-the-parts basis. If management is willing to direct the cash flows of attractive businesses into an open-ended strategic campaign, then those cash flows are worth less to shareholders.

The negative Homes.com valuation therefore reflects more than accumulated losses. It reflects a capital-allocation discount, a governance discount, and a fear that management’s definition of winning differs materially from shareholders’ definition.

That discount is justified to some extent. The question is whether it should be $10 - $15bn billion. And shouldn’t the company get at least some benefit of the doubt given their past record. This is all occurring in an era where AI firms are being valued at massive multiples, without any realistic possibility of knowing what the long term economics will be. So just bare that in mind as the market awards this company an all-time low relative valuation for pursuing something they are expert in.

Brief background

CoStar was founded approximately four decades ago and remains led by its founder and CEO, Andy Florance.

The company is sometimes described as the Bloomberg of commercial real estate. CoStar has assembled a deep, proprietary database supported by a large research operation, established brands, embedded workflows and extensive industry relationships.

Its core commercial real estate information businesses have characteristics associated with high-quality information services:

• proprietary and difficult-to-replicate data;

• recurring subscription revenue;

• high customer retention;

• substantial workflow integration;

• strong incremental margins;

• and limited credible competition in several important categories.

Over the last decade, CoStar has expanded beyond commercial real estate information and into consumer-facing property marketplaces. Apartments.com has been the most important example and, despite substantial initial investment, has developed into a valuable business.

The company has also acquired or developed residential assets in the United States and internationally, including Homes.com and Domain in Australia.

There is little serious dispute that many of CoStar’s earlier acquisitions and marketplace investments have created value. The dispute is whether Homes.com is comparable to those prior successes—or whether management is attempting to extrapolate a successful playbook into a market where the competitive structure is fundamentally less favorable.

How the company arrived here

The immediate source of investor frustration is the amount of money being spent to secure a meaningful position in the U.S. residential property portal market.

Homes.com has created pressure in several ways:

• it has depressed consolidated margins;

• it has made near-term earnings difficult to interpret;

• it has raised concerns about capital-allocation discipline;

• it has contributed to execution volatility elsewhere in the company;

• and it has damaged trust in management’s willingness to balance strategic ambition against shareholder returns.

The central issue is not simply that Homes.com is losing money. Investors will tolerate losses when the eventual return appears attractive and the spending envelope is sufficiently defined.

The issue is that the size, duration and ultimate strategic objective of the investment remain uncertain.

Investors do not know whether management would be satisfied with Homes.com becoming a profitable and relevant second or third portal, or whether Andy Florance views anything short of market leadership as failure.

That distinction could be worth billions of dollars.

The valuation has moved from excessive optimism to extreme pessimism

For much of the 2010s and early 2020s, CoStar traded at a substantial premium to both the broader market and most information-services and software companies.

Some of that premium was deserved. The company had durable growth, exceptional competitive positions, a successful acquisition record and a founder with a history of building valuable assets.

Some of it was also a function of the market environment. Long-duration growth assets received increasingly generous valuations, particularly during the years surrounding the pandemic.

That matters when considering the activists’ criticism of CoStar’s share-price performance. The starting valuation in the 2020–2024/5 period was unusually high. Weak returns from that point cannot be attributed entirely to operational deterioration.

Today, however, the opposite appears to be true.

At a share price below $28, CoStar’s valuation is at or near historic lows on several relative measures.

As above, over the last two decades, CoStar at times traded at approximately 3.5 times the EV-to-sales multiple of the S&P 500. Today, that ratio is approximately 0.7 times, more than two standard deviations below its historical mean and below the level reached during the global financial crisis.

The comparison is affected by the current concentration of the market in highly valued mega-cap technology companies. It is therefore also useful to compare CoStar with the equal-weighted S&P 500.

Excluding the pandemic-era valuation extremes, CoStar historically appeared to find support at approximately 3.5–4.0 times the equal-weighted market’s EV-to-sales multiple.

Today, that relationship is approximately 1.3 times. Remember that CoStar has had 50-60 quarters of double digits revenues growth whereas the median stock in the S&P likely does 5-6% and is more volatile.

The stock has not merely de-rated to a less demanding valuation. Its historical premium has disappeared.

The market is effectively arguing that the quality of CoStar’s assets has deteriorated, that the earnings from those assets are less likely to reach shareholders, or both.

There is some truth in the second claim. Homes.com is absorbing capital that would otherwise contribute to earnings or be available for buybacks and acquisitions.

But relatively little has changed in the underlying competitive position of the core franchise.

This is what makes the setup interesting. If the market becomes even moderately more comfortable with Homes.com, the stock could re-rate substantially even while remaining far below its historical relative valuation. A return to prior peak multiples is unnecessary to make this a compelling investment.

This is not a conventional AI disruption story

The de-rating also needs to be considered in the context of the broader pressure on software and information-services valuations.

Many investors now apply an “AI discount” to companies whose data, software functionality or customer workflows may be replicated or disrupted by generative models.

I do not view CoStar as especially vulnerable to that risk.

The software interface is not the principal moat. The moat is the proprietary information, the research infrastructure required to maintain it, the brands, the customer relationships, the integrations and the role CoStar’s products play in established workflows.

A competitor could build a modern interface relatively quickly. It could not readily recreate decades of commercial property data, the personnel footprint needed to maintain that information, or the installed customer base.

The experience of Homes.com and Apartments.com reinforces the point. Building the website or application is not the difficult part. Building the audience, brand, data and commercial organization requires billions of dollars and years of effort.

As above, over the last twelve months or so, CoStar has materially underperformed the IGV software ETF and most relevant information-services peers.

While CoStar’s revenue mix now includes lower-margin marketplaces, more than half of revenue—and a much larger proportion of profit—still comes from high-quality, recurring sources. I do not believe that CoStar has material ‘seat exposure’ to CRE brokers such as CBRE - these are relationship driven firms such that AI won’t have much if any impact. In any case they monetize their CRE data across a very broad set of users relative to the case 15 years ago.

A partial recovery relative to software and information-services peers would produce significant upside without requiring CoStar to return to its former premium.

The qualitative problem: management has lost the benefit of the doubt

The valuation decline cannot be explained entirely by financial forecasts.

There is also a breakdown in trust.

The market increasingly perceives CoStar as dismissive of legitimate investor concerns. This perception became particularly important during the company’s engagement with activist investors.

CoStar reached some accommodation with the activists, but they did not obtain the degree of strategic change they appeared to seek. Third Point subsequently exited, and it is possible that D.E. Shaw has also reduced or exited its position.

Andy Florance has described the activist process as a distraction. That may reflect his genuine experience, but it is unlikely to reassure shareholders who have watched the stock decline from approximately $95 to below $30 in around twelve months.

The first-quarter 2026 earnings call did not fully acknowledge the severity of the valuation problem. Management’s tone came across as too close to “business as usual.”

That matters because the market is not asking only whether Homes.com can eventually succeed. Investors are asking whether management recognizes that the present strategy has imposed a substantial cost on existing shareholders.

The company does not need to abandon Homes.com to address this concern. It does need to demonstrate in a more forceful way that the investment has limits and that shareholder returns influence the pace and scale of spending.

At the current valuation, a change in tone could be almost as important as a change in quarterly bookings.

Andy Florance is both the reason to own the stock and the principal risk

I have some historical familiarity with Andy Florance and his management team because, from approximately 2011 through 2018 or 2019, a fund at which I worked was among CoStar’s larger shareholders.

My impression is that Florance is intelligent, competitive, highly driven and unusually willing to pursue long-term goals despite near-term criticism. Those characteristics helped create the company that exists today.

They also create the current risk.

Florance has built CoStar over roughly four decades. He has successfully invested through periods when reported margins were depressed and investors questioned the economics. Apartments.com is the obvious example.

He has therefore earned some latitude.

But success in one marketplace does not ensure success in another. Residential home sales are structurally different from apartment rentals. Zillow is a stronger incumbent than the competitors CoStar faced in some of its earlier marketplace investments. The economics involve a different customer base, a different transaction frequency and deeply entrenched industry behavior.

Florance’s public comments reinforce investor concern about his willingness to endure pain.

In an August 2025 interview with the Australian Financial Review, he described himself as focused on long-term fundamentals, committed to his objectives and willing to “eat nails.”

That mindset can create exceptional long-term value. It can also lead a founder to continue investing after the probability-weighted return has deteriorated.

Many investors want the stock to recover now. Florance appears willing to tolerate substantial short-term pain in pursuit of his strategic vision.

The question is not whether he is sincere or determined. He clearly is.

The question is whether that determination will cause permanent damage to CoStar’s valuation and whether management’s concept of winning is compatible with an acceptable return on incremental capital.

Why management may have room to recalibrate

There is a plausible way for Florance to reduce the perceived ambition of Homes.com without conceding that the underlying strategy was a mistake.

Management’s initial revenue targets were set several years into the future. Providing five-year revenue guidance for a new marketplace was inherently risky.

The assumptions behind those targets also appear to have included expectations that changes to U.S. buyer-agent commission rules would materially disrupt Zillow’s lead-generation model.

In late 2024, Florance argued that Zillow and Realtor.com might be forced to find new revenue models because lead diversion could become much more difficult.

That has not occurred to the extent CoStar expected.

Despite changes in contracts and procedures, buyer agents continue to be compensated in most transactions, sellers continue to fund the economics either directly or indirectly, and aggregate commission rates have not changed dramatically.

The industry’s resistance to change illustrates how difficult it is to alter entrenched residential real estate practices.

It also weakens one part of the original Homes.com thesis. The rule changes did not force Zillow to abandon its model, and they did not create the structural opening CoStar anticipated as at 2024.

Management could therefore credibly argue that the external environment evolved differently from expectations and that the spending plan should be adjusted accordingly.

That would not require abandoning Homes.com. It would mean redefining success around profitable relevance rather than market leadership.

The CFO change adds noise, but may also signal greater margin discipline

CoStar recently announced that its CFO is leaving to pursue another opportunity and that he will be replaced internally.

The departure inevitably creates additional uncertainty immediately before earnings. Investors will ask whether it signals weakness in second-quarter results, disagreement over strategy or concern about the outlook.

I do not place much weight on those possibilities without further evidence.

The outgoing CFO had been in the position for only a couple of years and appeared to have more of a banking and capital-markets background than a long operating history inside the company.

The incoming CFO appears to bring more direct operating experience, including experience managing costs while continuing to grow the European business.

More interestingly, the company’s announcement repeatedly emphasized margins and the incoming CFO’s record of cost control.

That language suggests that margin restoration will receive greater attention on the next earnings call.

The appointment does not resolve the Homes.com issue. The CEO, not the CFO, determines the scale of the strategic ambition.

But it may indicate that the company understands the need to demonstrate better financial discipline.

Homes.com has missed expectations, but that does not mean it has no viable market position

The financial results are clear: Homes.com has not met the revenue expectations initially established by management.

The company has spent heavily, revenue conversion has been slower than anticipated, and profitability remains several years away.

Investors are right to be skeptical.

The more relevant question now is not whether the original plan was too optimistic. It plainly was.

The question is whether Homes.com can occupy a viable and eventually profitable position in the residential ecosystem.

I believe it can.

Homes.com does not need to displace Zillow across the entire market. It needs to attract a subset of agents and brokers who prefer its economic and branding model.

Zillow’s model has increasingly moved toward taking a substantial share of the commission generated from its leads. In some arrangements, Zillow can capture up to approximately 40% of an agent’s commission.

That model can work well for teams capable of processing large lead volumes and converting them efficiently. But many agents will dislike the economics and the loss of control.

The potential Homes.com customer is not necessarily the agent seeking the maximum possible volume of purchased leads.

It may instead be the agent who prioritizes:

• ownership of the customer relationship;

• retention of a larger share of commission economics;

• control over personal and brokerage branding;

• reduced dependence on Zillow’s lead-allocation rules;

• and direct engagement with consumers interested in the agent’s own listings.

Homes.com’s proposition—“your listing, your lead”—has intuitive appeal to that portion of the market.

Why agents may support an alternative to Zillow

Several factors could create a durable customer base for Homes.com.

Commission economics

Giving up 30% to 40% of a commission is economically and emotionally painful, particularly after an agent also shares revenue with a brokerage and, in some cases, a team.

The Zillow model may still be worthwhile when it produces sufficiently high conversion and volume, but it is not universally attractive.

Dependence and control

Zillow imposes performance and volume requirements on participating brokers and teams. Agents who fail to satisfy those requirements may lose access to preferred lead flows.

This creates the perception among some agents that they are effectively working for Zillow rather than operating an independent business.

That is a poor cultural fit for a profession that attracts people partly because they value autonomy.

Brand ownership

Agents and brokerages have reason to resist allowing Zillow to become the sole owner of the consumer relationship.

Even agents who use Zillow may support Homes.com as a second channel in order to preserve bargaining power and maintain their own brands.

Consumer experience

As a consumer, I have found it irritating to make an inquiry on a property portal and receive calls from an unrelated broker rather than the listing agent.

If Homes.com can offer comparable inventory, search functionality, content and virtual-tour capabilities, some consumers may prefer to communicate directly with the agent representing the property.

Matterport and CoStar’s content investments should help narrow any remaining product gap.

Conflict of interest

Under the traditional portal model, a consumer enquiring about one property may be routed to an agent who is not connected to that listing and who may have incentives to sell competing properties.

Homes.com can argue that its model better aligns the listing agent, seller and consumer.

Whether consumers care enough about this distinction remains to be proved, but the proposition is coherent.

The market does not have to be winner-take-all

Investors often frame Homes.com and Zillow as though the residential portal market must produce one winner and one loser.

That may be too simplistic.

Property portals are only one part of a broad real estate marketing ecosystem. Agents spend money on open houses, brokerage platforms, social media, local advertising, search-engine marketing, photography, video, direct mail and third-party technology.

A dollar spent on Homes.com does not necessarily have to be removed from Zillow. It could come from other areas of the marketing budget.

Different agents will also prefer different business models.

Some will favor Zillow’s higher-volume, performance-oriented lead model despite the large revenue share.

Others will prefer to retain more margin, maintain control of their brand and communicate directly with consumers.

Many may use both platforms.

The existence of Zillow does not eliminate the possibility that Homes.com can become a valuable business. The issue is whether CoStar can reach that position at a rational cost.

That is where the debate should remain focused.

The real underwriting question: what happens to losses?

The stock does not require Homes.com to meet the revenue targets established in 2022.

It does require the annual losses to become bounded.

Management currently expects Homes.com to achieve profitability around 2029/2030.

Investors are understandably reluctant to capitalize earnings that will not appear for several years. There is also no assurance that the current timeline will hold.

However, even a failure to achieve profitability by late 2029 would not automatically justify the present valuation.

By then, management would have several choices:

• reduce marketing;

• slow product and content investment;

• raise subscription pricing;

• narrow the target customer base;

• integrate the business more closely with other CoStar assets;

• seek a strategic partner (combine with Realtor.com anyone?)

• or reconsider ownership of the asset.

A business that fails to reach its target profitability date does not necessarily continue losing hundreds of millions of dollars indefinitely.

That is the flaw in treating the current loss run-rate as perpetual.

The bear case needs to explain not only why Homes.com will miss management’s targets, but also why CoStar would continue funding those losses without limit.

That outcome is possible under a highly determined founder. It should not be treated as certain.

What matters in the second quarter

Net-new bookings will receive considerable attention when CoStar reports second-quarter results.

A strong number could produce a substantial share-price reaction because expectations appear low and investor positioning is likely defensive.

I have done a detailed model of booking and flow into revenues and although it’s too much detail for this report, the bottom line is that I think in order to hit 2027 revenues, that the rate of bookings into 2027 has to be 30% more than what the current run rate has been.

This takes into account a 4% CPI kicker across all their revenues (and that should also be applied to actual bookings - in fact industry reports say that they raised pricing for new Homes.com members in May for example and it sounded like more than just CPI).

Given what could be tailwinds arising in core CRE as external conditions improve (per the likes of CBRE) and also assuming the salesforce seasons into a higher production mode, I don’t think a 30% improvement is a stretch in the circumstances. In general, analysts seem too focused on sequential bookings rather than y-y, we’ll see on July 28.

However, the more important information on the 2Q call may be qualitative.

Investors should listen for:

1. A better defined Homes.com spending plan with a credible path to margin improvement.
   Is consolidated margin expansion based on real cost action or merely optimistic revenue assumptions?

2. A more realistic definition of success.
   Does management describe Homes.com as a profitable alternative within the market, or continue to frame the objective in terms of defeating Zillow?

3. Evidence that core execution is improving.
   Are bookings, sales productivity and customer retention stabilizing outside Homes.com? Is the salesforce becoming ‘less rookie’ and more productive.

4. The incoming CFO’s mandate.
   Does the company explicitly connect the appointment with greater cost discipline and accountability? Are they listening to the stock price crashing sound?

5. Capital returns.
   Is management willing to use the depressed valuation to repurchase a meaningful amount of stock? I’m not as interested in buybacks now because ultimately, it’s the multiple which can move things materially from here, not reducing share count by 3%. Plus they already executed a lot of buybacks in 1Q and it didn’t help, nor did Andy’s open market purchasing in the mid $30s.

What the stock could be worth

The appropriate valuation framework is a conservative sum-of-the-parts analysis rather than applying a historical multiple to consolidated revenue.

The core CoStar businesses should be valued using a multiple that reflects their recurring revenue, strong competitive position and high incremental margins. However I use a number maybe 25% lower than what it used to bottom at over the years.

Apartments.com deserves a lower but still meaningful marketplace multiple although Zillow is also quite depressed - would that stock recover some if Homes.com didn’t have such an open checkbook?

Using 7x for the core CRE as above, and 2.3x (Zillow’s current 2026 multiple) and giving Homes.com zero value rather than negative $10-15bn, I get a $51 stock. Using the 5x that Zillow might trade at if the competitive outlook became less of a threat, I get $66. So maybe you award Homes.com a more level headed valuation of negative $1bn - well for every $1bn in negative value you want to make it, that’s worth about $2.50 a share.

At $28 or so I don’t care much if Homes.com’s booking are soft this quarter - maybe that would help change the tone about getting to a dominant position, and instead make them pivot to a focus on margins (which is implied in the release about the CFO departure anyway).

What could invalidate the thesis

The principal risks are not difficult to identify.

Homes.com spending remains open-ended

If management refuses to establish limits and continues increasing investment despite weak revenue conversion, the capital-allocation discount may persist or deepen.

The core business is weaker than it appears

If net bookings, retention or pricing deteriorate meaningfully in the commercial real estate businesses, the argument that Homes.com is merely obscuring a high-quality core becomes less credible. And investors will likely blame Homes for the distraction that causes it.

Management defines success as market leadership

If Florance is unwilling to accept anything short of becoming the number-one residential portal, the eventual spending requirement could be far greater than investors expect.

Homes.com lacks genuine customer demand

Agent dissatisfaction with Zillow does not automatically translate into a willingness to pay CoStar. The company must prove that its proposition supports attractive retention, pricing and customer acquisition economics.

The valuation remains structurally impaired

Even if earnings recover, investors may permanently apply a founder and capital-allocation discount to the stock.

Management credibility deteriorates further

Additional executive departures, missed targets or dismissive communication could prolong the buyer’s strike.

Conclusion

CoStar’s current problem is serious but unusual.

The company’s core franchises are not obviously being disrupted. Its commercial real estate data moat remains intact. Apartments.com remains a valuable asset. The broader company is still growing, and consolidated EBITDA expectations appear to be moving higher.

The valuation collapse instead reflects the market’s belief that management will continue directing the earnings and balance-sheet value of those franchises toward a costly and uncertain effort to challenge Zillow.

That concern is legitimate.

But the current price appears to go considerably further. It assigns an extraordinarily negative value to a business representing only a small portion of prospective revenue and assumes that management will allow the losses to remain excessive for many years.

The market may also be underestimating the possibility that Homes.com can establish a profitable niche without defeating Zillow. There is a logical customer base among agents who dislike Zillow’s commission economics, lead-allocation rules and control of the consumer relationship.

More importantly, CoStar has the ability to change the financial outcome. It can slow spending, redefine success, improve margins and repurchase stock.

The company does not need to prove that its original Homes.com forecasts were correct. It needs to demonstrate that the investment is bounded and that management remains accountable to the owners of the company.

The valuation is now low enough that a partial restoration of confidence could produce a substantial return. The stock could rise 50% to 70% and remain inexpensive relative to its own history. Under a more meaningful normalization scenario, a doubling is not implausible.

None of that will happen automatically. Something has to change.

But the current share price appears to assume that nothing will.

Note: The thoughts and analysis herein are original, however some of the text was rewritten using AI to spare readers from my verbosity.

In my active accounts I have modest positions in a handful of payments names and a concentrated FLYW position. I believe I have edge in FLYW still, whereas most other stuff seems very macro and AI thematic (and throw in stablecoin for good measure). I have pointed out how a big buyback and a small valuation (in GPN for example) can be meaningful, and many stocks (such as BILL) are good value when you normalize for proforma margins in a takeout scenario. Despite reasonably constructive prints, such stocks have continued to be lagging the momentum trade, so thought I’d just add my 2c of color as it does seem that there are some signs of a thaw in the “AI disruption trade” which affects many payments and fintech companies to varying degrees.

• Retail traders are more powerful than ever and they seem to have no interest in these cheap stocks, especially when the stocks have not made anyone any money in recent memory. Retail loves stuff that has made them money and that’s regardless of how expensive the stocks may have become. And the opposite holds true, cheap valuations just validates that they are not good companies - right? That said, if these names start to work for whatever reason, there’s already few retail in them, and so when they start to move, it could be a good move….

• I started my career in the mid 90s and so was able to witness the dot.com era. I was covering financials at a GARP manager, and at one point when we were maybe not as heavy in tech as we could have been, I clearly remember my PM ask me (or was it threaten me) “why are you bringing me these deep value names”. Now if you get over the fact that at least one of these names went BK in the GFC (Wamu) that timing was memorable to me because not long after that conversation, the market for tech stocks did what it did and those deep value names doubled or more, and so the relative performance was pretty strong. Searching back it seems like at the end of 1999 WAMU was at a PE of about 8x times the current year earnings (which rose 12% the next year). Regardless of whether there are excesses in the AI trade, the fact is that AI, plus perhaps all the big IPO to come, has been sucking up liquidity from other areas, just like the dot.com era. The point is that it might take the same type of rotation event such that at least some funds flow down to less exciting areas of the market. Consider that just in the US public markets, $5-8tn of mkt cap has been added due to the AI trade since Jan 2025 per my math - and if even a small portion of this vast wealth is reallocated to currently out of favor parts of the market it’ll be very noticeable. Are we starting to see it right now?

• The valuations such as GPN at around 5x NTM EPS may be in the same boat as WAMU was in 1999 - I clearly though 8x was stupid cheap back then. By the way WAMU went to 15x in 2001, and GPN is at 4.3x 2027 right now - stupid cheap anyone? Nobody wants it today, just like my old PM didn’t want my deep value ideas in 1999.

What’s going on in the payments/software space right now: to summarize - the “AI will kill software” trade seems to be less threatening.

1. I will add this recent development as a #1 because it seems to have had an effect on the market today. It is that Jensen Huang once again talked about how AI won’t disrupt software, and in fact he thinks that now is a good time for software firms. It’s not the first time he’s said this - but the fact that it is resonating more with investors this time. That likely says something about investors being more educated about what’s going on than 2-3 months back - and/or just less ‘deer in the headlights’ when it comes to the AI bogeyman.

2. It’s not just high-profile Jensen Huang that is talking the market back from this AI disruption narrative. An article in the WSJ by Christopher Mims on May 22 (“The AI Superstars Who Say a ‘Vibe Slop’ Crisis Is Coming”) is also worth highlighting. It discusses the rise of vibe slop, the proliferation of poor-quality and potentially dangerous code generated by AI. Leading software engineers warn that relying on natural language prompts to build applications often bypasses essential design and testing phases, leading to massive technical debt and future security vulnerabilities. I have written about this before - and this kind of article, and likely more of them to come, is the type of thing that will cause Boards to adopt a cautious approach to allowing AI written software to encroach too much on the offerings of current software suppliers.

3. Not all AI software will be ‘slop’ however, especially in existing companies with established standards. But remember that mantra “software is free” - it might be if you are designing a basic app for your personal use, but it’s not if you do things according to high standards in a corporate setting. nCino, on their 1Q27 call recently, said that “We estimate writing code represents roughly 30% of the work it actually takes required to ship a product. The rest is product definition, quality assurance, security reviews, compliance validation and the coordination required to deliver Tier 1 mission-critical software to financial institutions of all sizes.” Not all of that 30% of code writing will be AI written, especially when there’s integrations involved, so maybe they are saving 15% of the software development costs. nCino likely has higher technical difficulty because they have to integrate and validate into large financial institutions, but they are a good example for the payments space as it’s the same standards required there.

