But time and again, history has shown us that when a general-purpose technology arrives, it rarely delivers immediate gains. Instead, it triggers a long period of stagnation—a structural pause where society attempts to force a radically new capability into old institutional containers. The breakthrough itself is merely the spark. The real transformation happens in the decades that follow, through an invisible reorganization of how human beings structure work, time, and authority.

Neil Postman noted that technological change is never simply additive. It is ecological. If you remove caterpillars from an ecosystem, you are not left with the same environment minus caterpillars; you have reconstituted the conditions of survival. A new technology does not just perform existing tasks faster. It changes the entire habitat.

To understand where AI is taking us, we have to look past the immediate capabilities and trace the invisible reorganizations of the past.

When steam power arrived in the late 18th century, the visible invention that resulted from it was the engine. But the shift that actually reshaped society was the clock. Not following? Before the industrial revolution, work was outside factories and it was task-oriented, dictated by daylight, seasons, and natural rhythms. Inside the factory that the industrial revolution created, work became clock-timed. Employers instituted fines for lateness, locked doors at the bell, and transformed time from something lived into something spent. Data shows us that working-class real wages remained stagnant for nearly 70 years after the industrial revolution began—a period known as Engels’ Pause. Society was accumulating output, but the institutional shift was so disorienting that the benefits took generations to materialize. The steam engine may have produced textiles; but the clock radically shifted how we worked and resulted in the modern workday.

A century later, electricity followed the exact same trajectory. When power grids spread in the 1880s, factory owners simply unbolted their central steam engines and wired up single electric motors in their place. For 40 years after that, industrial productivity barely budged. It was only in the 1920s, when a new generation of managers realized they could discard the old multi-story architecture entirely, that output exploded. They redesigned factories into sprawling, single-story modular plants with unit-drive motors on every machine. Beyond the factory floor, cheap power collapsed the boundary between day and night, giving rise to shift work, 24-hour cities, and eventually, the synchronized national radio audiences that created modern mass culture. Again, the same pattern: the lightbulb illuminated rooms; but what really came out of this was the creation of a power grid that enabled national media.

By the time the personal computer and the internet arrived, the pattern repeated itself again. In 1987, economist Robert Solow famously remarked that the computer age could be seen everywhere except in the productivity statistics. Macro-productivity gains did not surge until the late 1990s—a 20-year delay. The lag was not a failure of the microchip; it was the time required for organizations to dismantle their paper-based operations. In that transition, the pre-digital landscape dissolved. Print newsrooms collapsed, retail migrated to digital platforms, and the traditional concept of a career began to unbundle into contract and platform labor.

Across every era, the sequence remains identical. First comes the capability, which is immediately overhyped and misapplied. Next comes the productivity delay—the uncomfortable J-curve where old organizations break down before new ones take hold. Finally, the true outcome emerges: an invisible container that no one predicted at the start.

Look at the financial markets in early 2026 for proof that we are living through this exact J-curve lag right now. After years of record capital expenditure into AI infrastructure, Wall Street reached a tipping point. Heavyweights from Big Tech to chipmakers faced sharp market pullbacks as investors began demanding tangible ROI instead of promises. The market’s reaction was entirely predictable: companies scrambled to defend their valuations through aggressive cost-cutting and mass layoffs, slashing tens of thousands of roles to pad margins and pump stock prices.

Yet, despite the headlines and short-term share price recoveries, none of this downsizing has triggered a structural breakthrough. Trimming headcount to soothe public markets while grafting AI tools onto legacy org charts is not an institutional shift. It is the modern equivalent of unbolting a steam engine and dropping an electric motor in its place—a brute-force financial reflex that attempts to manufacture productivity out of accounting tricks rather than genuine reorganization.

This historical pattern reveals the fundamental flaw in how we evaluate AI today. We are busy measuring token speeds, benchmark evaluations, quarterly earnings calls, and corporate rollouts, expecting a 2-year spike in productivity. But capabilities are cheap, and raw technical power rapidly commoditizes.

The real transformation of AI will not be faster software engineering, automated document drafting, or corporate cost-cutting. It will be the decomposition of the traditional firm. The industrial revolution gave us the corporation; electrification gave us mass media; the internet gave us digital platforms. AI will produce a new organizational container—hyper-lean, agent-orchestrated entities that execute complex operations with a fraction of the human overhead previously required.