4. And if look into how much the actual coding cost of a software firm entails as a percent of total revenues, for a mature public SaaS company, pure code-writing cost is probably only 3–7% of revenue, depending on how engineering-heavy the company is. So even a very large AI productivity boost to coding alone is worth maybe a low-to-mid single-digit margin event, not a 20–30 point cost reset, unless AI also reduces support, implementation, QA, DevOps, sales engineering, customer success, finance, legal, and product management. And there are additional costs to developing software this way - governance, extra testing and maintenance, that reduce the initial productivity benefit.

5. And so there you have it - yes a new startup is going to get a relative boost as they can lower their software development costs relative to the prior experience so they have a lower cash burn rate. But then what? Years ago when SaaS was replacing manual processes it was green field territory. Now, a new AI native firm is not going to have a green field, in fact they will have to try to compete against a proven incumbent that is also able to benefit from lower cost and faster shipping of new software.

6. So let’s assume that some of the startups manage to build some scale (likely by giving it away) - they will not be as efficient as established companies for many years even then. Are they going to have an edge in their offering for long - no, if you believe the bear thesis on software it is that software is free etc etc - as such the incumbent will just replicate any new features a newcomer might come up with - in 2 weeks. OK so what does the newcomer have to play with in terms of costs? Well you price your product based on your all in costs, not just software coding. Let’s start with how much of software coding as % revenues might typically be - you’ll have to look this up yourself but I come to 3-7% of revenue for a mature SaaS company. We are hearing that such costs are being reduced by 30-50% - yes some parts are ‘free’ but you have to look at the whole workflow not just a portion of it. So, with the benefit of added productivity, a software company has 1.5% -3.5% of margin to play with due to those lower costs. But guess what - that’s going to mostly be passed onto the client. And guess what else - you won’t be able to compete with the incumbent as in addition to having access to the best AI tools as well, they also have the trust of their clients, there would be meaningful switching costs involved, and they are actually best placed to evolve the client workflow in a responsible way rather than having some AI assisted newco with limited domain expertise some along and try to displace everything overnight - it’s not happening.

7. Bottom line - if only a select few had access to AI it would be a problem for incumbent software providers, but everyone has access - if anything it strengthens the incumbent as they can more easily develop products that go deeper into client workflows, all the while offering the ‘one throat to choke’ if something goes wrong. This is especially the case when it comes to security issues - a CEO who lets his accounting department vibe code a new accounts payable solution for example - it’s on him if there’s a ransomware attack on the firm resulting from that code. Not so if the vulnerability was with a large incumbent SaaS or payments firm - it’s their fault then - but at the same time it’s highly likely they will have better security as they can spread it out over many clients. It’s not even close.

   Are we there yet on AI fears? Not quite for the payments names yet, but with the IGV software ETF back to levels before the panic really set in, it’s likely only a matter of time before adjacent areas like software centric payments names catch more of a bid too.

8. Stablecoins - is that another bogeyman that’s been overplayed as a threat? The payments space has been hurt in some areas by the thought that just like AI written software “it’s free”. It seems that ‘tech bros’ love to narrowly focus on a single aspect of a transaction (the ‘free’ bit), even as they ignore the high fixed costs of running a business including all the compliance. I won’t dwell too much on it here but I would highlight a couple of things. One is the cross-border payments seminar that CorPay (in early May?) - it’s on their website, it’s not super long but it’s solid. These guys know what they are talking about, and they highlight the all-in costs of a cross-border transaction, and the real-world issues that come up. Essentially stablecoins only solve for a modest portion of the cost of a transaction. Not that there isn’t a place for them - it’s just not going to replace how cross border commerce is done today is my read, and even if used as the settlement rails, the value added is elsewhere on the transaction chain.

   The second thing is what Brazil did earlier this year as regards stablecoins - if you are interested just look it up, but effectively they killed any advantage stablecoins had other than speed. And let’s face it, a big part of the use case really involves avoidance of various taxes and other costs, and if they grew to be a large part of the economy that growth would effectively be ceding sovereignty of their financial system to whoever has the ability to influence the technology behind it - ie the US. And one last point, look at what the US did as regards Iran’s holdings of USDT - these were Tether issued and the US intervened (twisted Tethers arms) to have $344m in funds frozen. This is another example of how in reality these tether funds are not really the censorship-resistant dollar cash that they were said to be. Stablecoin will grow, but they won’t take over.

I am long the stock as I am attracted by what I see as a 20% grower that is priced at only 2.2x EV/NTM revenues, down from 6-9x pre the student visa issues of the past two years. I have a $30 PT, and I believe that the Street, with targets at $14 - $22, are anchoring to a story that no longer exists. As always, do your own DD.

The pivot the Street hasn’t priced in

The bear case from early 2025 was that visa policy in the US, Canada, and Australia structurally impaired FLYW's growth algorithm. It was correct as regards Canada and to a lesser extent the US, but wrong on Australia which ended up growing double digits rather than being down 30% as initially guided (out of caution). In the even the overall company delivered 16% organic growth which for a sticky revenue compounder, is pretty good. If things were so bad then how did they do 16%, which is much better than say Visa or MasterCard not to mention many other payments stocks?

• Three years ago, FLYW was effectively a cross-border education payments business with the US and Canada representing 40% of revenue. Today it is a diversified payments platform across four verticals where:

• Travel is approaching 25% of revenue in 2026 (vs 21% in 2025), with Sertifi attaching payments to a hospitality software base that grew 35% last year. There are a lot of synergies ahead with respect to Sertifi.

• Domestic education in mature geographies (US, UK, Canada) is now a comparable revenue stream to cross-border. Flywire’s growing reputation is creating efficient momentum in attracting new clients.

• The size of the cross border business in the US and Canada is maybe 13% of revenues now - it can’t do that much more damage yet may recover some day.

• B2B is 5% of revenue but contributing 3–4 points to FY26 FXN growth, the highest growth contribution per dollar of base in the company

• The healthcare segment was not growing much 2-3 years ago, now it is doing well, with marquee client above ensuring that Flywire is on every healthcare provider’s RFP list

Flywire is delivering strong revenue growth across every vertical and these revenues will compound and prove sticky over time (1% turnover among Education/Travel clients). The resulting acceleration in growth in 2026 and beyond is broad based and is not dependent on a rapid rebound in visa numbers in places such as Canada and the US, it is just helpful to not have them lose 75% of visas as we say in Canada in particular.

They have reached a scale where incremental margins are high (35% but this could rise as they scale), earnings quality is good (SBC is declining as % revs and could be 7% in ‘27), cash flow conversion is a healthy 75%. Geographies have expanded, and they believe that their TAM has expanded 10x per management. Their credibility as a supplier to educational establishments broadly, and in the travel space, is helping to attract clients. They are more deeply embedded than ever, and AI is helping them help clients and is making Flywire more efficient.

One final thing - some growth stocks such as TOST are very beholden to certain KPIs such as location count, SSS, etc as we saw last week. FLYW by virtue of its breadth of territories and products is less exposed to single item KPI risk these days, given its breadth both product wise and by geography. In addition their markets are not nearly as competitive as restaurants.

The Street has missed this pivot. They will catch up in time. I think the company is hinting at the need to host an investor day soon as they certainly appreciate that a lot has changed in the last couple of years.

1Q26 top line

Normally, many CFOs wouldn’t raise the year because of 1Q, but that’s hard to do with this momentum (30% organic growth in 1Q albeit not being extrapolated). At a guided higher midpoint of 21% ‘26 revenue growth FLYW (at the high end of their range) has got to within 1% of my initial growth estimate for 2026 (from my ‘research heavy’ January piece). They have several products still early in terms of monetization, and of course on top of a pretty high growth outlook for travel on a core basis, they have a lot of synergies to come from Sertifi over the next number of years. How does this change the outlook?

In aggregate, they are guiding to a midpoint of 21% y-y FX neutral top line growth (18%-24% range), up 300 bips. That’s $730m, although FX is helping some on top, and so consensus has settled at $746m. My January buy rec had a $755m estimate for the year, and consensus was $692m, so consensus top line is up 8% since early January, mostly organic, some FX, some as below (deferrals from 2025 showing up in 2026).

Consensus for 2027 is now $855m, up 7.5% from January. I was at $938m, 18% ahead of consensus, but now less than 10% ahead - still a source of alpha if correct, which I believe it is of course.

Flywire no longer has serious drags to top line, is the bottom line. And importantly, these sources of upside are enduring - FLYW is a top quartile, perhaps even top decile, grower from what I see, yet it is not yet priced remotely as if it is.

Within the revenue guide for 2026 there is the following to consider:

• 3-4% of that growth is the payment processing ramp from Cleveland Clinic. While this will add to the recurring base going forward, the pace of increase will be a tougher growth comp starting in 2H26 and 2027, esp 1H.

• The 9% revenue beat in 1Q was due to two main things - education had a big January intake, and Sertifi is starting to deliver.

Half of the beat was from education, and they called it out as timing related due to strong January intakes in mature countries. The likely reason is because a lot of the deferrals from 2025 turned up in January. Recall that due to constraints the US State Dept put on visa processing times, particularly in India, this caused colleges to offer deferred entry. Some of these students likely opted to study in the UK or Australia.

This may be ‘one off’ in the sense that next year won’t have this ramp, however this effectively increases the base of revenues that would otherwise belong to the cohort starting in Fall 2025, so the 2025-26 cohort is a tad larger than originally reported and will add to future revenues. As such, by raising the ‘25-26 cohort a bit, it helps 2027, even if the growth comp in 1Q27 will be tough.

2026 guide remains ‘prudent’

Management continues to assume that US visas will be down 30% in 2026 (worth about $10m v’s flat assumption), Canada visas down 10% (although revenues up 10% due to share gains), with UK/Australia flat. That’s very conservative to say the least yet it also demonstrates how little impact such assumptions have on Flywire’s revenue - a central part of my January thesis was that the visa issues are not critical to the company’s growth going forward, and this clearly demonstrates that. In recent months, the noise level has abated as regards student visas, and although there’s tweaks still ahead, overall ‘the storm’ has passed both in absolute magnitude but as important, it’s practically impossible to have the same impact on Flywire given the now modest relative size of the geographies that were causing drags in cross border revenues.

Travel started to deliver synergies

The other half of the 1Q beat came from the travel segment, with Sertifi being helped by synergies as well as continuing strong growth. The company provided some welcome disclosures in the 2025 4Q slides. among which was that travel was 21% of revenue that year - so it’s about 22% proforma for a full year of the Sertifi deal. They also said during the past year that the travel segment would be close to a quarter of the company in 2026 - my model has it at 24.6% at $190m ‘26 contribution. That analysis was outlined in my January ‘25 initial FLYW report which was more or less the same number.

Sertifi grew 35% last year, and legacy travel grew 31%, this per the slide deck from 4Q25. One thing I did do was lower the synergy ‘cushion’ as below from $15m to $10m as they seem to be getting the payments revenue now whereas I thought this might be back half loaded. Offsetting this I did lower the 2025 combined travel base a bit per their new disclosure on the slide deck from 4Q, so the delta from January is minimal. This is my latest travel segment 2026 growth bridge:

The amount of synergies they have articulated as regards Sertifi is around $180m spread out over multiple years (they have been vague on timing, and admittedly I’m not sure most would class a synergy as something that happens in year 5 but regardless) and so alongside a solid growth rate from the core of the segment, the augmentation from the synergies will result in a multi year period of high growth from this space is what I see. I think this segment’s outlook is one reason management seems more outwardly confident.

The company has been hiring and building overseas presence to exploit the revenue opportunities within Sertifi and I would expect to see the benefits of international roll out later this year, the monetization of existing US clients seems to be taking place now. What is the value of the synergies? If you take the $180m figure, the gross margin is maybe 45% so less than the 60% they did in 1Q, however the EDITDA margin is likely around 35% as customer acquisition costs would be modest. So say $63m of EBITDA at some stage - at a market average 15x that’s $945m. That’s in addition to core Sertifi too. It’s spread out over multiple years, and maybe my margin, or multiple, is too high - it would still show that the $340m acquisition price may not be the poor capital allocation that was the consensus in early 2025.

I would add one final point about growth in areas like travel and education - these are enterprise clients rather than fickle retail. The churn rate is 1% which is extremely low, and that in turn makes it easier to have an attractive net revenue retention, and it is helpful to margins of course as they don’t have to keep marketing to replace lost clients.

What is core growth in 2026?

FXN guide for 2026 is now 18-24%, with about 1.3% being inorganic (Sertifi deal). The ramp in payment processing, mainly from Cleveland Clinic, adds 3-4% to growth in 2026 up from the prior 2-3%. FX adds an extra point or so in ‘26.

As in the first spreadsheet clip below, I have 2026 organic growth of 26%. If you want to exclude the payments ramp it’s 22-23% - although that’s permanent revenues that will keep rising going forward, not one off in the sense its going away next year.

Because gross profits on the payments revenues that are coming on board are less than company average due to mix of pass-through interchange (credit card network fees etc), the organic gross profit growth is more like 22% in ‘26.

The important question - what is Flywire’s longer term growth potential? Is it a rule of 40 company (yes)

Some sell-side analysts have labelled FLYW as a lower to mid teens grower (and they think this is not good enough). So what is it? Some of the sell-side will wait to be told this, likely at an investor day. Here’s my take:

The part of the story these analysts are essentially drawing their conclusions from is the previously troubled cross border segments. The thing is that those segments are either bottoming (the US) or in the process of recovering (Canada) and in any case they are relative modest in size nowadays. Or in the case of Australia, it was a false alarm in 2025 as the initial guide that helped crater the stock in February 2025 said it could be down 30% but it actually grew double digits in 2025. And the UK, which has had some tweaks to visa policy, has other drivers that far outweigh that aspect of the growth (50% growth in 2024, 25% in 2025).

Meanwhile the rest of the business ex US cross border and Canada grew 31% in 2025, and overall education grew 12% even if you include the big drags from Canada and the US. International students in those two countries are arguably at levels which will rebound over time, plus in any case there’s tuition inflation, new client wins and of course the SFS domestic expansion that’s driving growth in those geographies too. Doing 16-17% organic constant FX growth in 2025 is not so bad when you face such policy headwinds.

Looking ahead beyond 2026 - here is my revenue model.

I am getting an organic growth rate in the low 20’s% for Flywire in the next 3 years.

My 2028 revenue estimate is 16% ahead of consensus. Travel will be the biggest driver, driving about 1/3 of growth and being a quarter of the company in ‘26/’27. With lots of revenue synergies on top of good core growth, and with it being a huge TAM, this one looks fairly well set up. Travel will drive about 1/3 of growth and reach 25% of the company by 2026-27, with the Sertifi synergy stack (sized above) extending the runway for years

Next is the UK driving 20% of the growth. I have prudently assumed that this growth rate declines just due to being more mature, but with tuition inflation likely strong, and SFS wins adding to core growth in cross border, I don’t see any reason to be too cautious here, particularly with Flywire opening up the market by one third extra through additional ERP integration work recently. Flywire’s UK business is probably 50:50 domestic v’s cross border and may be going more domestic over time. They have a virtual monopoly position serving the UK’s education clients - a very valuable and enduring revenue source.

Next in terms of growth contributor is the non-big-4 education segment which contributes 15% of my growth estimates, with B2B adding 10% of the growth over these two year ahead even though it was only 5% of revenues in 2025.

My 20%+ growth estimates actually doesn’t need a lot of growth from the likes of the US, Canada, Australia, it’s just needs them not to be down 30% like Canada was recently. And what’s not included here is additional M&A - the company has done well buying assets that can be better monetized at Flywire. Sertifi looked expensive but (as above) not if they can deliver the synergies in a handful of years. I would expect more bolt on deals where they acquire the software and monetize the payments.

Flywire’s chart relative the market

This is a chart of FLYW relative to the RSP (equal weighted S&P500). A lot of fintech/payments stocks such as BILL have also gone down over this timeframe, however these stocks are all quite different. FLYW’s issues that related to student visas are largely in the rear-view mirror as above. FLYW is not as exposed to the economic cycle as most stocks - it doesn’t have float revenues that could go away if rates fall, it has little if any small business exposure in the US, it has little mass-market consumer exposure (they cater to the luxury space in travel). And it is not priced for the mid teens growth that the sell side has factored in never mind the 20%+ growth I think is demonstrably more likely. As such I would expect the stock to outperform the market going forward.

Conclusion - Flywire has evolved, the stock will catch up to this reality in time

I’ve referenced above how an investor in 2026 is underwriting a very different company than in 2022-23. Investors as a class have yet to fully appreciate this. As such there is a high chance they do an investor day this year, certainly by May of 2027.

My target price remains $30 for the next year. As we see often in the space, sometimes these stocks double in a quarter, but of course timing that in advance is another matter. But FLYW will likely be an example of where 80% of the return comes in 20% of the time, so it’ll pay to be there ahead of time.

Why $30? Analyst targets range from $14 - $22 today. I think a lot of that is anchoring to what went on in 2025. Another key part of my January work was that the company was badly modelled by the Street - that has already proven itself out in the two reports since. I would say today that although the Street has closed some of the gap between my numbers and theirs for 2026, they continue to badly model the out years hence my reference to being 16% ahead of the 2028 consensus.

As above, $30 would be 4x forward NTM revenues at the end of 2026, using fully diluted share count (3.2x at the end of 2027). Prior to the student visa volatility the stock traded at 6-9x the same basis. And since then the broader market has increased in valuation. As such $30 makes sense to me.

What is $30 using EV/EBITDA? It’s 14x 2028, or 19x adding back all SBC. That’s at a 27% margin which is going to be going up 2% a year for some time. What is it then at proforma 36% margins at a more mature state - it’s 10.5x ev/ebitda, or 13x including all SBC. I think the way the world works is that at $30 when the chart looks good, these analysts will mostly love it and they will be putting 25x multiples on the stock, chasing it higher. At that point I may start selling some.

What are the main risks?

If the Sertifi synergies are not really synergies but are just future growth that would come anyways (at the same pace or less than the core travel) then my 20%+ expectation may be too high.

The gross margin decline, even though it has been explained by management as mix related, may confuse investors (or quants!) who may take it as something to be frightened of in this hypersensitive market

B2B grew 100% in 2025 but a lot of that is likely monetizing existing client flows - so if the growth falls to company average instead of remaining a high growth segment it might hurt some

Visa news will likely be more headlines than have potential to move numbers much, but it affects sentiment a lot.

The payments space staying out of favor, or small caps in general staying out of favor, is likely part of why they stock remains so inexpensive, and there’s no telling when this changes sustainably.

To be up-front, this is not a clean compounder - it still needs to pivot up market to avoid the smaller company churn, it needs to fix its gaap margin profile. It’s a sentiment-reversal trade where the bar to rerate is low given the starting point of 2x sales. It remains a modest position mostly because I have higher conviction in Flywire which is a concentrated name for me, but don’t dismiss Bill because stocks do go in and out of favor as we know. Ramp (not a direct competitor btw) grew at 100% recently and was valued at 32x sales in the private market - BILL went through that period as well don’t forget, as they say ‘this too shall pass”. Far from being disrupted by AI, isn’t there at least some chance that BILL can vibe its way to being a serious Ramp competitor? I am being flippant of course - but the market sometimes wants to see things as black and white - BILL’s current valuation v’s Ramp or Brex’s recent sale, suggests that’s what is going on here.

BILL reports this week - if you listened to their one conference appearance at Morgan Stanley this quarter, you may hear them sound quietly confident. I liked that tone - it was a lot better than the hedging and caution of two years ago.

Why the stock is disliked by the market

BILL rode the post-Covid SMB formation wave, then watched its customer base churn harder, spend less, and accept lower take rates as supplier pushback on virtual card pricing intensified. Public market tolerance for adjusted EBITDA companies running 16%+ stock based comp to revenues collapsed, especially given the AI threat arriving and with interest rates proving to be sticky (although that does help BILL’s float income, which I will ignore in this write up on the assumption that rates won’t move much). Competition keeps multiplying: Intuit, Ramp, Brex, banks, vertical SaaS, AI-native workflow tools. Investors used to love this name, rational investors sold it way too early in its early years as a public (yip, I did) and often the software analysts picked it up from the financial types and they couldn’t get enough of it at $200-300.

When you think about it - who is the natural buyer for BILL stock these days? Activists - yes, but aside from those guys this likely doesn’t appeal to value investors due to the SBC and the gaap margins. But is it even in the value indices yet - I doubt it. What about the growth guys? Well, it went from hypergrowth darling to landmine status. It’s not a clean growth stock any more, and that institutional memory of when it went from being in the pitch book as a top holding at $278 or whatever to being in the ‘alpha detractors’ part of your quarterly report to investors for a year or two - yeah that’s really why they dislike this stock, because it reminds them of past mistakes. As such, right now I don’t think BILL has much of a natural investor base. Retail, a cohort who could care less about SBC, they prefer a chart going up and to the right. It’s disliked because it hasn’t made investors money is the bottom line. Even I, someone who has liked the stock since it became valued rationally, I can’t make it a large position as I don’t have enough conviction. I have seen how the market can treat it over issues that while important, are hardly existential warranting a 50% haircut. But in a way I’m comforted by this sentiment backdrop, including my own, it just says that the bar is low.

Top-line: not exactly ex-growth

December quarter core revenue grew 17%. That is not hyper growth compounder territory, but it is also not the trajectory of a business in structural decline. It runs comfortably above median S&P 500 revenue growth of 5-6% of 2025 (7% cap weighted). If you look near the end of this note you can see that despite having a higher consensus growth profile than the likes of Corpay (CPAY) it has gone from a premium on EV/Sales to 40% of CPAY’s multiple - it seems irrational to fear for BILL’s future and not fear for CPAYs.

The 2023–2025 SMB backdrop was not good. Higher rates, demand uncertainty, tariffs and hesitancy on new software adoption hit BILL’s customer base directly, and the company-specific monetization reset compounded the damage. But recent bank commentary now points to SMB and lower mid-market customers beginning to invest again - they are tired of waiting for more certainty which won’t come.

The longer-term picture seems fine to me:

• Bill is broadening its product set, likely moving to be a complete ‘office of the CFO’ solution, much like FLYW has become the ‘office of the college bursar’. More products to offer = less dependence on a single product take rate, more overall stickiness, more revenue stability

• AI likely further accelerates digital adoption among SMBs - this is helpful to BILL. But don’t believe these silly ideas that companies will vibe code their workflows - sure certain companies with a tech bent may do some vibing for adjacent tasks, but few if any will do anything important in terms of solving for payments. It’s just not worth it, because it’s not remotely free despite the headlines, especially for the typical mid sized company. There’s other reasons such as security, maintenance and compliance, and the desire to avoid the security headaches that fall back to those that authorized such software vibing in the first place. It’s just not going to be a big deal for companies like BILL, in fact AI will likely be a positive. It will reduce the cost of BILL’s product development and maintenance, it will allow them to add products that clients find useful, helping to build a more sticky relationship. Agentic AI is talked up - but many SMBs still use checks - that’s how fast things move. The idea that most SMBs will move to an agents doing their back office work, settling payments, bargaining with other agents - that’s happening at some companies no doubt, but history suggests it’ll be decades before it becomes the norm. Meanwhile when it happens, perhaps BILL will be the prime candidate to help bring it into a client in a controlled way, allowing the CEO to sleep at night.

The platform is wider than the bear case treats it

The market likely still models BILL as a single-product accounts payable (AP) vendor. That framing is at least two years stale. The current product surface spans AP, AR, spend management via Divvy, and connective workflow across accounting platforms — closer to an Office-of-the-CFO layer than a point tool. You will see references to them adding tool such as travel management - isn’t it the case that software is easier to develop than ever - so why can’t BILL be the ones to broaden out the product set rather than allow some no name start up to come into the picture.

The repositioning is undersold for two reasons. First, it opens multiple ARPU expansion paths that do not depend on take rate alone. Second, sitting across accounting systems rather than inside one is potentially a strong place to be, contrary to what some might think. SMBs and lower mid-market firms migrate ERPs as they grow. A workflow layer that travels with the customer is more durable than one welded to a single back-end. BILL does not need to be Intuit. It needs to be the layer above whichever ledger a customer happens to run. Many SMBs might choose BILL because their solution is portable as it ties into all ERPS. Plus in any case many SMBs especially Middle market firms, have more than one ERP due to acquisitions among other reasons. Structurally speaking BILL doesn’t have a small niche to attack, they are capable of addressing the broad needs of all ‘office of the CFOs’ within SMBs.

Stock-Based Comp

An important variable in this name is stock-based comp. FY26 SBC at roughly 16% of revenue is the reason most generalists never open the model, although it has been higher in past years. They are not wrong to flag it - especially at a time when terminal values are being questioned due to what AI might come up with. Management seems to realize that they have to at least get this down over time, and they have made a start.

However, it could be argued that reported SBC number overstates the forward economic dilution, although that depends on the amount of new grants, but the company seems committed to discipline in this regard:

• FY26 SBC was guided to ~$290m, then cut by $30m mid-year.

• A mid-year cut to a number largely pre-locked by prior-year RSU vesting implies new grant values are being reduced more aggressively than the $30m headline suggests.

• Significant portions of current vesting expense reflect grants struck when BILL traded at ~$202 (2022 - although these may have rolled off already), ~$120 (2023), and ~$91 (2024). The P&L charge carries those grant-date values; actual share issuance and any offsetting buyback occur at today’s $40. This is why as per this chart, share count is declining recently.

The economic dilution math, in other words, is materially better than the GAAP optic. Buying back stock at $40 to neutralize awards struck at $150 is not the expense the income statement describes. It’s also not the outcome an employee would feel good about - hence there’s also pressure on management to get the stock up for this reason.

I think they can get SBC down to below 10% of revenues by 2028 - it may require a redistribution from stock towards cash comp however so it’s not like its a given that gaap margins will be that much higher unless they can show good operating leverage going forward. Yet BILL doesn’t have to become a margin-pristine SaaS asset to get away from the current distressed multiple of sales - but they have to demonstrate intent. It requires evidence the company is no longer being operated primarily for its employees. Activist pressure raises the probability the board treats this as urgent rather than aspirational.

Sustainably repairing the equity’s valuation is the only way equity comp remains a usable retention tool. Management’s interests are aligned with shareholders here, not opposed.

Margins - they just have to show a path to progress

Margin upside is necessary regardless of the SBC impact. Headcount discipline has been there (flat over four years, and BILL’s career site implies modest growth in hires in the near term. Incremental margins could therefore be quite good if revenues grow in the mid teens. AI-driven productivity in support and onboarding, and natural maturation of newer products will also help.

Under a strategic owner, the cost base would compress further. That is a separate scenario, not the base case, it should limit the downside - except I would have thought that at 3x revenues and here we are at 2x. In any case, valuing a stock that’s clearly not at scale is not very generous - the market should really be factoring in at least a decent chance that they get acquired.

Without going into detail on margins - it’s a topic that can be addressed when they start showing us some progress - and this is what the stock needs down here - just evidence that they are going to get good incremental profits as revenue expands

Strategic value = skew support

A takeout pitch by itself is speculative, even for a stock such as BILL, which is in a consolidating space, with two or three activists pressuring for change. AvidXchange (a peer with more focus on the mid market, and perhaps slower growth than BILL) went out at roughly 5x revenue before the deeper SaaS derate. More broadly, large banks have shown willingness to pay up for workflow-plus-payments assets that deepen commercial relationships, given how hard it is to build SMB distribution organically.

For a public market investor, BILL is an impaired SaaS name. For a bank or large processor, it is distribution into millions of SMBs, real payment flows, workflow tenure, and meaningful data adjacency.

I think the likes of WFC or even USB would be a good fit. I won’t go into too much detail, but BILL’s existing bank distribution is not an issue - some could be retained where WFC’s footprint is not as strong, but WFC has a lot of potential clients to offer BILL’s products to. The data that a BILL can capture for WFC’s other purposes would be quite desirable, as would the treasury and lending/deposit products they could offer. At a minimum the COF/BREX deal opens up more potential acquirers.

AI: more likely a moat extender than a moat killer

The reflexive AI bear case treats BILL as a SaaS workflow that gets disintermediated. The mechanics suggest the opposite. Agentic finance workflows require clean data, controls, audit trails, approval chains, exception handling, and trust. SMBs do not build those internally. They adopt them through embedded vendors with existing payment rails and accounting integrations. I have written before on Substack about the security concerns CIOs and CEO (and Boards) will have when smarter AI models are used to do damage to company IT systems. Experience has shown that fraudsters and ransomware attackers can innovate, and allowing either internal software to be vibe coded, or done via two weeks of work by a startup, won’t be the responsible thing to do in order to save what is likely to be a modest amount.

AI will likely compress point-tool value. It also raises the value of the trusted control layer those tools sit on top of. BILL has a credible claim to be one of those layers — not the only one (Intuit and larger ERPs are also well positioned), but credible, and also a neutral, portable choice versus locking yourself into Intuit.

Valuation: the bar is low

Skate to where the puck is going to be in a couple of years: normalized growth in the mid-teens, a more diverse revenue base across several new products that create a solution for the CFO and team, more mature margins, SBC tracking toward 8–10% of revenue, further consolidation in the space. On that profile, sub-$40 prices at 2x EV/Sales suggests permanent mediocrity. Stabilization plus incremental discipline, and the realization that BILL is more likely to be the purveyor and orchestrator of AI solutions for SMBs, does suggest that the bar is low. The stock could double and will still be considered attractive as long as metrics are trending in the right direction without as much apparent controversy.

One peer of interest is Corpay - it’s not an exact match but it is a corporate payments name that is mature, and also acquired a stake (alongside PE) in BILL’s direct peer AvidXchange. Obviously at one point BILL traded at a premium, but now it’s at 40% of CPAY’s EV/Sales (NTM). BILL’s gross margins are even higher than CPAY’s (which are also high) so given the synergies in such a deal the math would work at a much higher price. I’m not suggesting that CPAY buys BILL given the AvidXchange overlap, but who knows because Avid was more mid market and there would not be any competition concerns given the competition we all hear about?

And as below, BILL’s consensus forecasts say that it is a faster grower than CPAY - it’s maybe not a bad deal even with AVDX in the mix. They could likely pay double at least for BILL and make the numbers work with synergies.

Risks specific to the name

• Virtual card take rate compression accelerates as supplier pushback continues.

• SBC reduction is achieved by shifting toward cash comp, leaving total compensation flat.

• Mid-market expansion stalls; customer mix stays skewed to very small SMB.

• An AI-native AP entrant (Ramp or Brex extending further, or a credible startup) wins greenfield mid-market deals at scale.

• Strategic optionality forecloses, either via regulatory posture or by the board’s own choice.

• The 17% December core growth print proves noisy rather than directional. Iran war impact on costs and confidence puts SMBs back into ‘wait and see’ mode.

My conviction has stepped up materially overall the more we see issues like cyber risk being brought up in the debate about AI disruption. Many stocks in this space are just stupidly priced in my opinion. The ‘per seat’ issue and monetization is a separate topic, but from what I see many firms have already pivoted away from this, and software centric payments names in particular never had much of this seat exposure (I’m really coming at this from the standpoint of payment names where I have exposure - SaaS was never my thing).

The bottom line is that vibe coding will not displace established vertical SaaS and payments incumbents at any pace that matters to the investment cases for the foreseeable future. If anything, the AI cyber threat curve makes the incumbent moat wider, not narrower. And this moat it not just technological, but something that will endure as it encompasses a fear that is grounded in human nature - that of taking the least risky decision for your career. Vibe coding will certainly be a thing for certain tasks - but not for anything that allows unfettered access to systems of record, payments, employee data etc.

The decision-maker calculus has shifted

Put yourself in the seat of a CFO, CIO, or audit committee chair in 2026. The greed case for replacing an established SaaS vendor with an internally built or VC-backed-startup alternative has always been thin once you honestly account for switching costs, integration, compliance overhead, and opportunity cost. The fear case has now become overwhelming. Every board member has read about ransomware events, BEC-driven wire fraud, and supply-chain compromises. And now we see that the Fed has brought the banks together to discuss the threats they could face specifically from the likes of Anthropic’s Mythos. It should be pretty clear to everyone that a software system designed, tested and curated over time by an established software or payments firm is going to let a CEO sleep better at night than any system that’s vibe coded or comes from a startup without a history and robust risk infrastructure. None of them want to be the director of record when the post-mortem identifies an unvetted internal build as the entry point of a cyber attack

This is what stock market technicians call a “point of recognition” - the moment the market collectively decides a narrative is real and the chart accelerates. For vibe coding in regulated or operationally critical workflows, that point has now passed in my opinion as it does not stand up to many real world frictions, but the “Mythos Bogey Man” should focus attention on the fear of the risks, and these are more of a driver for decision makers than trying to save a few bips of margin. The mystique (and lots of PR) of frontier models being able to compromise more systems than ever before is, paradoxically, the best thing that could have happened to BILL, TOST, FLYW, WDAY, INTU, NOW, the card networks, and the broader vertical SaaS complex.

The regulatory stack has independently tightened the screws

This is the part the bull case on vibe coding tends to ignore. The compliance perimeter around enterprise software has expanded dramatically in the last 24 months, and each new requirement raises the documentary and operational bar in a way that favors scaled, audited, established providers: (AI came up with this list - it looks legit - but a broader point is that we’ve had the ‘techies’ dictate the narrative on vibe coding / software is free thing of late, and now it’s the turn of the ‘normies’ to curb their enthusiasm )

• SEC cyber disclosure rules (Item 1.05 8-K) — in force since December 2023, requiring four-business-day disclosure of material cyber incidents. Boards are now personally exposed.

• DORA (Digital Operational Resilience Act) — live in the EU since January 2025, imposing strict ICT third-party risk requirements on financial entities, with mandatory registers of information and concentration risk monitoring.

• NYDFS Part 500 amendments (2023) — expanded CISO obligations, MFA requirements, and incident reporting for any entity touching New York financial services.

• EU AI Act — high-risk system obligations now applying to AI components in financial, HR, and critical infrastructure software, with documentation and conformity assessment requirements that effectively preclude undocumented internal builds.

• PCI DSS 4.0 — mandatory since March 2025, with materially expanded requirements around custom software, change control, and authenticated scanning.

• OCC/FRB/FDIC interagency third-party risk guidance (June 2023) — making banks accountable for the cyber posture of their commercial customers and vendors, which cascades down the supply chain.

The cumulative effect is that “we built it ourselves with AI” has become a sentence that triggers regulatory, audit, and insurance friction at every step

Insurance is the underappreciated chokepoint

Cyber insurance carriers — Beazley, Chubb, Coalition, At-Bay, AXA XL — now condition coverage on specific controls: MFA on all remote access, EDR deployment, immutable off-domain backups, documented vendor risk management, and increasingly, software bill of materials for critical systems. Unvetted internal builds frequently fail the underwriting questionnaire or trigger coverage exclusions when an incident does occur. For a mid-market firm, losing cyber coverage — or having a claim denied — is often a worse outcome than the breach itself. This is a hard constraint that the “AI makes software free” thesis simply does not address.

The mid-market is structurally exposed, and everyone knows it

JPMorgan last explicitly disclosed cyber spend in 2019 at roughly $600M and 3,000 dedicated staff. Proportioned to today’s tech budget, the run-rate is plausibly north of $1.5B. JPM is constantly under attack and absorbs it as a cost of doing business spread across $170B+ of revenue. A $200M revenue mid-market firm cannot replicate that defense posture. Median ransomware dwell time has compressed from 200+ days a decade ago to under 10 days today, with full domain encryption now achievable in hours from initial access. Recovery routinely runs 21 days of median downtime and 3–6 months to full productivity. A meaningful percentage of mid-sized firms hit with severe ransomware do not survive the following 12 months.

The implication: the marginal mid-market CFO evaluating a vibe-coded replacement for an established SaaS workflow is not weighing 7 bps of margin against a tail risk. They are weighing it against a 1–3% annual probability of a materially disabling event for which their insurance carrier may not pay and their lead bank may use as grounds to exit the relationship. Even if the ‘technies’ say that the vibe code is secure, it doesn’t shift the mindset of the CEO - this is about tail risk and perception, and that’s what will sway his/her decision

The de-banking vector is real and underdiscussed

Under the 2023 interagency third-party risk guidance, banks are accountable for the operational risk introduced by their commercial customers — particularly those with wire/ACH origination privileges or API connectivity. Banks routinely exit commercial relationships citing weak cyber controls, typically framed as a breach of an information security covenant or unacceptable third-party risk. There is no general right to a commercial bank account in the US, and IT security is a defensible, neutral commercial reason that survives state-level “fair access” challenges. A firm whose cyber posture deteriorates because it replaced vetted vendors with internally built alternatives can credibly find itself unbanked, with all the M&A, liquidity, and credit consequences that follow.

Appendix. The following are answers that AI provided in relation to issues raised above about ransomware attacks (as an example) and the risk of de-banking due to risk controls at client firms (if you assume vibing adds risks this is an issue)

1. How easy is it to disable a mid-sized firm with a ransomware attack:

Very easy, and the asymmetry is the whole story. A mid-sized firm (call it $50M–$1B revenue, 200–2,000 employees) is essentially the sweet spot of the ransomware economy — big enough to pay a meaningful ransom, small enough to lack the security maturity of a Fortune 500. The median time from initial access to full domain encryption is now measured in hours, not days.

Why mid-market is the target cohort:

• Security spend gap. Large enterprises run 24/7 SOCs, EDR on every endpoint, segmented networks, tested IR plans, and dedicated threat hunting. Mid-market firms typically have a small IT team (often 5–20 people), one or two security generalists, MSP-managed endpoints, flat networks, and backups that have never been tested under fire. Security budgets at this tier run roughly 3–6% of IT spend versus 10%+ at large financials.

• Cyber insurance coverage but limited self-insurance. They can pay — either directly or via carrier — which makes them economically attractive targets. Coveware and Chainalysis data has consistently shown median ransom payments in the mid-six-figures for this segment, with tails into the tens of millions.

• Vendor/MSP concentration. A single compromised MSP (Kaseya 2021 being the canonical example) can detonate hundreds of mid-market clients simultaneously.

The actual attack timeline — and why “disable” happens fast:

The modern ransomware kill chain for a mid-sized firm typically runs:

• Initial access (minutes to days): phishing, a compromised RDP/VPN credential bought on a broker market for $50–$500, or exploitation of an unpatched edge device (Fortinet, Citrix, Ivanti, SonicWall vulnerabilities have all been mass-exploited in the last 24 months). The “initial access broker” economy means attackers don’t even need to do this themselves.

• Privilege escalation and lateral movement (hours): tools like Mimikatz, Cobalt Strike, and increasingly legitimate IT tools (AnyDesk, ScreenConnect, PowerShell) used to look like normal admin activity. Most mid-market firms have flat Active Directory and over-privileged service accounts — domain admin in a few hops is the norm.

• Backup destruction (hours): this is the step that turns “incident” into “existential.” Modern ransomware crews specifically hunt Veeam, Commvault, and cloud backup credentials before encrypting. If backups are on the same domain (which they often are), they’re gone.

• Exfiltration (hours to days): double-extortion is now standard — data is stolen before encryption so the threat isn’t just “you can’t operate,” it’s “your data will be published.” This neutralizes the “we have backups” defense.

• Encryption (hours): the actual ransomware deployment is the loud, visible end. By the time anyone notices, the firm is already down.

Total dwell time has compressed dramatically — Mandiant’s M-Trends data showed median dwell collapsing from ~200 days a decade ago to under 10 days now, with ransomware dwell often under 24 hours.

What “disabled” looks like in practice:

• Operational halt: ERP down, email down, file shares encrypted, VoIP often down (it runs on the same network), badge access systems sometimes affected. Manufacturing firms lose production lines. Distributors can’t ship. Professional services firms can’t bill or access client files.

• Recovery timeline: the headline statistic from industry IR firms is 21 days median downtime, but full restoration to pre-incident productivity routinely takes 3–6 months. Rebuilding Active Directory from scratch alone is a multi-week project.

• Cost: the ransom is usually the smallest line item. IR/forensics ($500K–$2M), legal and breach notification, regulatory fines, customer churn, and lost revenue typically dwarf it. IBM’s Cost of a Data Breach report has the mid-market average around $3–5M all-in, and that understates tail outcomes.

• Mortality: estimates vary, but a meaningful percentage of mid-sized firms hit with severe ransomware fail within 12 months — either outright bankruptcy (Lincoln College, KNP Logistics in the UK), forced sale at distressed valuations, or permanent loss of customer base.

The tie-in to your coverage:

This is exactly why the cyber-as-a-debanking-trigger conversation matters. A bank looking at a mid-sized commercial customer with weak controls is staring at a 1–3% annual probability of a materially disabling event, and if that customer has wire origination or API connectivity, the bank is downstream of the blast radius. It’s also the underappreciated part of the vertical SaaS / vibe-coding bear case pushback — a SaaS vendor that handles your payroll, your patient records, or your payment flows is a much harder switch when the alternative is “build it yourself and become responsible for the security posture.” The compliance and operational-resilience moat is a real asset for incumbents like TOST, BILL, and the payment processors, and it’s why the “AI makes software free” thesis has real friction in regulated/operational-critical workflows.

The uncomfortable truth: most mid-sized firms are one phished credential away from a 30-day outage. The ones that survive are usually the ones that had immutable, off-domain backups they’d actually tested — and that’s still a minority.

2. Do banks “de-bank” commercial clients over cyber and other such risks?

Short answer: yes, and it happens more than most clients realize. Cyber/IT hygiene is a legitimate — and increasingly common — basis for a bank to exit a commercial relationship, though the mechanics depend on jurisdiction, contract terms, and whether the bank frames it as a credit, operational risk, or AML/sanctions issue.

Here’s how it actually works in practice:

The contractual basis is almost always there. Commercial banking agreements (deposit account agreements, loan covenants, treasury services agreements) virtually all contain “termination for convenience” or “termination on notice” clauses — typically 30 to 60 days, sometimes immediate for cause. Banks rarely need to articulate a reason at all. When they do cite cause, weak IT controls usually get framed as either (i) a breach of an information security covenant, (ii) heightened operational/fraud risk, or (iii) an unacceptable third-party/vendor risk under the bank’s own regulatory obligations.

Why IT security has become a real exit trigger:

• Regulatory pressure on the bank. Under the OCC’s third-party risk management guidance (2023 interagency final guidance) and the NYDFS Part 500 cybersecurity rules, banks are accountable for risks introduced by their commercial customers — particularly customers with API/host-to-host connectivity, treasury workstation access, or large ACH/wire origination privileges. A customer that’s been popped, or that fails a cyber assessment, becomes the bank’s problem.

• Business email compromise and wire fraud losses. Banks have absorbed enormous losses (and litigation) from BEC-driven fraudulent wires originating from customers with poor email security and no callback controls. Many banks now require attestations on MFA, email authentication (DMARC/SPF/DKIM), and endpoint controls for customers with wire/ACH origination.

• Ransomware contagion risk. If a commercial customer is mid-ransomware event, banks worry about (a) fraudulent transactions from compromised credentials, (b) OFAC exposure if ransom payments touch sanctioned wallets, and (c) being drawn into the customer’s incident response.

• Cyber insurance and the bank’s own underwriting. Lenders increasingly require evidence of cyber controls as a condition of credit, particularly for asset-based lending where collateral monitoring depends on the borrower’s systems.

What “de-banking” for IT reasons typically looks like:

• A questionnaire or assessment (often via a third party like SecurityScorecard, BitSight, or a bank-specific KYC-cyber overlay) flagging the customer

• A request for remediation within a defined window (90–180 days is typical)

• If unaddressed: restriction of services first (loss of ACH origination, lower wire limits, removal of API access), then full account closure with notice

• Loan facilities may be handled separately — banks are more reluctant to call a performing loan, but won’t renew at maturity

The nuance that matters for your clients: there’s no general legal right to a bank account in the US for commercial entities (this differs from the UK post-Farage, where the FCA has tightened rules on debanking notice periods). Banks can exit. What they can’t do is exit on a prohibited basis — race, religion, protected political activity in some states (Florida and Tennessee have passed “fair access” laws aimed at this, though their reach to cyber-based exits is untested). IT security is a defensible, viewable-as-neutral commercial reason and would almost certainly survive a fair-access challenge.

The relevant angle for fintech/payments coverage: this is part of why the BSA/AML and operational risk overlay has become such a moat for incumbents and a cost burden for challengers. PAYO and FLYW both deal with this on the customer-onboarding side — their SMB and education-vertical customers are exactly the cohort most likely to fail a major bank’s cyber screen, which is part of why specialized cross-border players exist in the first place. It’s the same regulatory-cost moat thesis you’ve been working through on the stablecoin and vibe-coding pieces — compliance infrastructure (including third-party cyber risk management) is a real, durable cost that doesn’t go to zero just because the software gets cheaper.

This note is about CoStar stock (CSGP $36 April 13 2026) and is intended for those with some background knowledge of the stock already. Take a look at this chart below. It shows the 10-year historical relationship between CoStar and the software sector, specifically the valuation, as measured by EV/NTM Sales. The IGV ETF used for this comparison contains the likes of MSFT, PLTR, but also NOW, CRM, WDAY and other SaaS stocks that have been under pressure in recent months or more due to AI fears among other things. CoStar’s valuation is well over 2 standard deviations below its average over this timeframe. The average over 10 years was 1.5x and the stock is now at 0.7x and is one of the most oversold stocks in the S&P500. I do not have a position but might go long for a trade soon

CoStar’s revenue revisions have been positive in the last year and they just did their 59th quarter of double-digit top line growth. Profitability revisions have been down (around 10%) due to investment choices the company has been making in residential (Homes.com). In terms of business mix over this timeframe, back in 2016, three quarters or so of the business was what CoStar has long been known for - commercial real estate data and analytics subscriptions, listings and related services. The other quarter of the business in 2016 was Apartments.com which was in the earlier days of ramping.

Today, 55% of the business is that ‘legacy’ business, one that remains a quasi-monopoly and continues to grow well, with high incremental margins and strong cashflow characteristics. But the 45% which is essentially residential real estate doesn’t quite have the same characteristics as the ‘legacy’ - is that part of why the multiple has declined? This is an important question and indirectly lies at the heart of the activist fight with CoStar. Is the company diluting its business mix too much with a revenue source that needs large investments in branding for a consumer audience that already has a known and trusted portal into what’s going on the housing market - as in Zillow?

The Apartments.com business is a good one and that makes up most of the revenues of this 45%. However the TAM in the residential sales side of the business is going to be higher in the long term if they can make inroads. But it seems pretty evident that this is going to be a long slog - and you have to ask if it is worth taking the business to 55% (and rising) residential in the years ahead - might they be giving up on getting back that multiple which as below averaged 1.5x that of these other software companies?

If you look at Zillow’s multiple - it’s been all over the place in recent years and lately the stock has been hammered - perhaps by fears of a war with CoStar among other things. But even with CoStar’s recent performance, it remains valued above Zillow (see chart below of relative valuation) and in the past 4 years it had a multiple of revenues 2x that of Z. Is this what Andy Florance wants - a lower multiple business with an ‘endless war’ with the like of Zillow (and a couple of others)?

A lot of investors love founder led companies for obvious reasons. Whenever a leader has replayed what seemed like a winning formula over and over again, it adds to the aura of control, their market insight, their sheer grit in achieving goals. But maybe there’s a sweet spot for the founder, maybe it’s when the company is still of a size that can be managed through force of personality and personal drive, when the TAM is mostly ahead of them, and they are generally still travelling rather than arriving. A healthy valuation for a company’s stock is the ultimate validation for a CEO and there’s not a lot a shareholder can say when that’s the case.

They did the Apartments.com deal and made it work; there was some overlap in the institutional CRE ownership of many of the apartments and it was a fragmented market with weaker players. But more lately they have been attacking the big prize - the US residential sales market. With its embedded ways of doing business, entrenched broker system with lots of incentive to keep the status quo (despite recent regulatory driven changes to the model, at least on the surface), and the consumer in the loop, it’s a different animal, and a very large one. When a consumer pursuing a rental unit it’s not the same financial and emotional commitment as buying a home - could it be that persuading the market to adapt a new tool is coming up against genuine consumer inertia. It’s too early to say, it could be that CoStar’s Homes.com is indeed the way the US market will go (as in the same as the UK, Australia etc) but that doesn’t mean other players would stand still.

Having had history with CoStar over the 2011-2018 or so timeframe when we were top holders, I sensed that Andy Florance had a ‘thing’ for Zillow and its ex-CEO and now Chairman Rich Barton. Andy is a colorful figure and fun to share meetings with, so it was all good-natured banter. But maybe the rivalry, to the extent there was one, in part at least, led to CoStar making a strategic mistake.

At the moment, CoStar stock is at multi year lows from a valuation standpoint, so the CEO and Board may have lost that validation from the market. Of course many stocks regarded as high quality are being sold off at the moment. That CoStar is relatively immune from AI is likely a consensus view, and while growth stock valuation comps are lower, a larger part of the CoStar decline relates to increased concern about their investments in Homes.com. That the company chose to consolidate reporting of this into a larger sub segment seems to have upset the activists even more and also could arguably signal a hardening of the company’s attitudes.

What the activists are saying

The following in italics is an AI summary of their argument (I have left out the CEO compensation and Board stuff for brevity).

Activists D.E. Shaw and Third Point share several major concerns regarding CoStar Group’s management, board oversight, and capital allocation, primarily centered around the company’s costly expansion into the residential real estate market.

The main issues raised by these activist investors include:

• The Homes.com/Residential Strategy is Destroying Value: Both D.E. Shaw and Third Point argue that CoStar’s strategy to build a residential real estate (RRE) portal through Homes.com is “ill-conceived,” poorly executed, and destroying billions of dollars in shareholder value. Third Point notes that CoStar has sunk roughly $5 billion into the RRE segment over five years with a “horrendous return on capital,” generating only about $60 million in revenue in 2024 and missing initial targets. D.E. Shaw highlights that CoStar is on pace to spend approximately $4 billion on Homes.com through 2030, an investment that has severely eroded the company’s overall profitability.