The leaders who build lasting value in this moment will not be the ones racing to adopt the fastest model or cutting headcount to satisfy a quarterly earnings target. They will be the ones who recognize the institutional lag, accept the operational tradeoffs, and design the workflows, governance structures, and habits that wrap around the new technology.

Stop asking what the machine can do today. Start looking at what it is quietly forcing you to rebuild.

Hi, I’m Tanya, founder of PurInsights. I work as a fractional strategy and operations partner for founders, helping them bridge the gap between big technical capabilities and the actual operational containers needed to execute. If you’re staring down the “how do we actually reorganize around this” question, you know where to find me.

I was in the trenches with them. On Sunday nights, my husband and I made runs to Costco to stock the office pantry. I organized snacks while he swiffered the floors before clients walked in on Monday. I helped in any way I could, while also holding down a demanding role at Credit Karma.

In those early days, there was no budget to outsource any of it, even though it was unglamorous. Every dollar spent on “operations” was a dollar that couldn’t go toward growing the company. The founders went without salaries for a full year to protect payroll, keep the lights on, and extend runway long enough to make their first critical hire.

That experience shaped one of the core beliefs behind PurInsights: your time is not free.

Founders are often incredibly disciplined about cash, and surprisingly undisciplined about their own bandwidth. I see founders spending hours on contracts, NDAs, bookkeeping, invoicing, and administrative work because paying someone else feels expensive. But doing it yourself has a cost too.

Every hour a founder spends reconciling QuickBooks is an hour they are not selling, recruiting, building partnerships, talking to customers, raising capital, or making the decisions only they can make.

The question isn’t simply whether you can afford to hire help. It’s: what is the highest-value use of your time as a founder, and what is it costing the company when that time gets spent somewhere else?

That’s a big part of the work I do at PurInsights. Yes, I build runway models and financial forecasts. But really I’m helping you see, via the data, what the tradeoffs actually look like. If your round closes in August versus October, what does that mean for adding capacity? If hiring someone buys the founder back 15 hours a week, what could those 15 hours produce?

Where should you stay scrappy, and where has “scrappy” started quietly constraining growth? It’s a nuanced question and there isn’t always a clean framework to reference.

Having lived through the bootstrap grind, I have enormous respect for every dollar a founder spends. I also know there comes a point when protecting cash too aggressively starts costing the business more than it saves.

Founders: what are you still doing yourself that your company may have outgrown?

I’m Tanya, founder of PurInsights. I help post-raise founders figure out where to deploy capital when the growth path isn’t obvious, and find the risks hiding in the plan before the board does. If that’s the seat you’re in, you know where to find me on LinkedIn.

Subscribe now

NFX, one of the seed-stage VC shops that helped canonize the moat conversation, built their Network Effects Bible around it and argued that network effects account for roughly 70% of the value created in tech since 1994. Every pitch deck and almost every bizops deck I worked on had to have a data flywheel slide: more users, more data, better model, more users. Uber’s routing data was a moat. Google’s clickstream was a moat. Palantir was selling data integration as defensibility. When I was at BlackRock and later Credit Karma, “we have proprietary data” was the answer to almost every question a board could ask about long-term durability. It wasn’t wrong, exactly. It was just the same answer almost everyone was quoting.

Then in May 2019, a16z’s Martin Casado and Peter Lauten published “The Empty Promise of Data Moats” and quietly marked an end to that era. Their argument was that what most companies had was not a data network effect but a data scale effect, and in their own words, “the exact opposite dynamic often plays out with data scale effects: the cost of adding unique data to your corpus may actually go up, while the value of incremental data goes down.” Implying that data alone doesn’t retain users. Their prescription for what actually does: packaging differentiated technology, verticalizing across industries, winning the go-to-market race, and winning the talent war. Every one of those turns out, in retrospect, to be a workflow argument in disguise.

Then GPT arrived and exacerbated it. If a competitor can spin up a comparable model in a weekend, it doesn’t matter if you have proprietary data. The proprietary data itself becomes commodity input for someone else’s model.

Which is why, imo, the new moat is workflow depth.