• Distraction from the Core Commercial Business: The activists argue that the massive financial focus on residential real estate has diverted resources and management’s attention away from CoStar’s highly profitable core Commercial Real Estate (CRE) franchises. D.E. Shaw specifically points out that pushing the core sales force to sell Homes.com subscriptions distracted them, which negatively impacted the performance and organic growth of the core business and Apartments.com

What should an investor do now?

It could be that the activists got into CoStar stock at the wrong time, the stock looked pretty expensive for quite some time, as did many stocks of course. Although for all I know they hedged well, and in any case these are large funds and CoStar won’t make much difference to their performance. Regardless, a new investor doesn’t have to deal with that baggage, what’s important is what’s priced in today, and whether CoStar stock stands out against the opportunities among similar stocks as there are many that have sold down to multi year lows.

CoStar historically never wants to be #2, especially a distant #2 or #4 for that matter. The US residential market, as big an opportunity as it is, might well be a step too far in applying CoStar’s pattern of capturing share through aggressive investment.

CoStar and its CEO faces an asymmetric risk from this situation - if they show progress in Homes.com it’ll likely be a longer-term thing and the question is if investors have patience/confidence for this. If they continue to invest in Homes.com, and it fails to achieve anywhere near the plan, it could be catastrophic in terms of turmoil at the company. CEO Andy Florance has a bunch of years left at the firm, he’s loves his job. He would have time to grow the business in other ways in the future, the question is whether his ego will allow him to do the “Metaverse Maneuver” that made Meta’s stock rebound significantly when they cut back on capex in late 2022. While admitting defeat on this battle, he could still win the war in the longer term if he somehow ring-fenced the risk that they deploy too much capital into a vanity project.

So where do I stand on all of this? Take my view for what it is - not grounded in weeks of deep research, even though I knew it very well at one point. I have only done several hours of catch-up work over the last few days. But from what I see I think now is an attractive time for me to consider taking a modest position given a valuation that appears attractive, and a technical setup that suggests it is washed out for now.

I don’t think there’s much wrong at CoStar in terms of overall net new sales, underlying competitive dynamics. I do think they are doing interesting stuff with AI, including at Homes.com. Overall, the stock has been hammered and there’s a good chance that it has been overdone - as in it’s not a certainty that Homes.com won’t be a decent business over time - but at this price does it matter as much? As below, CoStar has suffered more from multiple compression than most, yet revisions for revenues are stable and growth is better than average, and valuation on consensus 2027 EV/EBITDA, despite all the hand wringing about the investments in residential, doesn’t seem demanding for this company. So yes, now seems a good time to buy (for me, you do your own DD!)

Beyond the short term, I think they should consider splitting up the company into CRE and residential, that’s my gut feeling. I could change my mind on this with time but in any case, I don’t expect this to happen any time soon so it’s a moot point here but may as well get it off my chest. I know there would be some dis-synergies, but at the same time you wouldn’t have a CRE salesforce being asked to ramp up residential sales and neglecting the CRE even temporarily (that’s been rectified but still, it’s an example of disruption in focus). A standalone residential company would do OK given the strength of the apartment rental business, and without the cashflow from the CRE side, investors wouldn’t worry that they would overinvest for years - ie it would have to be more disciplined. But it would be more focused and if it did badly at least employees in the CRE business would not suffer from seeing their CoStar stock getting hurt due to issues outside their segment.

The CRE side is a high quality quasi-monopoly and would be valued generously by the market. The residential side would likely track Zillow. As the activists have pointed out, even at a stock price meaningfully above this mid $30s price, there’s negative value implied for the residential business. Andy Florance could stay Chairman of both businesses the residential business if he wished - but I don’t think most holders want CoStar to become Zillow when they could just as easily invest in that company.

Bank of New York (BK) on the other hand - it’s one of the most overbought stocks in the S&P (#2 in fact and peer STT is the #10 most overbought). I have an idea as why this is occurring - some is specific to each company’s changing investor perceptions, some is AI disruption (or not) related, some is interest rates, some is the starting valuations, and maybe it’s just one stock going to the high of its traditional range and another going to the lows the lows of the range at the same time. The question is where to put a dollar for the next 2+ years….

This is a ‘performance chart’ in stockcharts.com (so one stock’s relative performance v’s another)

Shows that BK has outperformed CSGP by 500% in a little over two years

Wonder how many knowledgeable investors would have predicted that

The math of a dollar in each

As well as showing the math of a compelling buyback scenario, the appendix details examples of several larger caps that executed buybacks which had mostly positive effects on performance, but offer caution should organic growth go backwards

Global Payments - When Buybacks Meet Bottom of the Barrel Valuation

Few investors have any illusions about Global Payments (GPN) the company. It is not exactly Stripe, or Adyen. It has a lot of work to do in terms of integration; there’s a ton of noise. And the space it operates in is very competitive. But why is this being treated as new information?

You can go back 10-15 years in this space and see constant worries about commoditization of core payments - ISOs, ISVs, modern platform fintechs such as Square, race to the bottom…. the worries are not new. GPN would have grown faster had these players not been around, but then again, the company would have grown so fast the stock would be at 25x sales / 90x earnings or whatever. Instead, it’s been growing modestly but it’s only at 4x 2027 consensus adj EPS and 5x 2027 EV/EBITDA

An investor is tasked with finding the best stocks, and while sometimes these are also the best companies, it’s often not as simple as that. Most people know this, but quants, ETFs and other passives, most retail investors, trends followers etc - these classes of investors are more influenced by past performance than ever it seems. The result is that valuations for many out of favor stocks have compressed to levels which can be shown to be mathematically ‘not well grounded in logic’ shall we say. As we shall see below, the math of buying back stock, given the combination of GPN’s evident cashflows and current valuation, make it as close to a sure thing as an investor can expect. It’s just the math of buybacks, not a call on GPN the company being much different than it is today.

A handful of stocks in the payments space, including GPN, are now at prices which only make sense if it is almost certain that they will be displaced at some point in the medium term. Given GPN’s diversity, both by geography, sub segment exposure, and by customer type, and given their scale, the trust their clients have in them, that just seems like an absurdly pessimistic thought. Plus, as should be evident, the pace of change and the customer turnover in payments is not quite fast enough to displace anyone in the medium term.

There’s no evidence in the numbers that GPN is not growing - yes they may lose some customers over time to the Adyens etc, but they will grow with existing customers and it’s not like they can’t win any new ones. Finally, over this medium-term time frame, isn’t it possible that GPN could upgrade its tech stack to make it more competitive with Adyen among others? After all, software is now free or so we are told (it’s not, but let’s not let the facts get in the way of a good story).

GPN’s earnings quality is quite high in that it doesn’t add back a lot of stock comp expense to adjusted EPS. They have strong cash flow characteristics that you’d expect as a mature payments company with scale. As such this PE of 4.6x is ‘real’ and the PE relative to the S&P seems a bit ‘unreal’ at about 22% give or take (about 26% relative to the equal weighted S&P).

GPN deserved to be re-rated downwards (from the 20s PE) over this timeframe and for sure it was not an appealing stock for a long time. But I think that the above chart of the NTM PE is what happens when the price setting responsibility is taken away from discretionary managers and is handed to trend followers, quants, ETFs and other passives, and the same cohort of sell side analysts and retail investors, who would rather pay 28x revenues for Robinhood at $150, thinking it still has lots of upside, than pay a mid single digits multiple of earnings for a stock like GPN because the trend is the wrong one. This recent phenomenon is perhaps why valuations seem to more frequently go to illogical levels - it’s not because a class of investors has better insights into future numbers, it’s that they just extrapolate trends more than ever. Yet isn’t there a point where it doesn’t make sense to keep adding more and more to the valuation (or less and less as in the case of GPN?) just because the trend has been your friend recently?

There are cycles in investment as we know - the next chart is just one example. Back in 2012 or so, GPN stock was considered to be a legacy player that was going to be hurt by ISOs, ISVs and fintechs. The same competitive worries we see today.

As it happens that was a good time to get long as things changed and the threats didn’t overwhelm GPN - in part because they were not a static company, just like today they are not a static company that is incapable of adapting to change. As below, the stock outperformed the equal weighted S&P500 by a considerable amount over the following several years from 2012. I wouldn’t argue for anything close to the 20+ multiple the stock had in 2019-2021, but you don’t need that to do well. For reference in 2012 it was at 12x, not 4-5x, and PE relative was 80%, not 25%.

Global’s Big Beautiful Buyback

There are two aspects of the buyback program that I will explore. First is the math, which demonstrates why it is highly unlikely that GPN stock remains at this valuation. Secondly, in the appendix, the historical precedent for large buybacks that either did well or in a smaller number of cases, didn’t add value.

Going back to assertion from above, I’m assuming that even in a bear case that the medium term numbers for GPN will be fairly close to consensus given how slowly change occurs on the ground among their diverse mix of clients and territories. CEO Cameron Bready has been at the company in a senior capacity for many years and I don’t see much reason beyond standard risks when it comes to executing on the Worldpay combination. As things stand, the main variable when it comes to the outcome of the buyback program in the medium term is the average price of the stock bought back.

The Buyback Math

Here’s a model of the buybacks and related analysis. The current price is under $64 right now, and arguably if the Iran war ends without $200 oil, the stock could recover and could easily get over $100 in 2026. Indeed another couple of clean quarters which improve confidence could see the stock get well above $100 within the year - that’s actually what I’d expect as my assumption would be that the market won’t allow a growing company to buy back most of its stock in a handful of years. But this is where we are today - and the model assumes a rising average buyback price from here that comes about from the 10% return in line with past S&P returns, so 8% price improvement p.a. and a couple of points in dividend yield.

By 2028 the company would be buying back a quarter of the Jan 2028 starting shares still in issuance. It’s likely hard to do that pace from a practical standpoint but this is just an exercise in the math. Indeed if the stock was somehow stuck at $64 through 2030 they would buy back the entire company over that timeframe - that’s not going to happen, they might be forced to take it private (at a premium) at some point if they were shown to be on track to do that. Again, it’s just the math which serves to show how a big buyback and a small valuation can have a material outcome.

Note that I have assumed that ebitda/cash flows don’t improve after 2028 - even under this scenario, which is arguably what the valuation today is suggesting (if not much more pressure than flat). However even with this outcome, the EPS would grow as above at 27%, 38%, 39%, & 57% in each of 2027 - 2030

So what price of buybacks would I think the market would allow, on the assumtions that things are ‘more or less fine’ in terms of GPN’s numbers in future years? Well if I plug $82, $135 $165, $185, $205 into the model through 2030 it would allow around a 10% reduction in shares outstanding each year and EPS would be a nice 20% or so followed by 10-11% in the 2028-2030 period when I assume flat earnings and cashflow growth. But at $205 in 2030 the stock would still be under 9x trailing earnings. And a buyer today at $64 or so would have paid 2.6x the 2030 EPS.

Again, this is all a math exercise, it’s not a prediction. But as I like to say ‘math works’ and opinions are a dime a dozen.

The “Buybackers” - evidence from past big buybacks

The following is a sample of stocks that have a philosophy of buying back stock. The actual detail and nuance behind these have been appended in an appendix as it is AI derived. Note that GPN’s valuation is lowest by some way today, none of these companies could buy back all their stock through free cash flow in 4 years as GPN can (at current $64).

Appendix - the following are stocks that have or have had, a philosophy of aggressive buybacks

The following is what Anthropic’s Claude came up with when asked about the history of larger US stocks doing aggressive buybacks over multiple years. Everything from hereon is AI generated:

The “Mount Rushmore” of Aggressive Buybacks

1. AutoZone (AZO) — The Gold Standard

This is the comp everyone should study. AutoZone has reduced its share count by approximately 89% since 1998 Yahoo Finance — from 188M shares to under 17M today. Since 2020, the company has been buying back $3-4B annually, reducing its share count by nearly 8% per year, and by about 6% per year since 2015. The Investor’s Podcast Network

The results? EPS has compounded at a 17% CAGR over the past decade, meaningfully outpacing total earnings growth. Yahoo Finance The stock has gone from ~$150 in 2005 to over $3,500 today — a 20+ bagger.

Why it worked: AZO has a genuinely asset-light, recession-resistant business model with predictable FCF. The absence of cyclical forces in the auto parts industry frees the company to channel virtually all FCF into buybacks. Yahoo Finance Management never got distracted by acquisitions or empire-building. The consistency of execution over decades created a compounding machine where the buyback itself became a core competitive advantage for per-share value.

Relevance to GPN: High. Post-Worldpay, GPN is pivoting to a “simplified, FCF-generative compounder” narrative. The question is whether the payments business has AZO-like predictability — I’d argue it’s close, though more macro-sensitive.

2. Dillard’s (DDS) — The Sleeper Hit

This is the comp that most closely matches the GPN math. Dillard’s entered 2020 with about 24M shares outstanding at a stock price around $30 — a market cap barely above $700M for a company generating $250M+ in FCF. By mid-2022, total shares had dropped to 17.2M from 20.7M a year earlier. sec By end of FY2023, Class A shares outstanding were down to roughly 13.1M. Today the company has about 15.6M total shares outstanding CompaniesMarketCap (post-issuances), representing a reduction of roughly 35% from the 2020 base.

The stock performance was extraordinary: DDS returned 306% in 2021, 39% in 2022, and 32% in 2023. The 5-year total return CAGR is about 72%. FinanceCharts

Why it worked: DDS was buying back an enormous percentage of its float at absurdly cheap prices. When COVID hit, the stock dropped to $30 while the business was generating meaningful FCF. The Dillard family, which controls the company, treated it like a private entity — minimal communication with investors, maximum buyback intensity. They didn’t care about quarterly guidance theater. They just kept hoovering up stock.

Relevance to GPN: Very high. GPN’s setup is remarkably similar — deeply depressed valuation, strong FCF, management signaling massive buybacks. The key difference is that DDS’s operating business was also inflecting positively (gross margins expanded from 29% to 43% during this period), creating a dual tailwind. GPN needs the Worldpay integration to go well to get that second engine.

3. eBay (EBAY) — The Asset-Divestiture Buyback

This comp is structurally the closest to what GPN is doing. Between 2017 and 2020, eBay reduced its share count by roughly one-third Seeking Alpha, funded by a combination of operating FCF and asset sales (StubHub for $4B, classifieds for ~$9B). At one point the buyback yield was running at 14.4%, which combined with a 1.7% dividend gave a 16.1% total shareholder yield. Seeking Alpha

Stock performance was solid but not spectacular — the stock roughly doubled from $28 to $55+ over 2017-2020, then rallied further to $70+ in 2021. It’s now around $72.

Why it worked (mostly): eBay used asset sales to fund an accelerated buyback at a time when the remaining core marketplace business was undervalued. The structural similarity to GPN — divesting non-core assets and redeploying proceeds into buybacks — is striking.

Relevance to GPN: Extremely high. GPN is following almost the same playbook — selling Issuer Solutions, selling the payroll business, and reinvesting proceeds into the core merchant business + buybacks. GPN explicitly said the payroll sale proceeds went into a $500M ASR that was “incremental to the $7.5 billion of capital returns we are targeting between 2025 and 2027.” Global Payments Inc.

The Cautionary Tales

4. IBM — The Anti-Template

IBM spent roughly $110B on buybacks since 2000, amounting to nearly 80% of its market cap at time of writing. Shares were up only about 30% over 16 years, massively trailing the S&P 500. The Motley Fool From 2000 to ~2015, IBM’s market cap actually declined by 21%, even as the stock price rose ~44% — the entire return was engineered through share count reduction rather than value creation. The Liberty Advisor

Why it failed: IBM used buybacks to mask a business in secular decline. Revenue was essentially flat for 15 years while the company underinvested in R&D relative to peers. R&D spending as a percentage of revenue was roughly 6%, compared to 12-14% for North American internet and software peers. The Motley Fool The buybacks were eating the seed corn.

Relevance to GPN: This is the bear case. If GPN’s merchant acquiring business faces genuine secular headwinds from modern platforms (Stripe, Adyen, Square), then the buyback just delays the inevitable rather than creating value. The IBM lesson is clear: buybacks only create lasting value if the underlying business is stable or growing.

5. GE — The Worst Case

GE spent $46B on buybacks over a decade, including $24B in 2016-2017 alone, while the company’s debt load ballooned and its cash flow deteriorated. CNBC The stock collapsed from $30+ to $7. Buying back stock with borrowed money while the business is deteriorating is a recipe for disaster.

Relevance to GPN: Limited, but worth flagging the leverage angle. GPN carries $22B in debt against $8.3B in cash, and the debt/equity ratio is ~0.93. StockAnalysis The company needs to deleverage post-Worldpay. Management has to balance buyback aggression with balance sheet repair.

What Separates Winners from Losers?

Looking across 20 years of aggressive buyback programs, the pattern is clear:

• Buybacks work best when: The stock is cheap relative to intrinsic value (low P/E, high FCF yield), the business is stable or growing, FCF is predictable and recurring, and management isn’t leveraging up recklessly to fund repurchases.

• Buybacks destroy value when: They’re used to mask declining operations (IBM), funded by excessive debt in a deteriorating business (GE), or executed at peak valuations.

• The “cheap stock + stable FCF” setup is the sweet spot. AZO, DDS, and NVR all bought back stock aggressively at reasonable valuations while their core businesses remained healthy. NVR has reduced shares by ~7.5% per year while maintaining 33%+ ROE and minimal debt StockAnalysis — another textbook compounder.

Applying the Framework to GPN

Here’s what makes the GPN setup compelling and what makes it risky:

The bull case writes itself: At ~$64/share and ~275M shares, the company’s $7.5B buyback commitment could retire 40%+ of the current float if the stock stays anywhere near here. Adjusted EPS guidance for 2026 is $13.80-$14.00, implying the stock trades at ~4.6x forward earnings. If the buyback reduces the share count by even 25% over the next 2 years, and the business grows EPS at 13-15% annually, you could see EPS approaching $20+ by 2028 — and even at a modest 10x multiple, that’s a $200 stock. The company now processes $3.7T in annual payment volume across 6M merchant locations globally, which is genuine scale.

The risk factors:

• Post-Worldpay leverage needs to come down. The company has to deleverage to ~3.0x — that will constrain buyback pace.

• Integration execution is non-trivial. Worldpay has been passed around like a hot potato (FIS bought it, spun it to GTCR, now GPN). Each transaction introduces culture clash risk.

• The AI disruption / “software is free” overhang. Investors are worried that modern platforms can eat into GPN’s SMB merchant base. If organic revenue growth disappoints, the buyback becomes an IBM-style treadmill.

Net assessment: The historical comp set is actually quite favorable for GPN. The closest analogues — DDS, eBay, AutoZone — all generated strong shareholder returns when they bought back 20-40% of their float at depressed valuations. The key differentiator between success and failure is whether the core business holds up. If GPN executes the Worldpay integration and delivers MSD revenue growth, the buyback math at these prices is genuinely extraordinary. The stock is priced as if the business is in decline — if it’s not, this has the hallmarks of a DDS-style setup where you’re buying back a third of the company at trough multiples.

The most important thing to watch isn’t the buyback itself — it’s the organic revenue trajectory over the next 2-3 quarters. That’s what separates the AZO outcome from the IBM outcome.

Gemini’s Contribution :

The Tech Juggernaut: Apple (AAPL)

While Apple doesn’t have the “cheap” valuation multiples of a traditional value stock, its capital return program over the last decade is unprecedented in sheer size.

• The Buyback: Starting around 2012–2013, Tim Cook initiated a massive capital return program. Over the last decade, Apple has retired more than 35% of its outstanding shares, spending hundreds of billions of dollars.

• The Result: Apple’s core business grew, but the stock’s astronomical rise over the last ten years was heavily turbocharged by the shrinking share count. Every dollar of Apple’s net income is now divided among far fewer shares, making each remaining share intrinsically more valuable.

I hold several names in the payments space, with one of them (FLYW) being a fairly concentrated position. Smaller holdings include GPN, PAYO, PAY, BILL, and RELY. I’m underwater on BILL and PAYO, their valuations now only make sense if they are static companies unable to adapt to change. The space has not been the easiest to make big bucks in the recent past even if there’s been some good trades. Let’s put it this way - if you pitched most of these names to your typical ‘Robinhood Dude’ you’d be accused of being an out of touch boomer. I’m an X’er who grew up thinking valuation does matter, but those retail ‘hood guys like ‘exciting’ stocks and valuation doesn’t factor into their thinking, broadly speaking. With ‘exciting’ Space X, OpenAI, Anthropic, Databricks and others looking to tap the equity markets in 2026, competition for allocations among stock funds is heating up, even if they initially come with modest floats it could be a typical years’ worth of IPO dollars in a small handful of deals. And then when lockups expire there could be more supply. Maybe these funds don’t come from small caps, but Visa, MA, the big banks, maybe even the likes of GPN - valuations tend towards fairness over time but this crowding out effect could be something to think about this year at least. Although maybe the market is already factoring much of this into stocks today? And arguably the funds will come out of other mega caps given many funds’ need to avoid concentration - many firms have to abide to certain industry rules on this topic.

As for the starting points of valuations for the broad fintech and payments space, I think many of these stocks are trading at levels where high returns over the next few years are possible - that would require a shift in sentiment from Mr. Market, but such shifts are normal.

The title, borrowed from Animal House, which likely borrowed it from Oliver Twist, hints at the past few years’ environment where many names in the payments space have endured repeated beatdowns. Sometimes it was the companies’ own doing, other times sentiment was hit by peers, macro factors including policy and inflation (there’s always macro!), endless wars, AI fears hitting software broadly (more to the damage in those stocks than just AI), strategy pivots, unwinding mergers, management shake-ups, rapid shifts in the payments landscape, and bigger spending by VC-funded startups. Forward growth estimates have also been trimmed in places, never helpful when happening, and many investors who acquired some of these stocks in 2021 and 2022 have been badly burned. The specific challenges faced by larger companies such as PayPal, and Fiserv serves to confirm investors doubts about the broader space, and anchor stocks like Visa/MA have fallen to the very bottom of their historic valuation range v’s the indices recently

So, plenty of thrashings in parts of fintech/payments, but now the question is - are these ‘cheap’ names cheap for reasons that will prove enduring or is the above just a collection of headwinds that’ll prove to have been more cyclical. Have growth expectations been reset enough and can most of the payments names deliver ongoing growth and/or margin upside (or at least stability)? How does all this compare to the broader market and how do the valuations, especially on a relative basis, look in relation to this outlook?

My bottom line: broadly, these stocks are attractive versus the market and could outperform meaningfully simply by hitting 2026 targets and keeping 2027 estimates intact. I’m not going to dissect each company’s moats and detail where the growth is from as that would be unwieldly, but I will focus on a handful of names as examples - Visa, Global Payments and Flywire.

Let’s start with Visa / MA - the anchor payments names.

Koyfin is a resource that I use a lot. It allows you to chart valuations over time including relative multiples - it even works well using equal weighted indices (RSP/QQQE).

To me, from the first chart, Visa appears 12% cheap (22.8x) v’s its own typical NTM PE from 2014 - 2026 of about 26x (ex Covid and recovery). It is down 18% from its closing weekly high of 2025, and the multiple has declined about 24% from that high given growth. In light of recent developments in terms of the Iran war and the AI concerns that affected the software space, this is what I would call on the high side of a plain vanilla correction. Although travel around the affected areas such as Dubai has been severely hit, substitution is going on to some degree, airfares are higher (helps Visa), FX volatility is high (normally good for eps), gas prices are up (helps Visa) and buybacks will go further at this price (worth 75 bips to eps growth annualized). As things stand cross border fees may slow some but there’s likely offsets and I doubt there will be much delta in forward guide.

The Nuance - Is Visa Cheap? How should you Measure Cheap?

The above is in absolute terms, many investors are focused on relative performance and hence relative valuation, and this is where things are more nuanced, and it applies across the market as most appreciate. As in the next chart, over time, Visa’s relative PE v’s the S&P500 has been almost 150% on average. At my old firm I would regularly trade the ranges when it was 15-20% either side of this. In times of stress the market often gravitated towards the quality of V/MA and they would get to 175% (200%+ for MA) of the S&P multiple, or maybe as low as 130% at times. So today at close to 110% Visa is cheap relative to its historical range - in fact getting back to say 150% would lead it to outperform the market by 36%, not too shabby for what is a lower risk stock. But wait, there’s more….

We all know the market is concentrated, not that it matters in some regards, but maybe it does matter to many. Maybe the Mag 7 - soon to be Mag 10 or 11 perhaps, are just different, at least for the time being. So charting Visa’s forward multiple in relation to that of the equal weighted S&P reveals that it traded at about 160% on a median basis over time, but right now it is at 130% or so. Going back there from here would yield a 23% relative outperformance, so somewhat less than the 36% above. But I think the main observation in this chart is that it shows you that Visa’s valuation has not yet had what would look more like a structural breakdown in terms of what the market was willing to pay - vs the typical stock as per the equal weighted indices. It’s possibly just at the low end of the range especially if you consider that the 2020-2022 era was distorted by Covid and the recovery thereafter.