Sequoia’s Sonya Huang and Pat Grady called it back in 2023’s “Generative AI’s Act Two.” They argued that the inherent value of a company would move away from foundation models toward the application layer, and the durable thing at the application layer would be workflows and user networks, not the underlying model. In January 2025, a16z followed up with a claim that sent the markets crashing - that the per-seat SaaS model itself is dead: if AI can automate work rather than assist a human doing it, then software eats labor budget, not seat budget, and the moat has to be whatever binds the customer’s labor to your product. Stanford Law’s Mandal and Sinha put it into a framework in a June 2026 paper: five moats for vertical AI companies in ascending order of durability, with workflows plus UX plus vertical-specific tooling sitting above raw data or brand every time.

2025-2026 shows us how this thesis is playing out.

Cursor, the AI-native code editor (basically a Copilot competitor that owns the whole IDE surface, not just the autocomplete), hit over $500M ARR at Series C in June 2025 with a team of roughly 60 people, crossed $1B ARR by early 2026, and a year later, SpaceX made a bid to acquire them for $60B, the largest venture-backed startup acquisition on record. From what I can tell, their model isn’t dramatically better than what GitHub or Google ships. They just own the tool developers already live in, all day. Every keystroke, every accepted suggestion, every custom rule a team writes into their .cursorrules file becomes part of a switching cost that a competitor has to overcome. Now that is what SpaceX paid roughly 15x revenue for.

Similarly, Harvey, an AI platform built specifically for how law firms draft contracts, research case law, and review documents, went from $100M ARR in August 2025 to about $195M by January 2026, and is now deployed at 50% of the Am Law 100. Once a firm has trained its associates on Harvey and wired it into their document management system, ripping it out costs more than the license. There’s a compounding effect I find interesting here. Harvey reports that median seat count at a customer firm roughly doubles within 12 months of the initial deployment → the product goes viral inside the customer’s org making the switching cost that much higher.

Gong is one of the older workflow-embedded AI companies and still one of the cleanest proof points. It crossed $500M ARR in 2025 with over 55% YoY growth, its tenth straight quarter of acceleration, and publicly reports Net Revenue Retention consistently above 120%. Their moat is the accumulated conversation corpus, the coaching rules customers configure over years, and the way non-sales seats (customer success, executives, revops) now make up an increasing share of its expansion. Once your revenue org lives inside Gong’s playbook, you’re not switching for a better transcript.

Ramp and Rippling are worth pulling into this too. Ramp reached about $1.5B in annualized revenue by May 2026 and closed a Series F at a $44 billion valuation. Its moat is the ledger plus card plus procurement plus approval-tree architecture serving as the system of record for spend. Rippling crossed $1B ARR in 2025 and explicitly frames its moat as the “Workforce Graph”: a single source of truth for the employee record that’s wired into device management, app provisioning, payroll, and identity. Ripping either of them out isn’t a software migration, it’s a re-architecture.

Meanwhile, at the shallow end of the market, AI-native retention is falling apart. AI-native products priced under $50 per month retaining at roughly 32% NRR (meaning a cohort worth $100 today is worth about $32 a year from now i.e. not good), while products priced over $250 per month held near 85%. What actually separates them is the depth of the workflows - a $30 per month AI writing tool is easy to cancel while a $500 per month product your team has configured, integrated, and built playbooks inside, is not. To me, there is a story to be found in this bifurcation: workflow-embedded products compound.

The clearest counter-example is Jasper. In late 2022 it raised $131M at a $1.5B valuation on the strength of AI writing capability. Revenue reportedly peaked around $120M in 2023 and then fell to somewhere between $35M and $55M in 2024 after ChatGPT commoditized the wedge. Jasper had traffic, brand recognition, and a real product. What it did not have was configuration surface that customers had built anything durable inside. When the underlying model became commodity, so did the retention.

So, if you’re a post-raise founder thinking about how to explain your moat to your board, a few things I’d push you to consider:

1. Own the daily cadence, not the feature

2. Store the customer’s configured logic inside the product. Permissions, integrations, playbooks, prompt libraries, approval trees. That’s what makes the switching cost real, and it’s what Cursor, Ramp, and Rippling have in common.

3. Log every action the agent takes on the user’s behalf. The audit trail of what your AI did becomes the switching cost the customer feels most viscerally when they think about leaving.