Visa is also relatively cheap against its history v’s the XLF or the bank index but interestingly against the equal weighted Nasdaq 100 (QQQE) it’s about in the middle of the range of the last three years - perhaps due to the collapse in SaaS stock multiples in recent months. In fact if you chart Visa’s NTM PE relative to a software ETF such as IGV - as below - Visa is at the highest relative valuation in 10 years, being at about 1.05x compared to a median of about 0.8x.

So is Visa cheap / mispriced? It depends what you are comparing it to of course, but on balance I’d say it’s relatively attractively priced here, but it’s hardly being given away, and if you like cheap, there’s maybe better candidates….

=====================================================================

Next up is Global Payments (GPN)

This one reminds me of those historical animated maps of Europe which shows how the borders changing dynamically over the centuries - there’s just been lots of change going on here. But maybe it’s like when East and West Germany recombined, maybe the GPN- WorldPay merger will result in a stable and growing entity which although won’t be the most exciting enterprise, could become a grinder of a stock that gets rewarded over time.

When people talk about the payments space being cheap, they likely think of GPN among others. Look at this consensus valuation data from Koyfin - if it stays here the stock will be at 4x consensus adjusted EPS in twelve months time. As several other analysts have opined about recently, this is a cheap stock worth buying because even if you don’t think it’s one of the great companies, does it really need to be at this price? And is this just the opposite of the ‘Robinhood dude’ who ignores valuation on a stock that has gone to heights that are likely 95% sure of losing him serious money over the next 12m+ (eg HOOD at $150 last year). I just think that the diversity and slow moving nature of client relationships means that regardless of whether they can do $4bn or their stated goal of $5bn in FCF by 2028 - that’s just going to allow them to buyback a crazy amount of stock unless the valuation expands meaningfully. It could still remain relatively out of favor, and relatively cheap vs history if it doubled and they achieve more or less what they are telling us.

There’s a lot of noise in the numbers at this company and you have to take some things on faith - such as the $5bn in free cash flow by 2028. That’s a huge number relative to their sub $20bn market cap. But perhaps investors have discipline about needing to be able to see through the numbers in their due diligence. I admit in my own case that I don’t have any edge in the numbers here - but the current run rate numbers and the expected synergies make sense to me or at least don’t appear crazy.

Plus I think I can see why sentiment is so bad and I’m therefore willing to look ahead to a time when names like GPN settle back into a phase where they grind revenues modestly ahead of nominal GDP, they buy back stock at very low multiples, and they use AI to help them grow efficiently as well as help them improve their software/tech stack to be ‘less legacy’ and more cloud and AI native. This was the basis of my note from late January - even starting at $75 and going up 10% a year they could buy back half their stock by the end of 2030. Is the market really going to allow GPN to do this?

Maybe it’s the longer dated risks that concern you - maybe you think somehow, AI in all its forms will just eat a company like GPN. I could suggest the same thing about Nvidia - after all it’s a digital-only company that designs stuff using software, and eventually AI will do it all! You’d likely say ‘no way’ - and you’d be right, AI’s ability to write software will cut the software cost part by 50% or a bit more, but that’s it (we’re talking enterprise here not some weekend vibe coded project). There’s a lot more to running a company than producing software, which is a modest part of the total expense of such companies. AI won’t directly process payments - AI is a probabilistic model rather than deterministic and is not suitable for payments, plus it’s not like payments don’t work well today, it actually seems to work very well in my experience as a user, and it’s fairly cheap already. Sure there will be change, but the payment landscape is getting more complex not less, so leave it to those with domain knowledge - leave it to the likes of GPN.

Here’s the bottom line - look at this chart. It shows that over time GPN has grown revenue per share (blue solid) at about 6% cagr since 2016, and since then ebitda margin has gone from mid 30s to high 40s%. Meanwhile the valuation (right most y-axis label) has gone from about 12x NTM PE in 2014, then it went up nicely to the 20s, and now it’s gone to 5x (v’s market in the low 20s). Note that 2024 dip in rev/share was due to divestitures. I get that there’s lots of noise here, but this is nuts.

=========================================================================

Last but not least…. Flywire

The next one is FLYW, which is my #1 position by some way. I have done a lot of work on this name, and so far, my numbers have been pretty accurate, as such I would say that I have some edge in this name, mostly because the Street was too lazy or consensus hugging to do any real work on it. Some stocks like GPN may be super cheap because they are going though yet another period of transformation, some other like FISV are show me stories (that one seems more stable recently fwiw) and yet others such as PYPL just seem to be up against large trends that are against them and they are too cumbersome to pivot fast enough. Other like BILL have small clients where switching friction is not huge and where churn is a constant issue. But Flywire - it’s super cheap and growing well, executing well, has very sticky clients and is winning new clients and share of wallet at existing ones. It’s developing new products with real momentum, expanding margins meaningfully, it is more diversified than ever…. I could go on. Even after a year of so called headwinds in 2025 as student visa counts were down in a couple of markets, they did 16% organic growth. They are guiding (prudently in their words) for organic growth of 17% in 2026, however I can see 22-25% in 2026 per my math and that’s likely their goal.

Below is the consensus valuation taken from Koyfin. It’s practically at the same EV/EBITDA as GPN for 2027. Now I would point out that stock comp add back inflates that EBITDA for FLYW such that the valuation is just over 8x when it is included as an expense that it is. On the other hand that is based on a margin estimate that is perhaps 10 points lower than it could be in a handful of years, such that it would totally offset that SBC add back to expense. Plus in a few years the stock comp impact on revenues will be less than it is today.

It might seem reasonable to let GPN trade at 5-6x for a while as nobody is going to acquire that company. But FLYW is totally buyable for both PE and a host of globally oriented fintech and payments names, even the likes of SOFI could totally justify acquiring it strategically and financially. As such it could really be trading at 3-4x post synergy ‘27 EV/EBITDA here which is nuts for such a high quality company

Consensus FLYW Valuation per Koyfin

I have written a couple of pieces on FLYW starting with my initial Substack contribution in early January this year. I think the fundamentals are solid, the headwinds from student visas hurting some revenues are largely in the past, the mix of business has evolved a lot towards other revenues in any case, management is strong, market positions are ‘nichey’ and don’t attract lots of competition. And yet the stock is priced for no or very slow growth rather than the 20% give or take that I can see for multiple years.

Is there an outsized threat from AI in any way? Of course every company can be impacted especially if they sit around and don’t do anything about it. But the vast majority of FLYW’s revenues comes from monetizing their vertical software using their global payment rails. For example in their biggest vertical, education, they are deeply embedded into client systems of record. They provide the international payment capability, and increasingly the domestic payments, they have several products that form a bundle such as collections, financing (loans to students via third parties), and various other products to alleviate burdens at the bursar’s office. They have several year contracts, and anyone wishing to compete would have to have access to the client systems in order to write the code. But then again is that newcomer going to vibe code a salesforce, and a support team. Are they going to vibe all the licenses and banking relationships in multiple countries, are they going to vibe their way to global compliance? Are they going to vibe their way to domain expertise? Let’s think for ourselves here people, it’s crazy to think FLYW’s education business is at risk from AI or AI generated software. If you think that then you don’t know the culture of a bursar’s office. I had the privilege of helping to audit the bursar’s office of one of the richest colleges in the UK, (Trinity College, Cambridge) when I was a young trainee accountant and I can tell you, those colleges move slowly, and a college choosing the established, trusted, proven provider in a Flywire is a no brainer, nobody else stands a chance. (The quality of wine at dinner there was exceptional by the way - helps to have relationships with the best French vineyards going back centuries).

What about the rest of the business? Travel is almost a quarter of the business. Again it largely, and certainly increasingly, monetizes via payments revenue. There’s not much if any per seat type revenues at Flywire, it’s more per institution and by volume of payments. If a hotel decided to vibe code a solution to replace Flywire, they would still need a payment provider, plus now they have ongoing maintenance and security concerns and they are distracted from doing other stuff that’s more revenue focused - and for what - to ‘try’ to save 5 bips. Or maybe it’s a net cost of 15 bips extra v’s Flywire - software is never free especially at an enterprise, especially when it eats up valuable time elsewhere, don’t believe that hype just yet!

There’s a reason that client churn in Education and Travel is 1% - think about that, 1%.

Meanwhile their healthcare and B2B businesses - same as above, why would they put themselves through this exercise to try to save 5 bips, if they can. What about switching costs, opportunity costs.

And then there’s this - why can’t a firm like Flywire not be a serious beneficiary of AI, both in terms of supplying clients with additional features that they might even pay for, or at least adds to the bundle they find attractive. And what about Flywire’s ability to more quickly add adjacent verticals by building instead of buying - after all they already have the global payments infrastructure, a good brand, all the licensing and compliance backbone, is the market focusing only on the risks, however remote, yet ignoring the fact that AI is making companies like Flywire both more efficient and more capable of moving fast in entering new verticals.

I’m not saying it’s likely, but I would not be shocked if Flywire’s valuation went to where it was pre the student visa issues, so 6x rather than 2x.

After all, since 2023-24, margins have more than doubled and are going higher (given incremental margins at 35-40%).

Plus, unlike many SaaS stocks and indeed some payments names, they can reduce share count through buybacks as they stock is so cheap that these more than offset stock comp related issuance.

============================================================================

This comp table is derived from data downloaded from Koyfin - it is focused on FLYW, but it’s clear that many payments stocks trade at MSD 2027 (FY2) valuations. I would just point out that many companies here have already reset numbers somewhat (FISV, PYPL, FOUR, RELY and although you wouldn’t just buy because they appear this cheap, they likely are this cheap in reality. Flywire’s consensus growth below (17%, 15%) appears too low by a few points to me, and I can’t see why you would put it alongside these more challenged Payments names. This is why I selected some SaaS names as comps - those with similarities including growth, margins, revisions.

In the above, FLYW is compared to select payments names and various SaaS names separately. SaaS is much more highly valued despite the correction the space has had. EV/gross profit in the best equalizer for these stocks and SaaS is 3.6x the multiple of FLYW (not 3.6x added multiple points but 3.6 times more expensive!) - and these are names that were chosen because they have similar growth and margin profile v’s FLYW. Even the selection of payments names is more than 2x that of FLYW, and on average they have a lower growth profile.

Using ‘headline’ EBITDA, the SaaS names are more than 4 times more expensive than FLYW using 2027 consensus (and I think that is way too low for FLYW fwiw). However the stock comp is proportionately higher for SaaS and adding it back as a true expense puts FLYW at higher margins than all but one of these names (third from right column). Finally, FLYW has had positive revisions of late that are much more robust than SaaS and certainly more than payments. I think there will come a point of recognition that FLYW has huge upside potential, investors need to start looking beyond the mega caps perhaps

S&P Valuation for Comparison / Consideration

And then there’s the market multiple -it has come in a bit recently. This from Koyfin is likely a bottoms up aggregate which yields a NTM PE of 20.2. Using top down yields a 2026 PE of 22x at the moment. The median growth rate of the stocks in the S&P is 6.6% on a 12m trailing basis. Top down calls for total earnings growth of 12% for 2026 and 2027 fwiw. When looking at these single digits multiples for these payments stocks ask yourself if they deserve to trade at half or less of the market multiple, many are priced lower than banks on a PE basis. It’s a bit topic so will leave it at that.

Do your own due diligence obviously but the above explains why I am concentrating my payments/software bets in Flywire.

The company adopted a prudent approach at that initial guide with an fx neutral organic growth assumption of 10-14%, and despite beats in 1H they waited until the 3Q report before increasing estimates for the year. In the event, 2025 ended with over 16% organic growth. When I look at a download of growth rates for the S&P 500 constituents in the last 12 months, the median top line growth is 6.6%. The top decile of growth starts at 21%, and only 70 S&P500 companies grew faster than Flywire in 2025, and yet this was their year of headwinds. What does that tell you about the company - I think it tells you that it remains a growth company, and the initial 2026 guide, which is in their own words ‘prudent’, suggests that investors will have to look at Flywire as a growth stock again (top decline of that S&P sample), and that has significant implications for the stock’s valuation.

I wrote up the case to buy the stock in January and following this linked report I have made a few updated remarks below - nothing much has changed since early Jan aside from the fear of AI etc - the stock is a gift here still

1. The guide for 2025 was prudent and they came in 4% ahead on the top line. The guide for 2026 is said to be prudent per the CFO - I firmly believe him. Specifically, the guide for US student visas issuance is -30%. That is likely a $10m headwind or 1.7% v’s those visas remaining flat. My own estimate is half that fwiw. They are ramping up the investments in go-to-market in order to capture the Sertifi synergies but there wasn’t much reference to these synergies on the call which stood out to me. These synergies amount to a hefty $150m-$180m of revenues over time (at lower gross profits of course). That would translate into a material amount of EBITDA as incremental margins ought to be quite high on that. The EV has already been impacted by the $340m of cash for the Sertifi purchase but most of the EBITDA synergy is in the future.

2. The company seems to be benefitting under the CFO (started in ‘24) in terms of profitability. Despite the dramatic fall in the stock in 2025, here’s something worth thinking about - the consensus estimate for the 2026 EBITDA before the stock had its big downdraft in early 2025m was $160m, it went to $138m by mid year but now it’s back at $158m. I think they will exceed this. The consensus growth in EBITDA for ‘26-’28 is +31%, +26%, +29%. In effect, from a profitability expectations standpoint, nothing has changed due to the 2025 headwinds they had. Yet the stock is very depressed still - how long will it stay mispriced?

3. If you plug FLYW’s basic growth metrics into an AI chat box and ask what multiple it should trade at based on the characteristics - its spits out 10x EV/Gross Profit or 28-32x EV/EBITDA. OK maybe that’s generous right now given the ‘AI trade” implosion of SaaS etc. But look, FLYW is trading at 4x 2028 consensus EBITDA - and I think that’s likely too conservatively modelled, we’ll see. Even if you adjust for stock comp and include that as a genuine expense that it is, the multiple would be 5.7x. So yeah maybe 30x is a stretch, but what about 20x or more, that seems very reasonable to me, and if reasonable you can see why you could triple your money in this stock.

4. What about FLYW getting acquired - I don’t expect that, I think they have enough growth ahead of them so there’s no compulsion to sell. But what if the multiple never recovered? That EBITDA consensus of $255m in 2028 would likely be $400m after synergies in a deal. Might someone pay 12x the post synergy EBITDA ($40) for a company that can growth well above the market median growth rate?

5. Looking ahead a couple of years, people worry about the economy. Maybe credit fears will increase, maybe interest rates will drop as AI induced unemployment picks up? What do people do when they can’t find a job - often they go to college, or do a masters - let’s assume it’s not a depression where families can’t afford such things as college. FLYW is still 2/3 exposed to education - I’d bet that it is less impacted by the economy than most things. And their healthcare revenues - unaffected mostly, even their travel is higher end consumers, certainly for the core luxury travel - it’s often rich people going on safari or heli-skiing, they may cut back some but hey maybe they will also have more time on their hands (they’ll still have the funds is a fair guess). As such, FLYW is a fairly safe stock to hold in a recession - if rates went lower in a weak economy it’s the kind of smaller cap growth stock that might get chased higher by that all else equal.

6. Looking at 2028 - how can you lose money if it’s really on 4-6x EV/EBITDA, isn’t this company quite diverse now, hasn’t the student visa reset already been quite severe? Maybe the medium-term risks are to the upside for some of those territories in terms of Visas - certainly the weaker US Dollar could be helpful for demand for the US. Those EBITDA figures don’t even assume mature margins (or margins after being acquired). They just did 16% organic in their worst year, they are still small with huge TAMS, they still have most of the Sertifi synergies ahead of them, they have ample net cash, they are cash generative, their stock comp dilution is normal for such a company (3% pa is their target albeit at this valuation they can retire more stock than they issue), and they should be quite resilient against the broad threats from AI. I will be doing a separate thought piece on the AI trade as it might impact payments names - I’m quite sanguine about the threat is my 2c opinion.

7. But on that topic - yes Flywire is a software centric name. But they are very deeply embedded into client workflows, and they have several products which bundle together making the whole package very hard to displace. They talked on this call about having a 1% churn rate in education and travel - compare that to other names which lose 10% of revenues every year to churn. In education, FLYW increasingly does the domestic payments flowing into the schools, as well as international, they have products that help with collections and also payment plans (funded by third parties), they are increasingly rolling out a more complete suite of software for the Bursars’ office that satisfies the expectations of today’s students. These products are all integrated and deeply embedded into the master system of record. These are already efficient and Flywire provides convenient and attractive pricing to families sending large sums of money cross border. Flywire has a payments infrastructure that is hard to replicate - they offer hedged fx rates to families so the correct amount gets settled at the school, they have many integrations with local banks allowing easy and confident money transmission, and they are trusted by schools and colleges everywhere. That’s all hard to compete again, certainly if you want to make a return.

Big Picture - Estimates are on the Rise

1. Despite all the noise in 2025, consensus expectations for Flywire’s 2026 EBITDA are just about back to where it stood before the ‘25 guide that caused the stock to decline.

The EBITDA estimates are also rising as below and the 2026 expectation is about where it was before the 2025 guide. This reflects the above better than feared macro outcome in 2025, solid execution throughout the business, and a deliberate push to improve aspects of the company’s opex. But we are also seeing the benefits of scale as incremental margins in the mid 30%s are on display despite continued investments

2. The company clearly wants to get back to the beat and raise mantra that everyone likes to see in a growth stock, and 2025 was exemplary in this respect. 2Q25 marked

What to take from the above data?

1. They are back to beating estimates. They always have a reason why the next quarter has tough comps or there’s been some pull forward, it is expectations management

2. Organic growth bottomed in 2Q25, but note it was still 11.6% - for reference, Visa’s revenue growth for all of 2025 was slightly lower than this. And overall organic growth was over 16% for FLYW in 2025.

3. Organic growth accelerated after 2Q25. 1Q26 has easy comps so they are growing 26-30% FXN, with 7 pts from Sertifi, so 19-23% organic FXN (but FX is adding 5% on top of that). But then 2Q has even easier comps. And although comps are not as easy in 2H26 the Sertifi synergies could be coming on strong by then. There wasn’t much talk about the Sertifi synergies on the 4Q25 call - my hunch is that these are largely not in estimates as yet and the company will be happy with that

Future Revenue Estimates

The following shows my revenue estimates - with growth rates and revenue mix. Note that I have US student visas down 15% and the company is assuming down 30% - that’s very conservative and is worth an extra $5m approximately. However I did take down visa growth in Australia to flat, in line with the company which is the prudent thing to do. I have slowed growth in many places from the 2025 pace - including ROW Edu, B2B, the UK. A lot of their growth is derived from market share penetration, new products and territories, which is made easier given their payments backbone has been built out already.

For 2027 I am at $920m top line v’ $828m consensus - I think consensus is failing to model (any) revenue synergies with the Travel/Sertifi space, plus I think places like Canada can grow better from here as it went down so much (although it’s not a material part of the business at 4% of revenues). Essentially I don’t see major drags going forward, so the organic growth expectations are likely to drift from the mid teens to the 20%s, although gross profits may grow a couple of points below revenues for the foreseeable future. The offset to this gross profit dynamic is that EBITDA margins are on a consistent upward path due to incremental margins that are almost twice the 2025 margin of 20%. And stock comp dilution at 3% p.a. could well be offset a lot should the stock stay in the teens, although 3% is quite normal for most smaller cap growth names and it’s offset by the fact that we are not close to operating on full margin potential.

On the subject of EBITDA margins, the company has gone from 6% to 20% over the last three years. They are guiding for 22.5% at the midpoint in 2026, and then 24-25% in 2027. When applying 35% incremental margin to my 2027 revenue estimate over and beyond what they did in 2025, I get ebitda of $230m in 2027 on an adjusted basis, or almost $150m after adding back stock comp. What should a stock trade at when it’s delivering 20%+ organic growth and has structurally rising EBITDA margins? I’d say that 25x is very reasonable especially when you consider it has stock comp added back, and yet the company is far from achieving its mature margin potential. I may go into the valuation in more detail the next time I write on FLYW but I get a target near $30. Much of the sell side are reducing targets because software stocks have gone down - but let’s face it, they tend to follow the stock price in case you haven’t noticed.

Appendix - FLYWIRE’S 2026 GUIDE (this following part is AI generated)

Flywire’s initial fiscal 2026 guidance signals a structural re-acceleration, projecting FX-Neutral Revenue Less Ancillary Services (RLAS) growth of 15% to 21% alongside robust Adjusted EBITDA margin expansion of 150 to 350 basis points. The guide is characterized by an intentionally defensive macro framework—modeling a severe 30% collapse in U.S. first-year student visas—offset by the scaling of domestic Student Financial Software (SFS) and new payment processing ramps in Healthcare and B2B. While bottom-line profitability and Free Cash Flow (FCF) conversion are accelerating, investors must underwrite a structural degradation in gross margins as the company trades margin percentage for absolute gross profit dollars.

FY 2026 Guidance: Key Financial Metrics

Management’s full-year outlook emphasizes operating leverage and GAAP profitability, supported by a normalization of the Sertifi integration.

Metric

FY 2026 Outlook

Key Context & Drivers

FXN RLAS Growth

15.0% – 21.0%

Includes ~1% inorganic contribution from Sertifi and ~2% from B2B/Healthcare payment processing ramps.

Adj. EBITDA Margin Expansion

+150 – 350 bps

Reaching an estimated 22.5% at the midpoint; driven by improved sales productivity and G&A automation.

Adj. Gross Margin

(200) – (300) bps

Compression driven by the early-stage ramp economics of new payment processing contracts. Excluding ramps, core decline is normalized at 100-200 bps.

Gross Profit $ Growth

Mid-teens

Focus remains on absolute dollar growth rather than percentage margins.

FCF Conversion

70% – 75%

Normalized to remove one-time integration and restructuring items.

GAAP Net Income

3x to 4x YoY Growth

Supported by disciplined net dilution caps (~3%) and reducing stock-based compensation to ~10% of revenue.

Q1 2026: Near-Term Velocity

The first quarter is modeled to carry massive top-line momentum, driven by lapping the pre-Sertifi baseline and immediate payment processing yields.

• FXN RLAS Growth: Guided to 26% to 30% YoY.

• Growth Composition: Includes a ~7% inorganic contribution from Sertifi and 3-4 points from the Healthcare/B2B payment processing ramps.

• Currency Impact: Current spot rates imply a material ~4% to 5% FX tailwind for the quarter.

• Adj. EBITDA Margin: Expected to expand 100 to 350 basis points.

Macroeconomic and Segment Assumptions

Management has explicitly de-risked the 2026 model by layering in worst-case geopolitical scenarios across the “Big 4” education markets:

• United States: Modeled for a 30% decline in first-year international visas. Despite this severe drag, total U.S. education revenue is expected to grow in the low-single digits, entirely insulated by domestic SFS penetration and new client wins.

• Canada: Modeled for a 10% decline in visas. However, revenue is guided to grow >10% YoY as the company annualizes past headwinds and expands domestic payment capture.

• United Kingdom & EMEA: Assuming flat visa growth. The region is expected to grow at or above the corporate average, driven by deep enterprise penetration and market share gains.

• Australia (APAC): Assuming flat visa volumes. Expected to generate modest low-single-digit revenue growth while closely monitoring tightening requirements for Indian students.

Critical Skepticism & Gap Analysis

• Gross Margin Dilution: The guided 200-300 basis point contraction in gross margin validates our prior thesis: Flywire’s shift into Travel, B2B, and Healthcare processing (e.g., Cleveland Clinic) fundamentally lowers the margin ceiling due to higher credit card interchange costs. Management claims this pressure is “temporary” and mostly contained to 2026, aiming to normalize in 2027, but execution risk remains high.

• H2 2026 Deceleration: The mathematical delta between the Q1 guide (26-30%) and the full-year guide (15-21%) explicitly demands a sharp deceleration in the second half of 2026. Management attributes this to tougher year-over-year comps and “prudent” macro buffers, but it creates a “show-me” dynamic for H2 organic volume.

It’s not perfect of course and you may miss a big relief rally at times, but Fiserv is a good example - a cheap stock just got cheaper as investors gave up on it (and estimates were reset). This is the purpose of running this prompt - it’s to keep you out of trouble, or maybe at least watch your position size when there is this type of capitulation risk. A pal asked me about FOUR - reports tomorrow. I ran this prompt for FOUR when it was in the $70s and it came up with the same thing it shows below more or less, however the price risk then was maybe down to where it is now and so maybe the valuation leaves less downside risk here - we’ll see. I don’t own FOUR, I have in the past, but it’s become a bit too complex and too much of a battle ground stock, the type that would occupy too much time

Here is what this prompt outputs:

As a Senior Equity Analyst and Behavioral Finance Expert, I have conducted an audit of Shift4 Payments (FOUR). The company is currently in one of the most precarious “Strike 2” positions in the fintech sector.