4. Instrument for retention early. If cohort-two NRR is below cohort-one, the moat isn’t there yet, no matter what the deck says.

The catch is that building this kind of embedded product takes patient design work, not just aggressive selling, and the risk of over-fitting to any one customer’s process is real. But the alternative is your NRR sliding down to 32%.

So would your customers actually notice if you disappeared tomorrow? What are you seeing on your side? If you’re in a board seat or a founder seat right now, I’d love to hear how the moat question is landing with your team.

I’m Tanya, founder of PurInsights. I help post-raise founders figure out where to deploy capital when the growth path isn’t obvious, and find the risks hiding in the plan before the board does. If you’re staring down the “what’s our moat now” question, you know where to find me.

Subscribe now

So I went back to the source material: Andrew Ng’s Stanford course, Andrej Karpathy’s lectures on large language models, MIT’s intro machine learning curriculum, and the papers behind the terms everyone throws around. Here’s what I learned along the way, that I felt might be useful to others:

AI is the broad idea. Since the 1950s, it’s meant computers doing things that normally require human intelligence: recognizing a face, translating a sentence, playing chess. It simply describes a goal.

Machine learning is the method that won. Computer scientist Arthur Samuel defined it in 1959 as giving a computer the ability to learn without explicitly programming every rule. You show the system thousands of examples, and it finds the pattern.

Large language models are one specific application of that method. Karpathy’s framing is the simplest one I’ve found: an LLM is two files. A parameters file, holding everything the model learned from predicting the next word across a huge amount of text. And a run file, the code that executes those parameters. Prediction, run at a scale large enough to look like something else.

What “a huge amount of text” actually means. Training data comes from a handful of buckets: crawled web pages (Common Crawl is the biggest public source), digitized books, code repositories like GitHub, and forums like Reddit and Stack Overflow for conversational text.

What happens after that first pass. Labs add a second training stage called RLHF, reinforcement learning from human feedback. It happens after a model finishes learning to predict text. Human raters look at pairs of the model’s answers and mark which one they prefer. Those preferences train a separate “reward model” that learns to score outputs the way the raters would. Then the original model gets fine-tuned to maximize that reward model’s score. It’s the step that turns a raw next-word predictor into something that follows instructions, stays polite, and holds a conversation, rather than just autocompleting whatever text came before.

What RLHF actually optimizes for. Raters are instructed to reward nuance and multi-faceted thinking. Research on this stage suggests raters routinely conflate the appearance of nuance with actual depth, so a response that uses contrast framing, “it’s not X, it’s Y,” scores better even when it isn’t saying anything more substantive. It’s also a low-risk move for the model: it validates whatever the reader already assumed, then tacks on an elaboration, which makes it hard for a rater to flag the answer as wrong or unhelpful. Cheap to generate, hard to penalize. A fair amount of what reads as “AI voice” traces back to this one training step.

Two behaviors worth knowing. In-context learning is the model picking up a new pattern from examples you give it in the prompt itself, no retraining required. Hallucination runs on the same mechanism, pointed at you: the model produces a fluent, confident answer with no connection to what’s actually true, because fluency and accuracy are two separate things it learned. Researchers studying this described hallucination as a structural property of how these systems generate text, rather than a flaw that eventually gets patched out.

What agents add. A chatbot answers once and stops. An agent runs a loop: think about what to do next, take an action like using a tool or running code, look at what happened, decide the next step, repeat until the goal is done or it gives up. That loop is what lets a system book a flight, debug a program, or research a topic across a dozen sources instead of producing one paragraph. Model Context Protocol, released by Anthropic in late 2024, has become the standard way agents connect to outside tools, a big part of why agents moved from demos to daily use this fast.

What’s actually changed since late 2025. GPT-5 shipped in August 2025. Gemini 3 Pro and Claude Opus 4.5 both landed in November. GPT-5.2 followed in December, and by mid-2026 the frontier models, GPT-5.5 and Claude Sonnet 5, had pushed context windows up to a million tokens, roughly 750,000 words a model can hold at once. The bigger shift is what’s built on top of the models: reasoning systems now show their intermediate steps before an answer.