Step 1: The Audit of Disappointments (Strikes 1 & 2)

Shift4 has spent the last 12 months eroding a once-pristine growth narrative through a series of “noisy” earnings reports and macro excuses.

• Strike 1: The Q2 2025 EPS Miss & Guidance Revision (August 2025). Shift4 reported an adjusted EPS of $1.10, a significant miss against the $1.22 consensus. Despite a top-line beat, the stock plunged 16% in a single session as the market punished the company for margin compression and a perceived “buying of growth” via expensive acquisitions (like Global Blue).

• Strike 2: The Organic Growth Deceleration & Multiple Compression (Jan/Feb 2026). While Q3 2025 was “in line,” the stock began a slow-motion capitulation in early 2026, hitting a 52-week low of $52.65 in February. This was triggered by Deutsche Bank’s downgrade citing concerns over unsustainable organic growth and a 47% year-over-year decline in share value. The “Strike” here is the market’s refusal to reward “in-line” results, effectively pricing in a structural failure of the bull case.

Step 2: The Credibility & Patience Check

• Management Rhetoric: CEO Jared Isaacman and the team are doubling down on a “Free Cash Flow (FCF) Hockey Stick.” They have guided toward a $1 billion FCF run rate within two fiscal years. However, the move to a single-share class and the waiver of Tax Receivable Agreement (TRA) rights in February 2026—while long-term positive—is being viewed by cynical investors as a “cleanup” ahead of a potential sale or a desperate attempt to lure new institutional buyers after losing the trust of the old ones.

• Analyst Sentiment: There is a deep “Show Me” chasm. Zacks recently flagged the stock with a Rank #4 (Sell), and price targets have been slashed from $117 down to the $60–$90 range. The consensus is no longer “How fast can they grow?” but “How much of this growth is actually profitable?”

• Valuation: Despite the price crash, the stock isn’t “deep value” yet. Trading at a forward P/E of ~27x (Simply Wall St data), it still carries a growth premium. If the “growth” part of that equation falters further, that multiple has a lot of room to fall toward a “commodity processor” level (10–12x).

Step 3: The “Strike Three” Scenario

The “Strike Three” event is scheduled for tomorrow: February 26, 2026 (Q4 Earnings).

• The Fragility: The stock is currently sitting on a “trap door.” Investors are anchored to the $1B FCF promise.

• Market Reaction: If Shift4 misses on End-to-End Volume or provides 2026 guidance that retreats from the $1B FCF target, the reaction will not be a standard 5% dip. It will be a Thesis-Breaking Capitulation (20%+). The market will stop viewing FOUR as a high-growth disruptor and re-rate it as a debt-heavy, acquisition-dependent legacy processor.

Final Verdict: High Risk (The Glass Jaw)

Shift4 has used up its “growth” hall pass. The technicals are broken (trading near 52-week lows), and the behavioral sentiment has shifted from “buy the dip” to “sell the bounce.”

Assessment: High Risk (The Glass Jaw). All eyes are on the February 26th report. Because expectations for the “hockey stick” FCF recovery are so high, even a “minor” guidance miss will likely shatter the remaining investor patience, leading to a complete re-rating of the stock’s valuation multiple.

Here is the prompt (you can obviously adjust this to use different definitions of what a ‘strike’ is - it’s totally subjective and may change over time or market type)

Role: Act as a Senior Equity Analyst and Behavioral Finance Expert.

Objective: Analyze (ticker or company name to be provider by user) to determine if it is currently facing a “Three Strikes and You’re Out” risk scenario.

Definitions:

A “Strike”: Any material negative event in the trailing 12 months, including a stock price drop >10% in a single session, a revenue/earnings miss, a downward guidance revision, or a delay in key product launches, degradation in margins from pricing pressure (rather than mix changes).

“Three Strikes and You’re Out” Risk: The risk that, due to prior disappointments, investor patience is entirely exhausted. Consequently, the next negative event (Strike 3) will not result in a rational adjustment, but rather a devastating, non-linear loss in value (capitulation) as the investment thesis breaks completely.

Task Instructions:

Step 1: The Audit of Disappointments (Strikes 1 & 2)

Review the company’s performance over the last 12 months. Look at whether the company beat expectations and whether revenue and earnings expectations rose. See if the stock reacted positively and sustained the move.

Identify specific instances where the company disappointed the market (e.g., missed consensus estimates, lowered guidance, operational errors, made too many excuses).

Output: List the dates and specific nature of these disappointments.

Step 2: The Credibility & Patience Check

Analyze management’s rhetoric following these misses. Did they promise a “turnaround” or “second-half weighting” that is now coming due? Is there a ‘hockey stick’ in terms of growth for each quarter for the guided year?

Assess analyst sentiment: Are sell-side analysts skeptical or “fatigued” in their recent notes? Look for phrases indicating a “show me” story.

Check the valuation: Is the stock still trading at a premium multiple (e.g., high P/E or P/S oe ev/ebitda or ev/gross profit) despite the recent misses? (A high multiple with low credibility increases the risk of a crash).

Step 3: The “Strike Three” Scenario

Based on the above, determine the fragility of the stock.

If the company reports another negative event (missed earnings or lowered guidance) in the upcoming quarter, predict the market reaction.

Would this result in a standard correction (3–5%) or a thesis-breaking capitulation (15–20%+)?

Final Verdict:

Provide a summary assessment using one of the following labels:

High Risk (The Glass Jaw): The stock has used up all investor patience. Any bad news will likely cause a devastating crash.

Moderate Risk (The “Show Me” Story): Investors are skeptical, but the valuation has already compressed enough to provide some cushion.

Low Risk (Reset): Expectations are already so low that bad news is priced in.

In terms of using Claude for Excel to update models etc - this below is by way of example, and it covers PAY’s results last night, it take a couple of minutes or so to get this and I didn’t even load the PDF of results I just asked it to find it on the IR site. Load the prompt below into the excel chat box for Claude (with the model open, it’ll add this tab) and out pops the following. I’m still going over these to ensure there’s nothing weird go on, this one seems ok, although just fyi on occasion Claude seems to do really well building a model and then it’s a bit off at times - keep the excel prompts as succinct and direct as possible, ask the AI for help in that regard as I find when they get too long or have repetitive aspects it leads to errors.

PROMPT to generate the above: Post-Earnings Quarterly Comparison & Commentary

Standalone prompt to run after each quarterly earnings print. Use in a separate conversation. Upload or paste: (1) your pre-earnings model estimates, (2) your consensus data file for this ticker, and (3) the earnings release / 10-Q.

Instructions - enter this into the Calude in Excel chat box opened in Excel

I am providing you with various inputs

1. My pre-earnings model with estimates which is open now - this is the stock in question

2. Consensus estimates for this ticker - you may already have it loaded into the workbook but if not it is most likely in a PDF upload which needs OCR to extract the data. Or if I don’t give you consensus specifically, try to extract the basic consensus information from the web (start with investor relations website)

3. The company’s earnings release, slide deck and/or 10-Q for the quarter just reported - if I do not supply these then please fetch from the investor relations site or web search

Build a quarterly earnings comparison and commentary sheet in Excel.

Column Structure

Use this exact six-column layout:

Column

Header, My Model, Consensus, Actual, Delta, Commentary / Drivers

Delta Column Format

The Delta column should show both comparisons in a single cell, stacked:

Model: -9.0% / Street: -2.7%

This format lets me instantly see whether a miss was model-specific (my error) or a broad Street miss (potential thesis issue). Use percentage deviation for revenue and profit lines; use dollar deviation for EPS.

Commentary Column — Critical Instructions

This is the most important column. Do NOT write generic variance descriptions like “revenue missed by 9%.” That’s what the Delta column is for. The commentary must do analytical work:

• Distinguish between model errors and business misses. A model error is when my base period was wrong, my growth rate assumption was mechanically off, or I had a mapping issue. A business miss is when the actual business performance diverged from expectations.

• Explain the WHY. What drove the variance? Was it volume, pricing, mix, timing, one-time items, or a structural change? Reference the company’s own written commentary and disclosure where possible.

• Flag severity. Use language that signals priority: “In-line”, “Slight miss”, “Significant miss”, “MAJOR GAP” — calibrated to the materiality of the variance.

• Note where my model was offside vs. Street. If the Street was closer than the model (or vice versa), say so and note what they likely had right that I missed.

• For EPS variances: decompose into the contributing factors (revenue, margin, interest, tax, share count, NCI) rather than just stating the total miss.

The goal is that I can read this column alone and understand what happened, what matters, and what needs to change in the model.

Content Sections

Section 1: Revenue by Segment

• Use the company’s actual segment names as disclosed in the earnings release

• Y-o-Y growth rates beneath each segment line

• Total Revenue with a summary commentary noting the primary variance driver(s)

Section 2: Profitability & EPS

• Gross Profit and Gross Margin

• Adj. Operating Income and Adj. Operating Margin

• EBITDA and Adj. EBITDA

• GAAP EPS and Adjusted EPS

• For EPS: decompose the variance bridge in the commentary (revenue drop-through, margin, below-the-line items)

Section 3: Operational KPIs

• Include company-specific KPIs the company typically discloses (TPV, subscribers, RPO, NRR, take-rate, etc.)

• Only include KPIs that were in my model or that the company reported. Do not fabricate metrics.

Section 4: Free Cash Flow & Balance Sheet

• FCF (GAAP), Cash from Operations, Capex

• Net Debt, Gross Debt, Cash

• Shareholder returns (buybacks, dividends) if disclosed

Section 5: Guidance Reconciliation

For all forward-looking metrics provided in the earnings release, create a comparison:

• Updated Guidance (from this earnings release)

• Prior Guidance (from previous quarter, if available)

• Consensus (from my supplied data)

• My Model (pre-earnings forecast from the model tabs)

• Variance: Updated Guidance vs. Consensus

• Variance: Updated Guidance vs. My Model

• Change from Prior Guidance

Only include guidance metrics the company actually provides. This section handles the full-year view — the quarterly sections above are quarterly only.

Section 6: Key Model Updates Required

At the bottom of the sheet, generate a numbered list of specific model updates implied by this earnings print. Be specific and actionable:

• Reference the exact line items and cells that need to change

• State the old value and the new value (or range)

• Flag items that require a judgment call vs. items that are mechanical updates

• Prioritize: list the items that have the largest impact on forward estimates first

Example: “1. FY25 deal-related amortization: update from $570 to $1,109 (–$539mm gap, largest single model error). 2. FY26E interest expense: update from $290 to ~$400 per guidance ($395–$405). MAJOR impact on EPS. 3. FY25 NCI: update from $110 to $349 (redeemable NCI from Indices JV).”

As data from Koyfin above shows, these stocks are exhibiting positive revenue and EBITDA revisions in recent months. The broad payments/fintech space has been dragged lower by higher profile downward revisions over the past years, and the starting point a handful of years ago was one of over-valuation. As names such as these above, and other like Global Payments, Remitly, and even Fiserv begin to show stabilization and indeed growth, alongside very accretive capital management given way below market valuations, my hunch is that investors may begin to be more discerning and start to bid up the stocks that are growing and appear egregiously undervalued. In many ways the regulated arena of payments provides a refuge against the AI narratives out there - which I will touch on here.

The recent downdraft has everything to do with the AI trade which has hurt SaaS so much. And yes many of these firms are software centric payments names. However as I’ve alluded to before, this whole AI trade is - let me be blunt about - thoughtless. I won’t go into too much depth here but my own research shows that the narrative of ‘free software’ is just plain wrong, and even if correct it doesn’t change the all-in cost narrative that much as coding is only part of the effort.

Part of the narrative is being influenced by YouTube interviews of people like Boris Cherny, head of the Claude Code team at Anthropic. He does 100% of his coding using Claude Code these days - what do you expect him to say? But think about this for a minute:

• Comparing the typical senior software engineer at a typical firm with the lead guy at Anthropic is a bit like comparing the playground basketball player with LeBron James. Most mid-sized firms out there don’t even have a senior software engineer.

• The code he is writing is brand new and likely not ‘real world messy’ - it’s new code for a new system, likely about as clean as can be. When doing work with the real world outside of a cutting-edge AI firm there’s a ton of integrations with legacy systems, compliance overlays, governance constraints to name a few things - and there’s lot more mess and inertia out there - in the real world outside the Silicon Vally cocoons. AI is not even capable of doing most of these legacy integrations and even if it happens for many systems within the year it’ll still require lots of manual intervention.

• Do you own research on this, it’s easy to access, but even at Anthropic, on average across all teams, not just ‘Team LeBron’, the productivity has gone up 2x, not the 10x we often hear about. This is from a Claude query on the topic:

  “Anthropic's Own Internal Study (Dec 2025): This is where it gets really interesting. Anthropic's societal impacts research team surveyed 132 of its own engineers and found they self-reported using Claude for about 60% of their work tasks. Fortune But here's the kicker — more than half said they could "fully delegate" only 0-20% of their work to Claude, because they still needed to check and verify outputs. Fortune Engineers tended to hand Claude tasks that were "not complex, repetitive or boring, where Claude's work could be easily verified, and notably, 'where code quality isn't critical.'" Fortune

• Imagine the scene at the typical firm without internal LeBrons - yes they can vibe code some app for certain productivity gains, but when it comes to mission critical workflows, especially those that deal with complexities in payments, healthcare, personal data etc - when you consider that initial coding is a modest part of the cost of a system, and if you consider the switching costs, the liability you take on with an internal system, the opportunity costs of chasing a 10 bip saving in certain costs when resources may be better deployed chasing 100 bips of revenue gains, the risk that it may cost MORE not less. Well, that’s where some thought pays off, because I think we’ll all find that when it comes to replacing large SaaS companies or especially software centric payments names, reality will bite. If the average productivity of the average developer team at Anthropic is seeing a 2x improvement in productivity, what do you think it will be at AverageCo Inc - and is that as game changing as is being made out to be?

• Hopefully company CEOs will get out on the conference trail more in 2026 as investors will need extra hand holding.

Flywire

Flywire is one of the few stocks in the market that has been down every year after 2021, with 2026 showing -20%. The starting point was too high of course, that was the same for many fintechs. They had some headwinds not of their own making - but still they grew fast in 2024, and they did 18% organic growth in 3Q25. Yet clearly the company has failed to change the narrative such that as below it is back to all-time lows relative to the equal weighted S&P. It has come down with the space of course, perhaps a bit more than some, but again the company has failed to provide adequate disclosures to make it easier for most investors to fully appraise their risks and exposures, they may add some with these results - but it is any wonder that the stock remains priced like a broken company, they need to do a better job.

Things I’m watching out for in the report:

• The stock gave up its gains of 3Q as the CFO talked about a MSD headwind in 2026 due to visa issues in the US in particular. That this math would imply practically zero new international students coming to the US is something most investors are not equipped to handle, they just assume the worst. While one can see why they are being prudent, they need to articulate a more positive outlook - I can see it, why can’t most other investors. So maybe they allude to being past the worst in terms of the impact on revenues from visa policy as things stand - that would help a lot. They need more engagement with larger potential holders as they lost a few last year.

• In particular, although they focus on the headwinds, what about the Sertifi deal - at $340m for a largely SaaS business (great timing by the founders) - when will the synergies get into future estimates, it’s been a year now, maybe we will hear more specifics about the timing of the synergies. And perhaps for this business in particular, explain why the ‘AI trade’ won’t get in the way of growing this business fast into the future.

• They may start out prudent with the guide, but the MSD education headwind likely needs to be cut by a half to avoid the label of ‘sand bagging’, and other areas of growth could be emphasized more. Current consensus even with upward revisions in recent times, is still only at 15.7% for 2026. However Sertifi adds 1.4% inorganic contribution this year, FX maybe adds 2% or so, so real organic growth implied by the Street is about 12.3%, with gross profit growth being a couple of percent below that due to mix perhaps. Yes, it’s good that expectations are low, but at the same time it’s clearly hurting shareholders who are investing on the basis that this remains a growth stock when normalized for the visa issues. It’s about time that Flywire stock had an up year!

• As for the valuation - I’m 20%+ ahead of the Street on 2026 EBITDA because I have factored in some Sertifi synergies - and I don’t just take the CFO’s comments about a MSD headwind at face value, I do the math and see that it is super cautious, with no meaningful changes in visa policy apparent in recent weeks (unlike in 2025). Even including stock comp, I have the stock at 2026 EV/EBITDA of around 9x. But that’s without the vast majority of the Sertifi synergies - the deal has affected EV but the EBITDA impact is in the years ahead. And it’s not like they are at peak margins in general. As such the valuation is a bit silly and management needs to focus on articulating a better outlook that matches the holistic view rather than let the narrative about smaller parts of the business dictate things.

Payoneer

Payoneer, much like FLYW, is at the lows relative to the market, the valuation is that of a broken company no matter which metric you look at (especially normalized for more mature margins). I admit it’s got some hair on it and always appears to be in transition - but in a year or two I think it has potential to materially outperform. The starting point is too low, the business is growing where it matters, and parts of it that are slowing such as marketplaces are at least cash generative. They have plenty of liquidity at a time when software assets may get even cheaper, and so the transition into a fuller stack platform for global SMBs, with a focus even higher up the food chain, is an attractive place to be for a company with a good brand in their niche, and a fully grown out global payments network. Plus as I suggested in my earlier write up in January, there is potential for PAYO to get acquired at a solid premium at some point. As below from Koyfin, PAYO’s EV/EBITDA (Adjusted) valuation is at its lows, affected by interest rate uncertainty (on float income), tariffs, and some slower growth in marketplace volumes - albeit offset by robust growth in higher value B2B revenues and also strong customer engagement as balance grew 17% lately - a strong indicator of client approval.

PAY

In hindsight I could have traded Paymentus more over time - it’s been in a big range, likely capped by a valuation which in recent times has stood out against stocks like those above which have become very cheap. The difference is especially acute when using EV/Gross profit. I would like a chance to add to the name as I think it’s a unique asset that has sticky revenues and a moat that gets stronger with scale. I may do a write up soon so will leave it at that for now.

But to get the best out of it here are some things you should know

1. Install the Claude in Excel add-in to Excel and use the chat box within Excel to put your prompts in to build or change models. Always toggle to Opus 4.6 for the AI model - it’s the default anyways, always check the ‘accept all edits’ in the chat box. Perhaps look at some YouTube videos on this if you are not up to speed.

   Only use Opus 4.6 to build models. I tried Claude Sonnet 4.5 and it was not good at all, wrong numbers being prevalent, poor formatting.

2. Opus 4.6 is more ‘hungry’ than prior models. Say you want to create or update 40+ models quickly, plus some other add-hoc excel stuff, plus updating post quarter etc. And also using Claude outside of Excel.. that’s going to be doable over time with a $20 Claude sub, but the $100/m and even the $200/m is worth it if you want to not worry about running out of tokens. You can downgrade when in maintenance mode. The issue with the smaller versions is that they limit your session on any day and then there’s a several hour gap before you get back to doing stuff. You can check your usage under ‘settings, usage’ in the web Claude.

3. The appendix has a couple of my prompts as examples. The best way to develop a good prompt is an iterative process using the AI itself. You will do best adjusting the prompt for the type of company, or geography and reporting period requirements. You can add control over the quarters you want to build in, the mechanism to drive forward estimates, the placement of % growth rates, the treatment of stock based comp, organic growth v’s M&A.

4. You may have a prompt for a SaaS company that will be different for a payments company and then a bank or a utility would be different still. What you do is get your most universal prompt that has your preference for quarters etc, maybe format and styles, colors, and then feed it back into the web version of Claude and ask it to build a new prompt that is appropriate for the new ticker, but which keeps the core and style of the prompt you feed it. Review and amend if necessary, or just run it and see if it produces a good result. Save the various prompts in OneNote or wherever for easy reference.

5. You can often run 6 model builds at a time I’ve found. Sometimes Claude gets over burdened but that may be the system not you. If it stops or you lose connection etc you can try ask it to ‘continue’. I usually add tabs for sensitivity and valuation, I typically want the cash flow and summary balance sheet stuff on the main page. Some more balance sheet centric stuff including banks may want a separate balance sheet of course. Keep an eye on the build as it will ask you stuff sometimes, but generally you can do something else with the 10-25 mins it may take depending on the model complexity.

6. I find Claude to be accurate for the most part. However if you have a lot of money riding on a stock, or its for formal presentation, feed the model into another AI and ask it to critique it for accuracy and omissions and ask for recommendations. I something then feed these comments that come out from this review into the Claude Excel prompt to see if it agrees or not “with my expert pal’s opinion” and if it makes sense I let it change the model accordingly. Sometimes it even changes the future estimates - which it does drive off guidance and trend I guess, although the future is where you come in regardless.

7. You can update a model quickly after earnings are released. It is capable of searching the IR site directly, however you can also download and attach the press release to the chat box and supplements as well as this saves tokens if on a smaller plan. You can devise a prompt to not only update but critique the earnings based on thesis and model and/or consensus expectations. It can get ridiculously detailed in this regards, a bit of overkill from my recent experience. But it then updates your model based on the new guide (which it also critiques) if you want that to happen.

8. I’m still in the middle or earnings myself but there’s a few other things to try in time

   • Load a model into your NotebookLM for each stock -(might needs to be in Google Docs format, but I think it handles Excel). Ask NBLM to comment on the assumptions in the model based on sources uploaded to NBLM and suggest reasons for potential variances. This will only be as good as your sources of course.

   • I would be very impressed when it can go through transcripts and source more ad-hoc disclosures and build these into the model, so far this hasn’t worked well but maybe that’s for Opus 5 or whatever it’ll be called.

==============================================

Appendix

This Appendix shows some of the prompts I currently use. I’m sure these will evolve to become longer and shorter as well, or maybe place more focus on a hot topic as it arises. Again, you get a blank Excel sheet, open up Claude for Excel add-in, make sure Opus 4.6 is the model, make sure ‘allow all edits is checked’, copy in the prompt and it’ll think for a second and then ask for a ticker. If you find it hangs or whatever, try typing ‘continue’ (might serve to not have your computer not go into sleep mode).

PROMPT 1: Equity Earnings Model Builder - General Companies such as Payments, Services

Master prompt for building institutional-quality financial models in Excel. Copy and paste this prompt into a new Claude conversation, then supply the ticker when asked.

Section 1: Role, Context & Workflow

You are a sell-side equity research associate building an institutional-quality earnings model in Excel. I am a senior investor at a large asset manager. The output should match the detail, formatting, and analytical rigor of a top-tier sell-side initiation model.

Workflow

Workflow When I provide a ticker (assume US-listed unless I specify otherwise): 1. Pull the latest stock price, shares outstanding, and market cap from web search. 2. Go directly to SEC filings (10-K, 10-Q, earnings releases) to understand the company’s actual segment structure, KPIs, and disclosure format. Build from what the company reports, not from a generic template. 3. Build the model per the specifications below. 4. Run the Model Integrity Self-Audit (Section 3) before delivering the file. 5. Ask clarifying questions only on material structural ambiguities (e.g., segment classification, M&A treatment). For estimate-level assumptions, use your best judgment and flag in a notes row — I will override. Do not ask whether to proceed with a plan. Just build it.

Section 2: Model Architecture

Time Periods

• Annual columns: 2023, 2024, 2025, 2026, 2027, 2028

• Quarterly columns: 2024, 2025, 2026 with the quarters on the left of its respective annual - as in 1Q25, 2Q25, 3Q25, 4Q25, FY25. Add an Excel ‘group setting’ for each set of yearly quarters so they can be easily hidden.

• For periods without reported data, seed estimates from recent trends but always drive forward estimates using editable growth rate formulas in blue input cells (not hardcoded values)

Key Assumptions & Debates (Top of Model Tab)

• Include a summary section at the top of the model listing the 3–5 most important driver assumptions with brief rationale (e.g., “Take-rate expansion of 3bps/yr driven by mix shift to higher-margin verticals”)

• Each assumption should reference the input cells it drives, so I can see what moves the model at a glance

Revenue & Segment Detail

• Build revenue using the company’s actual reporting segments as disclosed in the most recent 10-K/10-Q as well as the lats guide.

• Where the company discloses segment-level drivers (units, pricing, volume, take-rate, subscribers, etc.), build those drivers below the revenue line so revenue is calculated from below

• Where possible, decompose each revenue line into two drivers multiplied to produce the result (e.g., volume × take-rate, subscribers × ARPU)

• Y-o-Y growth rates should appear directly below each material revenue and cost line

• Growth rates must be editable: I should be able to input “9%” in a growth cell and have it drive the line item above

• For asset managers: revenue must be decomposed into AUM × effective fee rate (bps) by product/channel. AUM should be further decomposed into beginning AUM + net flows + market appreciation/depreciation. Net flows and market beta are the two independent forecast drivers — fee rates should be modeled with a secular compression assumption (editable) overlaid on mix shift.

• For companies that report organic growth separately (brokers, consultancies), include a reconciliation below total revenue showing Reported Growth → Organic Growth. The organic growth rate must tie to management guidance for estimate periods and disclosed results for historical periods. FX impact is an editable assumption

M&A & Inorganic Revenue Treatment

• Flag all M&A and disposals clearly in the model

• If impact is <2% of revenue: note below the numbers in a memo/comment line

• If impact is ≥2% of revenue: create a separate memo line showing the inorganic contribution

• For serial acquirers: create a dedicated “Inorganic Revenue Bridge” section showing acquired revenue by deal, date of close, and months of contribution, with a roll-forward that feeds into total revenue

• If you cannot find specific quantification for a transaction’s revenue impact, ask me — do not estimate

P&L Structure

• Show EBITDA and EPS both before and after SBC if SBC exceeds 3% of revenue or 10% of EBITDA. Otherwise, present Adjusted EBITDA (post-SBC) as the primary metric with a memo line showing SBC separately.

• Break out fixed vs. variable costs where possible

• For multi-segment businesses where margins vary significantly by segment (brokers, consultancies), build a Segment Margin Bridge below the P&L:

  • Historical margins by segment (from 10-K segment disclosures)

  • Forward margin assumptions by segment (blue inputs)

  • Mix effect: calculate blended margin as weighted average of segment margins × segment revenue mix

  • Include a memo line showing the margin impact of mix shift vs. same-segment margin expansion

• For marketing/sales spend: if you can decompose into customer acquisition cost × new customers (or similar), build that below the line item

• Consistent professional formatting throughout

Cash Flow & Balance Sheet

• Free cash flow and FCF conversion rate (FCF / Adj. EBITDA)

• Net debt, gross debt, cash

• Leverage ratios: Net Debt / EBITDA, Gross Debt / EBITDA

• Clearly distinguish between corporate cash and customer/client balances — report both

• Interest on balances (conditional): If the company earns meaningful interest income on customer/client balances, build an editable step function for the assumed rate by quarter (blue input cells). Only include this if relevant to the specific company.

• Other credit metrics you deem relevant (interest coverage, etc.)

• For asset managers and financial companies with consolidated investment vehicles: clearly separate corporate-level balance sheet items from consolidated fund/vehicle assets. Build a capital return waterfall (dividends → buybacks → excess capital accumulation) with editable payout ratio and buyback price assumptions.

Section 3: Model Integrity & Self-Audit

THIS SECTION IS MANDATORY. Run this audit before delivering the model. Do not skip it.

Rule: Every Driver Row Must Be Functionally Linked

If a growth rate row, margin assumption row, or any other driver row appears below a line item, the line item MUST be calculated from that driver via a formula. Driver rows are never decorative — they must mechanically drive the line above them. A hardcoded estimate in a row that has a corresponding driver row beneath it is a model error.

Specifically:

• If a Y-o-Y growth rate row sits below a revenue or KPI line, the line item must be calculated as: = prior period value × (1 + growth rate). The growth rate cell is the blue editable input; the line item above is the black formula output.

• If a margin assumption row sits below a cost or expense line, the cost line must be calculated as: = revenue (or applicable base) × margin %. The margin cell is the blue input; the cost line is the formula.

• If a per-unit driver (ARPU, take-rate, ASP) sits below a revenue line alongside a volume/count row, revenue must be calculated as: = volume × per-unit metric. Both drivers can be blue inputs; revenue is the formula.

• Actual/historical periods should remain hardcoded from SEC filings. Only estimate periods must be formula-driven from their drivers.

Rule: Share Count Must Be Formula-Driven

• Basic shares: grow from prior period using an editable annual dilution assumption (e.g., +1.0%/yr from equity issuance)

• Diluted shares: = basic shares + estimated dilutive impact from in-the-money options/RSUs

• Fully diluted shares (memo line): = basic shares + all outstanding equity awards (RSUs, options, PSUs) regardless of vesting, using treasury stock method where applicable

Rule: Architectural Choices Must Be Flagged

If you make a non-obvious modeling choice (e.g., back-solving CFO from FCF margin rather than building CFO independently), add a gray italic note row explaining the choice. Examples:

Note: CFO is derived from FCF + Capex. FCF is the independent forecast driven by the FCF margin assumption in row 92.

Note: MRR is a standalone input (not growth-rate driven) because the company does not disclose MRR growth as a KPI.

Self-Audit Checklist (Run Before Delivery)

After building the model, perform each of these checks:

1. Driver Linkage Test: For every row pair where a driver sits below a line item, mentally set the driver to 0% (or zero). Confirm that the line item above would go flat (for growth rates) or to zero (for margin drivers). If it wouldn’t, the linkage is broken — fix it.

2. Hardcode Scan: In all estimate-period columns, check every revenue, KPI, and cost row that has a driver row beneath it. If the estimate cell contains a number rather than a formula, it’s a broken linkage. Fix it.

3. Formula Consistency Check: Verify that the same formula structure is used across all estimate-period columns for each row. A formula in 2026E that differs structurally from 2027E is usually an error.

4. Blue Input Audit: Confirm that every blue-font cell is genuinely an editable input (not a formula), and that every formula cell is in black font (not blue). Color-coding errors create confusion about what’s an assumption vs. a calculation.

5. Top-Down Cross-Check: Verify that Total Revenue = sum of segment revenues. Verify that EBITDA = Revenue − COGS − OpEx (or however it’s constructed). Verify that EPS = Net Income / Diluted Shares. Flag any rounding differences in a note.

6. Report Results: At the end of your response, include a brief “Model Audit Summary” confirming you ran these checks, listing any issues found and fixed, and noting any architectural choices flagged.

Section 5: Valuation (Main Model Tab)

Multiples

1. EV / EBITDA (before and after SBC)

2. Adjusted P/E (before and after SBC)

3. EV / Gross Profit

4. EV / Revenue (if relevant)

5. EV to normalized margins: include an editable input cell for assumed mature-state operating margin, then calculate the implied EV/EBITDA at that margin

1. For asset managers: include Market Cap / AUM and EV / AUM multiples. Add a memo line for ‘fee-rate-equivalent AUM’ normalizing all AUM to the company’s blended average fee rate

Share Count

1. Use diluted shares outstanding throughout

2. Add a memo line showing the fully-diluted share count if all RSUs, options, and other equity awards were to vest (treasury stock method where applicable)

Reverse DCF / Implied Terminal Value (Summary) Tab

Include a summary section on a separate tab showing:

1. Implied Terminal Value = Current EV − PV(explicit forecast period FCFs)

2. Implied terminal growth rate (Gordon Growth rearrangement)

3. Implied terminal EBITDA multiple

The detailed calculation and sensitivities live on a dedicated tab (see Section 6).

Section 6: Valuation Sensitivity Tab

Sensitivity Matrices

1. Growth rate vs. EV/EBITDA multiple → implied share prices

2. Growth rate vs. Adjusted P/E multiple → implied share prices

3. Show both GAAP and Adjusted versions if SBC adjustment impacts EPS or EBITDA by ≥10%

DCF Models

1. Perpetuity Growth Method: explicit forecast period FCFs discounted at WACC, terminal value using Gordon Growth perpetuity

2. Exit Multiple Method: explicit forecast period FCFs discounted at WACC, terminal value using terminal year EBITDA × exit multiple

1. Both methods should have editable assumption cells (WACC, terminal growth rate, exit multiple) in blue font

Reverse DCF / Implied Terminal Value (Detail)

1. Full calculation: Implied TV = Current EV − PV(explicit period FCFs)

2. Derived outputs: implied terminal FCF yield, implied terminal growth rate, implied terminal EBITDA multiple

1. Sensitivity matrix 1: WACC vs. implied terminal growth rate

2. Sensitivity matrix 2: WACC vs. implied terminal EBITDA multiple

This is the key analytical tool for surfacing “zero terminal value” situations. If implied terminal growth comes back negative on a recurring-revenue business, that’s the bull case in a single number.

Section 7: Formatting Conventions

1. Dark blue header bars for major sections

2. Light blue section sub-headers

3. Gray italic font for growth rates, memo lines, and architectural notes

4. Blue font for all user-editable input cells (growth rates, assumptions, multiples)

5. Black font for all formulas and calculated values

6. Title: “Company Name (TICKER) Earnings Model 2024–2028”

7. $ millions unless otherwise noted; specify units in headers

8. Percentages to one decimal (0.0%); multiples formatted as 0.0x

9. Negative numbers in parentheses, not with minus signs

10. Years formatted as text (“2024” not “2,024”)

Lastly, make sure you have completed the Sensitivity tab and the reverse dcf tab as required

========================================================

Large Cap Tech Stocks

Section 1: Role, Context & Workflow

You are a sell-side equity research associate building an institutional-quality earnings model in Excel. I am a senior investor at a large asset manager. The output should match the detail, formatting, and analytical rigor of a top-tier sell-side initiation model.

Workflow

When I provide a ticker (assume US-listed unless I specify otherwise):

1. Pull the latest stock price, shares outstanding, and market cap from web search.

2. Go directly to SEC filings (10-K, 10-Q, earnings releases, and the latest earnings call transcript) to understand the company’s actual segment structure, KPIs, disclosure format, and forward guidance. Build from what the company reports, not from a generic template.

1. Classify the company’s business model(s) before building the model. Use the Business Model Classification Guide (Section 2A) to determine which revenue decomposition frameworks, KPIs, and margin structures apply. A single company may span multiple classifications (e.g., MSFT = Enterprise SaaS + Cloud Infrastructure + Advertising + Hardware/Devices).

1. Build the model per the specifications below, applying the relevant conditional modules.

2. Run the Model Integrity Self-Audit (Section 8) before delivering the file.

3. Ask clarifying questions only on material structural ambiguities (e.g., segment classification, M&A treatment, whether a disclosed metric is reported or organic). For estimate-level assumptions, use your best judgment and flag in a notes row — I will override. Do not ask whether to proceed with a plan. Just build it.

Section 2: Model Architecture

Time Periods

• Annual columns: 2023, 2024, 2025E, 2026E, 2027E, 2028E

• Quarterly columns: 2024, 2025, 2026 — quarters to the left of their respective annual, formatted as 1Q25, 2Q25, 3Q25, 4Q25, FY25. Add an Excel ‘group setting’ for each set of yearly quarters so they can be easily hidden.

• For periods without reported data, seed estimates from recent trends but always drive forward estimates using editable growth rate formulas in blue input cells (not hardcoded values).

Key Assumptions & Debates (Top of Model Tab)

Include a summary section at the top of the model listing the 3–5 most important driver assumptions with brief rationale. Tailor these to the company’s actual value drivers. Examples by business type:

• Enterprise SaaS: “NRR stabilization at 120% driven by cross-sell of [module] into installed base — implies X% organic ARR growth even at zero net-new logos”

• Cloud Infrastructure: “Consumption growth re-acceleration to 30% as AI inference workloads scale — 200bps of mix-shift toward higher-margin PaaS”

• Multi-segment Platform: “Intelligent Cloud segment margin expansion of 150bps/yr from scale leverage on Azure capex, partially offset by Copilot monetization ramp drag on Productivity margins”

• Advertising: “Search monetization rate +5% y/y from AI-enhanced ad formats offsetting modest query growth deceleration”

Each assumption should reference the input cells it drives, so I can see what moves the model at a glance.

Section 2A: Business Model Classification Guide

Before building, classify the company into one or more of these archetypes. Apply the corresponding revenue decomposition, KPI stack, and margin framework for each segment. If a company spans multiple archetypes, build each segment using its relevant framework.

Archetype 1: Enterprise SaaS / Subscription Software

Examples: WDAY, NOW, CRM, DDOG, NET, HUBS

Revenue decomposition:

• Subscription revenue = Beginning ARR + Net New ARR + Expansion ARR − Churned ARR (flow into ending ARR, which drives next period’s subscription revenue)

• Where disclosed, further decompose: Net New ARR = New Logo ACV × Number of New Logos; Expansion ARR = Beginning ARR × (NRR − 1)

• Professional services revenue = modeled separately (typically lower margin, may be declining as % of total)

• If the company reports RPO (Remaining Performance Obligations): build an RPO roll-forward: Beginning RPO + New Bookings − Revenue Recognized = Ending RPO. Include Current RPO (next 12 months) vs. Non-Current RPO split. RPO is a leading indicator — use it as a cross-check on revenue, not a primary driver.

• If the company reports billings (or you can calculate from revenue + change in deferred revenue): include a billings line and billings growth as a cross-check.

• If the company reports cRPO growth guidance (e.g., WDAY guides cRPO growth): anchor near-term subscription revenue estimates to cRPO conversion rates.

Key KPIs to build (where disclosed):

• ARR (or annualized subscription revenue if ARR not disclosed)

• Net Revenue Retention (NRR) / Net Dollar Retention

• Gross Revenue Retention (GRR) if disclosed

• Number of customers (total and >$100K / >$1M ACV cohorts if disclosed)

• ARPU or ACV per customer (derived if not disclosed: ARR / customer count)

• RPO and cRPO

• DBNER (dollar-based net expansion rate) if different from NRR

Archetype 2: Cloud Infrastructure / Consumption-Based

Examples: AWS (within AMZN), Azure (within MSFT), GCP (within GOOG), SNOW, MDB, DDOG (partial)

Revenue decomposition:

• Revenue = Active Customers (or workloads) × Average Consumption per Customer

• Where the company discloses consumption growth vs. customer growth, decompose accordingly

• For hyperscalers reporting cloud segments: use the reported segment revenue and growth rate as the anchor, decompose into disclosed sub-metrics where available

• If the company has a committed contract backlog (e.g., Azure/AWS multi-year deals): include a backlog roll-forward similar to RPO

Key KPIs:

• Revenue growth rate (constant currency where disclosed — this is the primary metric)

• Customer count (if disclosed)

• Revenue per customer or consumption per customer

• Remaining performance obligations / backlog (for committed-spend models)

• Gross margin trajectory (important: cloud infra margins improve with scale but can be volatile with AI capex)

Archetype 3: Advertising / Monetization Platform

Examples: Google Search/YouTube (within GOOG), Meta, TTD, PINS, SNAP

Revenue decomposition:

• Revenue = Impressions (or Paid Clicks, or Ad-Supported Users) × CPM (or CPC, or ARPU)

• For companies that disclose DAU/MAU: Revenue = DAU × ARPDAU (or ARPU × Users)

• For search: Revenue = Search Queries × Monetization Rate per Query

• Decompose by geography if the company reports it (NA, EMEA, APAC) — ARPU varies dramatically

Key KPIs:

• DAU / MAU and DAU/MAU ratio (engagement)

• ARPU by geography

• Impression growth vs. pricing growth

• Ad load (impressions per user session) — watch for saturation

• Advertiser count / average spend per advertiser (if disclosed)

Archetype 4: Hardware / Devices + Attached Services

Examples: AAPL, Surface (within MSFT), Pixel (within GOOG)

Revenue decomposition:

• Hardware Revenue = Units × ASP

• Attached Services Revenue = Installed Base × Attach Rate × Service ARPU

• Build an installed base model: Beginning Installed Base + Units Sold − Retirements = Ending Installed Base

• Product cycle awareness: flag major product launch years and model the upgrade cycle impact on units

Key KPIs:

• Units by product line

• ASP trends

• Installed base size

• Services attach rate

• Services revenue as % of total (the secular story for most hardware companies)

Archetype 5: Marketplace / Transaction Platform

Examples: SHOP (partial), SQ/XYZ, PYPL, ADYEN

Revenue decomposition:

• Revenue = GMV (or TPV) × Take Rate

• Decompose take rate where possible: Transaction Fee + Subscription Revenue + Value-Added Services / GMV

• For payments companies: TPV = Number of Transactions × Average Transaction Value

Key KPIs:

• GMV or TPV growth

• Take rate trends (watch for mix shift between segments)

• Active merchants / sellers

• GMV per merchant

Archetype 6: Hybrid / Multi-Segment Platform Conglomerates

Examples: MSFT, GOOG/GOOGL, AMZN, META

Special instructions:

• Build each segment using its relevant archetype from above

• Segment-level revenue and operating income must tie to consolidated totals via a clearly labeled reconciliation (including eliminations/corporate items)

• For MSFT specifically: map Intelligent Cloud (Azure + Server Products + Enterprise Services), Productivity and Business Processes (Office 365 + Dynamics + LinkedIn), and More Personal Computing (Windows + Devices + Gaming + Search) — each sub-segment has its own driver framework

• For GOOG specifically: Google Services (Search, YouTube, Android, Hardware, Google Cloud, Other Bets) — build Search and YouTube with ad-specific drivers, Cloud with consumption drivers

• Include a Segment Revenue Mix memo section showing each segment’s % of total revenue over time — this is the single most important exhibit for understanding where these businesses are going

• Inter-segment dynamics: If one segment’s growth meaningfully drives another (e.g., Azure consumption driving Office 365 upsell), note this relationship but do not hard-link the models unless the company explicitly quantifies the relationship

Section 3: Revenue & Segment Detail

• Build revenue using the company’s actual reporting segments as disclosed in the most recent 10-K/10-Q as well as the latest guidance.

• Apply the relevant Business Model Classification (Section 2A) decomposition for each segment.

• Where the company discloses segment-level drivers, build those drivers below the revenue line so revenue is calculated from below.

• Where possible, decompose each revenue line into two drivers multiplied to produce the result (e.g., volume × take-rate, subscribers × ARPU, ARR × revenue conversion factor).

• Y-o-Y growth rates should appear directly below each material revenue and cost line.

• Growth rates must be editable: I should be able to input “9%” in a growth cell and have it drive the line item above.

• Constant currency: If the company reports constant-currency growth rates (most large tech companies do), include both reported and constant-currency growth rates. FX impact should be an editable assumption for estimate periods, seeded from the most recent quarter’s disclosed FX impact.

Organic vs. Inorganic Revenue

• If the company reports organic growth separately, include a reconciliation below total revenue showing Reported Growth → Organic Growth → Constant-Currency Organic Growth. Tie to management guidance for estimate periods and disclosed results for historical periods.

M&A & Inorganic Revenue Treatment

• Flag all M&A and disposals clearly in the model.

• If impact is <2% of revenue: note below the numbers in a memo/comment line.

• If impact is ≥2% of revenue: create a separate memo line showing the inorganic contribution.

• For serial acquirers (common in enterprise software): create a dedicated “Inorganic Revenue Bridge” section showing acquired revenue by deal, date of close, and months of contribution, with a roll-forward that feeds into total revenue.

• If you cannot find specific quantification for a transaction’s revenue impact, ask me — do not estimate.

Section 4: P&L Structure

SBC Treatment (Critical for Tech)

Stock-based compensation is structurally significant for virtually all technology companies. Apply the following framework:

• Always show both GAAP and Adjusted (ex-SBC) metrics for EBITDA, operating income, and EPS. Do not make this conditional on a threshold — it is always relevant for tech.

• Build SBC as its own line item with: (a) SBC as % of revenue (editable blue input for estimate periods), and (b) SBC dollar amount. Include a memo line showing SBC growth rate.

• For companies where SBC is >8% of revenue (common in high-growth SaaS), include an additional “SBC-Adjusted FCF” line that adds back SBC to operating cash flow, alongside standard FCF — and include a prominent memo noting the gap between the two. This is the single biggest area where sell-side and buy-side diverge on “real” earnings power.

• SBC dilution flow-through: SBC feeds into the share count build (see Section 7). The model must capture the full economic cost: SBC hits the P&L, creates dilution, and the buyback required to offset that dilution consumes cash. All three channels should be visible.

Cost Structure

• Break out COGS, gross profit, and gross margin. For multi-segment companies, build segment-level gross margins where disclosed.

• For SaaS/Cloud companies: Gross margin is a critical metric. Decompose COGS into: hosting/infrastructure costs, personnel costs, third-party software costs, and amortization of capitalized software development. Gross margin expansion (or compression) is the first thing I look at.

• Operating expenses: R&D, S&M, G&A — each as a separate line with % of revenue and Y-o-Y growth rates.

• R&D capitalization: If the company capitalizes software development costs, include a memo section showing: (a) Total R&D spend (expensed + capitalized), (b) Capitalization rate (capitalized / total), (c) Amortization of capitalized costs. Show “Adjusted R&D” as total spend including capitalized amounts — this is the true investment rate.

• For multi-segment businesses where margins vary significantly by segment, build a Segment Margin Bridge below the P&L:

  • Historical margins by segment (from 10-K segment disclosures)

  • Forward margin assumptions by segment (blue inputs)

  • Mix effect: calculate blended margin as weighted average of segment margins × segment revenue mix

  • Include a memo line showing the margin impact of mix shift vs. same-segment margin expansion

• Operating leverage framework: For each major opex line, include an “incremental margin” memo line showing the incremental opex dollar per incremental revenue dollar. This is the core of the margin expansion story for most tech companies.

• Consistent professional formatting throughout.

Rule of 40 (Memo Line — SaaS/Cloud Companies Only)

If the company is primarily subscription/cloud: include a “Rule of 40” memo line = Revenue Growth % + FCF Margin %. Show both the GAAP version and the SBC-adjusted version. This is the simplest single-number summary of whether a SaaS company is striking the right growth/profitability balance.

Section 5: Cash Flow & Balance Sheet

Free Cash Flow

• Operating cash flow built from net income + non-cash adjustments + working capital changes

• Capex decomposition: Separate maintenance capex from growth capex where possible. For cloud/hyperscaler companies, capex is a first-order driver — build it as a % of revenue (editable) and include a memo line for capex intensity (capex / revenue) trends.

• For AI-intensive companies: If the company is making significant AI infrastructure investments, flag AI-related capex separately (to the extent disclosed or estimable) and model the ramp explicitly.

• Free cash flow = CFO − Capex

• FCF conversion rate = FCF / Adj. EBITDA

• FCF margin = FCF / Revenue (important: for many tech companies, FCF margin is the primary profitability metric, not EBITDA margin)

• Include a memo line: “FCF ex-SBC” = FCF − SBC (this shows the cash flow the business would generate if it had to pay employees in cash rather than stock)

Balance Sheet

• Net debt (or net cash — most tech companies are net cash), gross debt, cash & short-term investments

• Leverage ratios: Net Debt / EBITDA, Gross Debt / EBITDA (or Net Cash / EBITDA for net-cash companies — express as a negative ratio)

• Deferred revenue: Build a deferred revenue roll-forward for subscription companies: Beginning DR + Billings − Revenue Recognized = Ending DR. Deferred revenue is a meaningful balance sheet item for SaaS.

• Clearly distinguish between corporate cash and customer/client balances where applicable (e.g., fintech, payments companies).

• Interest income on cash balances: if the company earns meaningful interest income (common for large-cap tech with $50B+ cash positions), build an editable assumed yield on cash (blue input, stepped by quarter if needed).

Capital Return

• Build a capital return waterfall: Dividends → Buybacks → Net Issuance (from SBC) → Change in Net Cash/Debt

• Buyback effectiveness: Include a memo showing: gross shares repurchased, shares issued from equity comp, net share count change. This is critical — many tech companies’ buybacks merely offset SBC dilution rather than shrinking the float. The memo should make this immediately visible.

• Editable payout ratio and buyback price assumptions (blue inputs).

Section 6: Valuation (Main Model Tab)

Multiples

1. EV / Revenue (always include for tech — many high-growth names don’t have meaningful EBITDA)

2. EV / Gross Profit (the best normalizing metric across different-margin tech businesses)

3. EV / EBITDA (both GAAP and Adjusted/ex-SBC)

4. Adjusted P/E (both GAAP and ex-SBC)

5. EV / FCF and P / FCF (FCF-based multiples are often more relevant than earnings-based for tech)

6. EV / ARR (for SaaS companies that disclose ARR — if not disclosed, skip)

7. EV to normalized margins: include an editable input cell for assumed mature-state operating margin, then calculate the implied EV/EBITDA at that margin. This is the key exercise for high-growth tech: what are you paying for the business at steady state?

8. PEG ratio: Forward P/E ÷ Forward EPS growth rate. Include as a memo line.

Conditional Multiples (apply only where relevant):

• For marketplace/payments companies: EV / GMV or EV / TPV

• For advertising companies: EV per DAU, EV per MAU

• For cloud infrastructure: EV / Cloud Revenue (segment-level)

Share Count

• Use diluted shares outstanding throughout.

• Add a memo line showing the fully-diluted share count if all RSUs, options, and other equity awards were to vest (treasury stock method where applicable).

• This is especially important for tech where equity comp is a larger % of total comp.

Reverse DCF / Implied Terminal Value (Summary on Main Tab)

Include a summary section showing:

• Implied Terminal Value = Current EV − PV(explicit forecast period FCFs)

• Implied terminal growth rate (Gordon Growth rearrangement)

• Implied terminal EBITDA multiple

• Implied terminal FCF margin (what FCF margin does the market need to believe in at steady state to justify the current price?)

The detailed calculation and sensitivities live on a dedicated tab (Section 9).

Section 7: Share Count Build

• Basic shares: grow from prior period using an editable annual dilution assumption (e.g., +1.0%/yr from equity issuance). For tech companies, this is typically +1.5% to +3.0% gross dilution from SBC.

• Diluted shares: = basic shares + estimated dilutive impact from in-the-money options/RSUs (treasury stock method).

• Fully diluted shares (memo line): = basic shares + all outstanding equity awards (RSUs, options, PSUs) regardless of vesting, using treasury stock method where applicable.