The core mechanism hasn’t moved - it’s still prediction, still trained on text and therefore still capable of being confidently wrong. But with time, the scaffolding around it has got a lot better at acting on what it predicts.

How I actually use this in the work I do

Every 90-day fractional engagement starts in a similar manner. An example - a founder just closed a round, the board wants to see capital deployed, and nobody has time for me to spend six weeks getting oriented. The first 2 weeks are a diagnostic sprint: close the books, map where the money actually went, find the vendor contract nobody’s re-read since the Series A. That’s the same kind of work I did the first time I taught myself QuickBooks and closed a year of my client’s company’s books in two months. I didn’t use AI to replace my instincts, but AI made it faster to point at a new company’s mess.

A few concrete differences from doing this job inside a large company:

No procurement cycle. A finance team inside a 500+ person company waits months for a new tool to clear security review, vendor risk assessment, and a committee that’s never met the problem the tool is meant to solve. I open Claude and start categorizing a client’s expenses the same afternoon I see their chart of accounts.

One person covering what used to take three. A controller closes the books. An analyst builds the model. A strategist tells the board what it means. In a 90-day engagement, I do all three passes myself, and AI is what makes that single-person coverage realistic instead of a corner-cutting exercise. It drafts the first pass of a model. I catch what it got wrong and decide what actually belongs in front of the board.

The pattern-matching gets a force multiplier. Finding the thing nobody else noticed, like the offshore staffing arrangement that didn’t match what a client thought they were paying for, used to mean reading every contract and invoice by hand. Now I can point a model at a stack of documents and ask it to flag anything that doesn’t match what I was told, then go verify the handful of things it surfaces myself. The instinct is still mine. But I save time on the first pass across a hundred pages.

Again: none of this replaces my judgment. A model will hand me a confident, wrong answer in the same tone as a correct one, which is exactly why the checking still has to be a person who knows what “right” looks like for this specific company. That’s the part I get paid for as a fractional operator. The tool just means I can get to the judgment call in days instead of weeks, and that matters a lot when a board is waiting on an answer and the runway clock doesn’t pause for the diagnostic phase.

What’s the AI question you still don’t have a clean one-line answer for? I want to hear it.

Subscribe now

Hi! I’m Tanya, founder of PurInsights (www.purinsights.com). I help post-raise founders figure out where to deploy capital when the growth path isn’t obvious, and find the risks hiding in the plan before the board does. If that’s the seat you’re in, connect with me on LinkedIn (https://www.linkedin.com/in/tanyapuri/) or DM me here on Substacks.

Sources

• Machine learning, explained | MIT Sloan

• CS229: Machine Learning, Stanford University

• Machine Learning Specialization, Andrew Ng / DeepLearning.AI

• Intro to Large Language Models, Andrej Karpathy

• Introduction to Machine Learning, MIT OpenCourseWare 6.036

• Agentic AI, explained | MIT Sloan

• Hallucination is Inevitable: An Innate Limitation of Large Language Models (arXiv)

• Are Emergent Abilities in Large Language Models just In-Context Learning? (arXiv)

• LLM Training Data: The 8 Main Public Data Sources

• State of AI Agents 2026: Autonomy is Here, Prosus

• What is RLHF? Reinforcement Learning from Human Feedback, Weights & Biases

• Why do LLMs so often say “It’s not an X, it’s a Y”? (LessWrong)

What I didn’t appreciate at the time is how much smoothing happens at every layer between the work and the board. The person doing the work tells their manager the real version. The manager softens it slightly for their boss. By the time it reaches a leadership review three or four layers up, the rough edges are gone, and so is most of the useful information. I was just one more layer in that chain, doing what every layer before me had already done.

That correction is one of the things that has stuck with me from that job. Over years sitting behind leaders in strategy and operations roles, building decks, prepping talking points, sitting in the room while they presented up, I started noticing a split.

The leaders who didn’t shy away from the honest lowlights were the ones whose teams actually fixed things. When a launch was behind, they said it was behind, and why, and what they needed to get back on track. No, leadership didn’t love hearing it in the moment. But their problems got resourced instead of buried, and the same problem rarely showed up twice. And these leaders won respect all around.

The leaders who only wanted to sell and who treated every leadership review like a pitch. Every slide said “great,” every metric had to trend up and to the right even when the underlying numbers told a messier story. My job, as they had laid it out for me, was to sell the story. The sell-only leaders ran teams that hid bad news from them the same way they hid it from leadership. Problems surfaced later, bigger, and angrier. People left. The upfront leaders kept their teams, and their teams kept telling them the truth, which is most of what makes a leader good at their job in the first place.

Leave a comment

I used to think my role was to protect my boss from the discomfort of an honest slide. The leaders worth working for did not need protecting from the truth. They needed someone who would put it in front of them straight, the same way they were about to put it in front of their own leadership.

I carry that lesson into my own work now, at my own fractional practice, PurInsights, sitting across from founders instead of behind their shoulder. The instinct ahead of these meetings is always the same: lead with the metric that is up, bury the one that slipped. I push the other way. The deck should show the missed number next to the one that landed and a plan to address the miss. Founders who do this build more trust with their board and their investors.

It matters more right now than it used to. A founder who just closed a round is reporting to a board that expects capital to move and metrics to follow. That pressure is exactly the pressure that turned good leaders into smoothers, in every job I’ve watched it happen. The founders who resist it, who keep the lowlight on the same page as the highlight, are doing something harder than it looks and rarer than it should be.

If this resonated, I write here every couple of weeks about fractional work, career structure, and the financial-operator lens on building a company from seed through Series B.

I’m Tanya Puri, founder of PurInsights. Fractional finance and operations support for founders who need someone to build the report that tells the board the truth, highlights and lowlights both. Find me on LinkedIn too, I read every comment.

Subscribe now

A lot of people are confused. Nothing obviously “bad” happened. So why is the stock falling?

Here’s what the market is actually doing, and why it was always going to do this.

The bond filing did what bond filings do

A week after the IPO, SpaceX launched its first ever entry into the public bond market, a $20B debt offering. To do that, it had to file disclosures with the SEC. Those disclosures contained numbers that weren’t front and center in the IPO prospectus: $14B in negative free cash flow last year, more than double the prior year. $29B+ in total long-term debt. A March bridge loan that the bond proceeds would mostly go to repay.

That bridge loan, by the way, was taken out to finance SpaceX’s acquisition of Musk’s own xAI startup. xAI posted a $6B operating loss in 2025 and burned $13B in capex, wiping out all of Starlink’s operating profit in the process. All 11 of xAI’s original co-founders left before the IPO. Musk himself said in March that xAI “was not built right first time around.”

This is not a distress signal

SpaceX has over $100B in cash. Investment-grade ratings from all three major agencies. The bond sale is routine post-IPO balance sheet management: swap expensive short-term bridge debt for long-dated institutional paper at better terms. The company also announced a $6B compute contract with Reflection AI the same week. So by any conventional read, the underlying business is not in trouble.

So why is the stock down?

Three mechanics converged at once.

First, the bond filing surfaced the income statement. At 20x plus revenue, SpaceX was valued like a company that has already solved AI infrastructure monetization. The filing made clear the capex build is just getting started, with Goldman and Evercore projecting SpaceX capex exceeding $1Tn by 2031 and S&P projecting negative free cash flow through 2029.

Second, the options market on SPCX opened the week of June 17. Before that, there was no practical mechanism to short SpaceX or hedge a long position. The bond announcement gave short sellers their first real narrative catalyst right as they got their first real tool to use it.

Third, the float is microscopic. Only 4-5% of SpaceX shares are currently tradeable. In a market that thin, modest selling moves prices fast and hard. The 20%+ drop does not require massive institutional exits. It requires a thin market meeting a new bearish mechanism. Worth noting: insiders could potentially sell 44% of shares by early September, which would increase the float by roughly 900%. Markets are starting to price that supply shock before it arrives.

What this means more broadly

The AI infrastructure trade is not over, but it is maturing. Investors are no longer asking “will AI be big?” They are asking “who captures the economics, and when?” SpaceX has Starlink as a real cash-generating business, DoD contracts, and credible compute ambitions. But it is also burning $14B a year net of capex, competing in a compute market where AWS, Google, and Microsoft have years of runway advantage, and absorbing an AI subsidiary that lost $6B last year with all its original founders gone.