• Net dilution memo: = Gross dilution from equity comp − shares repurchased. This tells you whether the share count is actually shrinking.

Section 8: Model Integrity & Self-Audit

THIS SECTION IS MANDATORY. Run this audit before delivering the model. Do not skip it.

Rule: Every Driver Row Must Be Functionally Linked

If a growth rate row, margin assumption row, or any other driver row appears below a line item, the line item MUST be calculated from that driver via a formula. Driver rows are never decorative — they must mechanically drive the line above them. A hardcoded estimate in a row that has a corresponding driver row beneath it is a model error.

Specifically:

• If a Y-o-Y growth rate row sits below a revenue or KPI line, the line item must be calculated as: = prior period value × (1 + growth rate). The growth rate cell is the blue editable input; the line item above is the black formula output.

• If a margin assumption row sits below a cost or expense line, the cost line must be calculated as: = revenue (or applicable base) × margin %. The margin cell is the blue input; the cost line is the formula.

• If a per-unit driver (ARPU, take-rate, ASP) sits below a revenue line alongside a volume/count row, revenue must be calculated as: = volume × per-unit metric. Both drivers can be blue inputs; revenue is the formula.

• For ARR-driven models: If an ARR line drives subscription revenue, the formula must be: Subscription Revenue = (Beginning ARR + Ending ARR) / 2 (or the company’s disclosed conversion methodology). ARR is the input; revenue is the output.

• Actual/historical periods should remain hardcoded from SEC filings. Only estimate periods must be formula-driven from their drivers.

Rule: Share Count Must Be Formula-Driven

(See Section 7 for the full share count build.)

Rule: Architectural Choices Must Be Flagged

If you make a non-obvious modeling choice, add a gray italic note row explaining the choice. Examples:

• Note: Azure revenue is modeled using reported segment growth rates rather than bottom-up consumption drivers because Microsoft does not disclose workload-level KPIs.

• Note: CFO is derived from FCF + Capex. FCF is the independent forecast driven by the FCF margin assumption.

• Note: ARR is estimated as subscription revenue × 4 (annualized from quarterly) because the company does not disclose ARR directly.

• Note: SBC forecast uses % of revenue as the driver rather than per-employee cost × headcount because the company does not disclose headcount guidance.

Self-Audit Checklist (Run Before Delivery)

After building the model, perform each of these checks:

1. Driver Linkage Test: For every row pair where a driver sits below a line item, mentally set the driver to 0% (or zero). Confirm that the line item above would go flat (for growth rates) or to zero (for margin drivers). If it wouldn’t, the linkage is broken — fix it.

2. Hardcode Scan: In all estimate-period columns, check every revenue, KPI, and cost row that has a driver row beneath it. If the estimate cell contains a number rather than a formula, it’s a broken linkage. Fix it.

3. Formula Consistency Check: Verify that the same formula structure is used across all estimate-period columns for each row. A formula in 2026E that differs structurally from 2027E is usually an error.

4. Blue Input Audit: Confirm that every blue-font cell is genuinely an editable input (not a formula), and that every formula cell is in black font (not blue). Color-coding errors create confusion about what’s an assumption vs. a calculation.

5. Top-Down Cross-Check: Verify that Total Revenue = sum of segment revenues. Verify that EBITDA = Revenue − COGS − OpEx (or however it’s constructed). Verify that EPS = Net Income / Diluted Shares. Flag any rounding differences in a note.

6. SBC Consistency Check (Tech-Specific): Verify that SBC on the P&L → feeds diluted share count build → and the buyback offset flows through the cash flow statement. All three must be internally consistent.

7. ARR/Revenue Consistency Check (SaaS-Specific): If you built an ARR model, verify that implied subscription revenue from the ARR build matches the subscription revenue line on the P&L within 1-2% (the gap is timing/FX). If >2% gap, add a reconciliation note.

8. Report Results: At the end of your response, include a brief “Model Audit Summary” confirming you ran these checks, listing any issues found and fixed, and noting any architectural choices flagged.

Section 9: Valuation Sensitivity Tab

Sensitivity Matrices

1. Revenue Growth vs. EV/Revenue → implied share prices (most relevant for high-growth, pre-profit tech)

2. Revenue Growth vs. EV/Gross Profit → implied share prices

3. EBITDA Margin vs. EV/EBITDA → implied share prices (the “what if margins expand?” matrix)

4. EPS Growth vs. P/E → implied share prices

1. Show both GAAP and Adjusted (ex-SBC) versions for EBITDA and EPS matrices if SBC impacts either metric by ≥10%.

DCF Models

1. Perpetuity Growth Method: Explicit forecast period FCFs discounted at WACC, terminal value using Gordon Growth perpetuity.

2. Exit Multiple Method: Explicit forecast period FCFs discounted at WACC, terminal value using terminal year EBITDA × exit multiple.

1. Both methods should have editable assumption cells (WACC, terminal growth rate, exit multiple) in blue font.

1. For high-growth tech: Include a “fade schedule” memo showing the implied revenue growth deceleration path from current growth to terminal growth. If the fade is unrealistically fast or slow, flag it.

Reverse DCF / Implied Terminal Value (Detail)

1. Full calculation: Implied TV = Current EV − PV(explicit period FCFs)

2. Derived outputs: implied terminal FCF yield, implied terminal growth rate, implied terminal EBITDA multiple, implied terminal FCF margin

3. Sensitivity matrix 1: WACC vs. implied terminal growth rate

4. Sensitivity matrix 2: WACC vs. implied terminal EBITDA multiple

1. Sensitivity matrix 3 (Tech-Specific): Implied terminal FCF margin vs. terminal growth rate at current stock price — this is the “what does the market believe?” exhibit. For a 30x revenue SaaS company, showing that the market implies 35% terminal FCF margins and 4% perpetual growth is more analytically useful than a standard DCF.

This is the key analytical tool for surfacing “zero terminal value” situations. If implied terminal growth comes back negative on a recurring-revenue business, that’s the bull case in a single number.

Section 10: Formatting Conventions

1. Dark blue header bars for major sections

2. Light blue section sub-headers

3. Gray italic font for growth rates, memo lines, and architectural notes

1. Blue font for all user-editable input cells (growth rates, assumptions, multiples)

2. Black font for all formulas and calculated values

1. Title: “Company Name (TICKER) Earnings Model 2024–2028”

2. $ millions unless otherwise noted; specify units in headers

3. Percentages to one decimal (0.0%); multiples formatted as 0.0x

4. Negative numbers in parentheses, not with minus signs

5. Years formatted as text (”2024” not “2,024”)

1. Segment sections should be visually separated with thin borders and sub-headers

2. KPI sections below each segment should be indented and in a slightly lighter font weight to distinguish them from financial line items

Lastly: make sure you have completed the Sensitivity tab and the Reverse DCF tab as required. These are not optional — they are the analytical core of the model for an investor.

Hearing that Anthropic uses Workday, I asked Claude to analyze the buy v’s build case. I am not a SaaS analyst but I’m interested in this topic in as far as it affects other segments of the market. Granted they started to install WDAY a year or so ago, maybe they’d do it differently in 2028, but the key things to note are those ‘real world’ reasons rather than the code aspect. And in particular, those things people are forgetting right now which are the real world, all-in costs. The market is taking a very simplistic and knee jerk reaction to this whole AI v’s SaaS thing, some comparing it to the newspaper industry’s demise due to the web which when you actually think about it is just a weak but convenient argument.

A couple of caveats - Anthropic is a massive market cap company now and is maybe preparing to go public, that may play a part in going this way, and maybe Claude Code’s recent upgrades would change this decision? But at the same time this report suggests using WDAY is a no brainer - as the costs of using Workday at ~$15-25M vs. Build at ~$40-60M. I have seen a lot of variance in such numbers, but generally the all-in costs of an internal build always seems to come in well above the costs of using an existing SaaS, and it’s unlikely to change. Will there be other impacts on SaaS companies’ growth - maybe, but so far, it’s mostly opinion rather than evidential.

Here is my simple prompt for Claude to explore using Workday v’s internal build

I saw that Anthropic uses workday to administer various internal functions. When did this occur and given the ability of Claude Code to replicate this functionality, why would Anthropic do it, and what are reasons not to do it. Write a comprehensive report of the pros and cons and assign a score of importance from 1-10 for each, and then write a final recommendation. Consider the timeframe as 7 years. Don’t be guessing or extrapolating just come to a firm conclusion

This is what Claude Produced in Response

INDEPENDENT RESEARCH

Anthropic’s Workday Adoption

Build vs. Buy in the Age of AI-Native Infrastructure

A Comprehensive Evaluation of the Strategic Decision

to Deploy Third-Party Enterprise Software at an AI Frontier Lab

7-Year Horizon Analysis | February 2026

RECOMMENDATION: BUY WORKDAY

Executive Summary

Anthropic is actively implementing Workday across both HCM and Financials modules, scaling toward what the company itself describes as “public company readiness.” The company is currently hiring senior Workday specialists—including an HRIS Lead, Senior Workday Integrations Engineer, and a Senior Business Systems Analyst for Workday Financials—which indicates the implementation is either in-flight or in a late planning phase as of early 2026.

The question this report addresses is deceptively simple: why would the company that builds Claude Code—arguably the most capable AI coding agent in the world—choose to buy Workday rather than build its own equivalent? The answer, as we’ll show, is that the question itself misframes the problem. Workday isn’t a coding problem. It’s a compliance, legal liability, regulatory, and institutional trust problem that happens to be delivered as software.

Our recommendation over a 7-year horizon is unambiguous: Anthropic should buy Workday. The decision is correct, and attempting to replicate this functionality with Claude Code or any internal AI tooling would be a strategic error with a negative expected value of roughly $15-30M over the evaluation period after accounting for hidden costs.

Timeline & Context

Anthropic was founded in 2021 with 7 employees. By September 2024 the headcount reached approximately 1,035, and as of January 2026 various sources place the figure between 3,000 and 4,074 (depending on whether contractors are included). The company raised $30B in its Series G in February 2026 at a $380B valuation, with a stated $14B revenue run rate.

The precise Workday go-live date is not publicly disclosed. However, several data points bracket the timeline:

• Compensation Partner role (posted late 2024): Referenced “implementation of compensation processes in Workday,” suggesting Workday HCM was either live or in active deployment by that point.

• HRIS Lead role (posted 2025): Seeks someone to support “Workday implementation & future updates” across Core HCM, Benefits, Compensation, Payroll, Absence, and Time Tracking. This is a full-suite rollout.

• Senior Workday Integrations Engineer (2025): Focused on connecting Workday with Anthropic’s broader tech stack using Studio, EIB, Cloud Connect, and custom APIs.

• Finance Systems BSA (current, 2026): Internal functional lead for Workday Financials implementation, explicitly tied to “scaling toward public company readiness.” Involves FDM design, Chart of Accounts, Prism Analytics, and SOX compliance artifacts.

Best estimate: Workday HCM went live in late 2024 or early 2025. Workday Financials is currently in implementation, likely targeting a 2026 go-live. This is consistent with the standard Workday phased deployment pattern (HCM first, Financials second) and Anthropic’s IPO readiness trajectory.

The Core Question: Why Not Build It?

This is the question the BuccoCapital tweet surfaced and the one that matters for the SaaS bear thesis. If Anthropic can build Claude Code, why can’t it build its own HR and finance system? The answer falls into several categories.

Category 1: Regulatory & Legal Compliance

This is the single most important reason Workday exists and the one that kills the “build it” argument.

• Payroll tax compliance: Federal, state, and local payroll tax withholding, remittance, and reporting across every jurisdiction where Anthropic has employees. With offices expanding to Dublin, London, Zurich, Tokyo, and planned growth in Germany, Austria, India, Australia, NZ, Korea, and Singapore, this becomes a multi-country tax and labor law problem. Getting this wrong creates personal liability for officers.

• SOX readiness: The Financials job posting explicitly references SOX compliance artifacts—data lineage, controls, audit trails. A homegrown system would need to pass external audit scrutiny for SOX 404 internal controls over financial reporting. Auditors (Big 4) have pre-built testing frameworks for Workday. A bespoke AI-generated system? They’d need to build the audit program from scratch, at enormous cost to both parties.

• Benefits administration: ACA compliance, COBRA, 401(k) integration, ERISA fiduciary requirements. These involve regulated third-party interfaces with benefits carriers, retirement plan administrators, and government reporting agencies.

• Employment law: I-9 verification, EEO-1 reporting, FLSA classification tracking, state-specific leave laws (FMLA, CFRA, paid family leave), GDPR/CCPA for employee data. These aren’t “features”—they’re legal obligations that change constantly.

Category 2: Institutional Trust & Third-Party Ecosystem

Workday’s value isn’t just the software—it’s the ecosystem of trust around it.

• Auditor acceptance: Big 4 firms have dedicated Workday practice groups. When Deloitte or PwC audits Anthropic’s financials pre-IPO, Workday’s SOC 1/SOC 2 reports and built-in controls framework dramatically reduce audit scope, time, and cost. A custom system would trigger extended testing, qualified opinions, or material weakness findings.

• Insurance & banking: D&O insurers, lending institutions, and banking partners assess operational risk partly through the maturity of back-office systems. Running on a validated ERP is table stakes.

• Employee expectations: Anthropic is hiring world-class talent at median total comp of $336K. These candidates expect W-2s to be correct, benefits enrollment to work, equity vesting to be tracked accurately, and expense reports to be processed without AI hallucinations.

• Board & investor expectations: At a $380B valuation with institutional investors including Coatue, GIC, Fidelity, and D.E. Shaw, the finance and governance infrastructure needs to be institutional-grade. Period.

Category 3: Opportunity Cost & Engineering Allocation

This is where the math gets brutal.

• Engineering cost: Building an ERP equivalent would require a dedicated team of 15-25 engineers and domain specialists working for 2-3 years minimum. At Anthropic’s compensation levels, that’s $15-25M in loaded cost before you write a single paycheck through the system—and those engineers would be pulled from frontier AI research, the actual business.

• Maintenance burden: Tax codes change. Benefits regulations change. Labor laws change across dozens of jurisdictions. A homegrown system requires perpetual maintenance staff. Workday handles this through its bi-annual release cycle across all tenants.

• The counterfactual: Those same 15-25 engineers, deployed against Claude’s capabilities, would generate orders of magnitude more enterprise value than saving a few hundred thousand dollars per year in Workday licensing.

What Could Claude Code Actually Do?

To be fair, we should steel-man the build case. Claude Code and Cowork are genuinely capable tools. Here’s what they could theoretically handle:

• Custom internal dashboards: Headcount tracking, org charts, compensation analysis dashboards. Claude Code could absolutely build these faster than configuring Workday reports.

• Workflow automation: Approval routing, PTO request processing, expense categorization. These are relatively straightforward CRUD applications with business logic.

• Analytics layer: Prism Analytics is powerful but Claude could build bespoke analytics on top of raw Workday data exports that are more tailored to Anthropic’s specific needs.

• AI-enhanced processes: Resume screening, job description generation, compensation benchmarking analysis, policy Q&A bots. These are AI-native functions where Claude genuinely outperforms Workday’s own AI features.

The key insight: Claude Code’s strength is as a complementary layer on top of Workday, not a replacement for it. This is precisely the architecture Anthropic appears to be building—evidenced by the Workday Integrations Engineer role focused on connecting Workday to Anthropic’s broader tech stack including BigQuery.

Scored Evaluation Framework

We evaluate each dimension on a 1-10 importance scale (10 = existential, 1 = cosmetic) and assess whether the factor favors Buy (Workday) or Build (Claude Code / internal). The analysis assumes a 7-year horizon through approximately 2033.

DIMENSION

SCORE

ASSESSMENT & VERDICT

Payroll & Tax Compliance

10

BUY. Non-negotiable. Multi-country payroll with real-time tax table updates, garnishment processing, and automated government reporting. Error = personal officer liability. Claude Code cannot replicate the regulatory maintenance cadence.

SOX / Audit Readiness

10

BUY. Pre-IPO imperative. Workday’s SOC reports, pre-mapped controls, and auditor familiarity reduce IPO audit costs by $2-5M and timeline by 3-6 months vs. a bespoke system.

Benefits & ERISA Compliance

9

BUY. Regulated carrier integrations (medical, dental, 401k, FSA/HSA) with government reporting. Not a coding problem—it’s a regulatory interface problem with 50+ external counterparties.

Global Labor Law Compliance

9

BUY. Anthropic is expanding to 10+ countries. Each has unique employment law, data residency, and privacy requirements. Workday maintains jurisdiction-specific configurations.

Data Integrity & Single Source of Truth

8

BUY. Workday’s single-database architecture ensures one canonical employee record across HR, finance, and payroll. Bespoke systems invariably develop data drift between modules.

Institutional Credibility

8

BUY. Board governance, investor reporting, banking relationships, D&O insurance all require institutional-grade back-office infrastructure. This is a prerequisite for a $380B company.

Third-Party Integration Ecosystem

7

BUY (with nuance). Workday’s pre-built integrations with ADP, Fidelity, carriers, background check providers, etc. represent years of certified connections. But Claude Code could build custom integrations faster for non-standard use cases.

Opportunity Cost of Engineering Talent

9

BUY. Every engineer diverted to ERP development is one fewer working on frontier AI. At Anthropic’s stage, the marginal value of an engineer on Claude >>> marginal value on payroll software.

Vendor Continuity Risk

5

SLIGHT BUILD EDGE. Workday could raise prices, change terms, or be acquired. But the switching cost argument applies symmetrically—migrating off a bespoke system is equally painful. Workday’s $80B+ market cap provides stability.

Speed-to-Capability

7

BUY. Workday HCM deploys in 3-9 months. Building equivalent functionality from scratch takes 2-3 years minimum, and you still don’t have the compliance certifications.

AI-Native Workflows

6

BUILD EDGE. Claude genuinely outperforms Workday’s AI for tasks like comp analysis, candidate evaluation, policy generation, and anomaly detection. But this works as a layer on top of Workday, not a replacement.

Customization & Agility

5

BUILD EDGE. Workday is configurable but within its guardrails. Claude Code could build highly custom workflows faster. However, most “customization needs” reflect process immaturity, not genuine edge cases.

Long-Term AI Displacement of SaaS (7yr horizon)

4

NEUTRAL. By 2033, AI agents may handle more back-office functions autonomously. But the compliance substrate—tax tables, carrier interfaces, audit frameworks—will still require a certified platform. The UI layer may thin out; the system of record won’t disappear.

Total Cost of Ownership

7

BUY. Workday for ~4,000 employees across HCM + Financials likely runs $1.5-3M/yr all-in. Building equivalent: $15-25M upfront + $3-5M/yr maintenance. TCO over 7 years: Workday ~$15-25M vs. Build ~$40-60M.

Cost Sensitivity Analysis

Let’s put real numbers on this. The estimates below are based on publicly available Workday pricing benchmarks, Anthropic’s disclosed headcount trajectory, and standard implementation cost ratios.

Workday (Buy) — 7-Year TCO

Component

Annual

7-Year Total

HCM + Financials subscription (~4,000 FSEs @ ~$42 PEPM)

$2.0M

$14.0M

Implementation fee (HCM + Financials, one-time)

—

$3.0-4.0M

Internal Workday team (3-4 FTEs @ ~$250K loaded)

$0.8-1.0M

$5.6-7.0M

Implementation partner retainer / hypercare

$0.2-0.3M

$1.4-2.1M

TOTAL (Buy)

—

$24-27M

Internal Build — 7-Year TCO

Component

Annual

7-Year Total

Initial build (15-25 eng × 2-3 yrs @ $400K loaded)

—

$15-25M

Ongoing compliance/maintenance team (8-12 FTEs)

$3.2-4.8M

$22-34M

External legal/compliance advisory

$0.5-1.0M

$3.5-7.0M

Audit premium (extended procedures for bespoke system)

$0.3-0.5M

$2.1-3.5M

Integration build & maintenance

$0.3-0.5M

$2.1-3.5M

TOTAL (Build)

—

$45-73M

Note on headcount sensitivity: If Anthropic grows to 8,000-10,000 employees by 2033 (plausible given trajectory and international expansion), the Workday subscription increases proportionally but so does the compliance burden on any build alternative—the gap actually widens.

The Case Against Workday (Steel-Manned)

In the interest of rigor, here are the strongest arguments against the Workday decision:

1. The Optics Problem

Anthropic is the company that claims AI will displace 50% of entry-level white-collar jobs within 1-5 years. Buying Workday—the epitome of traditional enterprise SaaS—creates a narrative tension. If Claude can’t manage Anthropic’s own payroll, why should a Fortune 500 CFO trust it with theirs? This is a real but manageable tension; the counterargument is that even the best electrician hires a plumber.

2. Vendor Lock-In Over 7 Years

Workday contracts typically include 1% + CPI annual price escalators. Over 7 years, this compounds meaningfully. More importantly, the data model, integrations, and institutional knowledge all become Workday-specific. If a genuinely superior AI-native platform emerges by 2030, switching costs could be $5-10M+.

3. AI Capability Curve

The strongest version of the build argument isn’t about today—it’s about 2029-2033. If AI agents can reliably handle multi-jurisdictional payroll compliance, benefits administration, and financial controls with full auditability by then, Anthropic would be stuck paying for an incumbent platform while competitors run on AI-native infrastructure at a fraction of the cost. This is a real risk, but we assign it low probability for reasons discussed in the recommendation.

4. Workday’s Own AI Bet

Workday is aggressively deploying its own AI features. Anthropic may find itself in the awkward position of using a competitor’s inferior AI embedded in its own back-office system. This creates friction, potential data leakage concerns, and missed opportunities to showcase Claude’s capabilities internally.

5. Dogfooding Gap

There is a legitimate product development argument that using Claude to manage internal operations would generate invaluable feedback data, edge cases, and use cases that improve Claude’s enterprise capabilities. By outsourcing to Workday, Anthropic loses this feedback loop. However, the risk of using production payroll as a testbed for experimental AI systems far outweighs this benefit.

Final Recommendation

VERDICT: BUY WORKDAY. Confidence: HIGH.

The recommendation is firm and does not require hedging. Here’s why:

The compliance substrate is non-negotiable. Payroll tax compliance, benefits administration, and financial controls are regulated activities with personal liability for officers. No rational CFO or General Counsel would sign off on running these on an experimental AI system, no matter how capable. This alone is dispositive.

The audit path is clear. Anthropic is on an IPO trajectory. Workday provides auditor-friendly infrastructure that reduces IPO audit costs and timeline. A bespoke system adds $2-5M in audit costs and 3-6 months in delay. At Anthropic’s valuation, any IPO delay has an enormous carrying cost.

The opportunity cost is decisive. The 15-25 engineers required to build and maintain an internal ERP would generate more value—by orders of magnitude—working on Claude. At $14B revenue run rate, every marginal improvement in Claude’s capabilities has 1000x+ the ROI of a marginally better internal payroll system.

The 7-year AI displacement risk is overstated. Even if AI agents can handle more back-office tasks by 2033, the compliance infrastructure (tax tables, carrier APIs, government reporting interfaces, audit frameworks) will still require a certified platform. The user interface may become AI-driven, but the regulated plumbing underneath doesn’t go away. Workday will likely be one of the platforms delivering that AI-enhanced experience.

The optimal architecture is hybrid. Workday as the system of record and compliance engine, with Claude as the intelligence layer on top. This is exactly what Anthropic appears to be building, given the Workday Integrations Engineer role that connects Workday to BigQuery and the broader data infrastructure. This is the right answer.

The cobbler’s children wear shoes made by a shoemaker. That’s not a failure of capability—it’s a recognition that comparative advantage is the foundation of every rational capital allocation decision.

Implications for SaaS Investors

BuccoCapital’s tweet framing is correct in direction but incomplete. The fact that Anthropic uses Workday does not “kill the bear case” for SaaS—it merely kills one specific bear case (vibe-coding critical infrastructure). The remaining bears—platform commoditization, agentic layers cannibalizing system-of-record economics, and AI-driven compression of per-seat pricing—are all still alive and relevant.

What the Anthropic/Workday case does tell us is that for SaaS vendors deeply embedded in regulated workflows (payroll, financial reporting, healthcare, legal), the moat is significantly wider than for SaaS vendors selling productivity tools, analytics dashboards, or collaboration software. The competitive dynamics are fundamentally different. The former group’s moat is compliance; the latter’s moat is user habit. AI erodes habit; it does not erode regulation.

For your equity research work: This framework suggests overweighting SaaS names with deep regulatory embeddedness (Workday, Paylocity, Veeva, nCino) and underweighting those whose primary value proposition is workflow automation or UI convenience that AI agents can replicate.

Disclosures & Methodology

This analysis is based on publicly available job postings on Anthropic’s careers page, third-party employee databases (Tracxn, TrueUp, ZoomInfo), publicly available Workday pricing benchmarks, and standard ERP implementation cost frameworks. No material non-public information was used. Cost estimates are directional and based on industry benchmarks; actual Anthropic costs may differ. The author has no position in WDAY or any private Anthropic securities.