The selloff is not a verdict on SpaceX as a company. It is a repricing of certainty. The stock went public pricing in a best case. The market is now pricing in a range of outcomes. That is how public markets are supposed to work. It just usually takes longer than a week.

Subscribe now

Hi, I’m Tanya Puri, the founder of PurInsights, a fractional CFO and strategy practice for seed to Series B founders. I write about capital markets, founder finance, and what the headlines miss. Follow me here and at linkedin.com/in/tanyapuri.

We came at it from two different desks. Stephanie sees it in contracts that get rushed and legal that only shows up once something’s already broken. I see it in books that don’t get touched until the mess is six months deep. Different functions, same pattern.

Here’s what came out of an insightful 30 minutes of comparing notes:

1. Don’t rush contracts. Rushing a negotiation to “just get it done” is exactly how risk gets baked in. Take the time to think through what could actually happen between the two parties, whether you’re still working together in two weeks or two years. Skip that, and skip something like a cost-escalation clause, and you’re the one carrying the risk when prices move.

2. Hire ops before it breaks. Books or systems don’t get cleaned up cheaply once they’ve been ignored for a year. They get cleaned up at cost: months of work, real stress, and a bill bigger than the part-time hire would have been. Bring someone in early, even a few hours a month, and the math changes.

3. ERP is a game changer, until you overengineer it. A system like Odoo can transform how a business actually runs. But the instinct to track everything down to the millisecond turns the tool into a burden instead of a help. The goal is the right level of detail, not full surveillance of your own operations.

4. Treat legal like a strategic partner, not a fire extinguisher. Most founders avoid bringing in a lawyer early because they assume it’ll slow things down. But by the time something’s a “big issue,” it’s usually too late to do much more than damage control. The earlier legal is at the table, the less reactive the fix has to be.

5. Reactive vs. strategic legal is about stage, not size. Smaller companies tend to call legal only after the CEO flags a problem. The more strategic ones, think Disney, have legal in the room while decisions are still being made, not after they’re made.

I know the objection: you don’t have the budget for a fractional ops person or a lawyer on retainer right now. Stephanie’s answer to that, and mine, is the same. It doesn’t have to be a full-time hire. It can be a few hours a month, brought in before the contract is signed instead of after it’s a mess.

We didn’t set out to make this episode a referendum on waiting too long. It just kept being the answer, from both sides of the table.

Which one are you avoiding right now: the contract, the ops hire, or the legal check-in? Tell me in the comments.

I’m Tanya, founder of PurInsights (www.purinsights.com). I help post-raise founders figure out where to deploy capital when the growth path isn’t obvious, and find the risks hiding in the plan before the board does. If that’s the seat you’re in, you know where to find me, or connect with me on LinkedIn.

The new job paid more. Maybe that was the part that made it less prestigious.

In the world he came up in, prestige meant title, longevity, and proximity to the public good. You worked your way up over decades. Salary was beside the point — it wasn’t supposed to mean anything about who you were. The private sector, when it existed at all in pre-liberalization India, was where you went when you couldn’t make the government track work. License-Raj-era India was structured so that the most consequential decisions about an Indian’s life — what got built, what got priced, what got allocated — happened inside government rooms. Anyone serious sat in those rooms.

My father walked away from 17 years of seniority, with a young family to support, into a sector that almost everyone he knew thought of as a step down. The fact that it paid better was, to his peers, suspicious. It implied a quality of work that mattered less, and a willingness to be evaluated by markets rather than by the slow logic of the bureaucracy.

Today that move reads as obvious career-building. In 1991, it was radical. India’s liberalization had just begun under Manmohan Singh — the economy was opening, foreign capital was about to start flowing, the private sector was about to become the engine of the next two decades of growth. My father made his move before any of that was obvious from inside the government. Most of his peers thought he was being reckless.

Second definition

A generation later, I started my career at Morgan Stanley, then BlackRock, then post-MBA at Credit Karma.

The script had flipped.

For me and most of my peers, prestige was the logo on the resume, the total comp number, and the speed of the title progression. It had almost nothing to do with the work itself, or whether the work made anything in the world better. We measured prestige in offer letters, and we measured each other against them. The MBA programs trained us to think this way. The consulting and banking firms reinforced it. The LinkedIn-era public display made it competitive.

My parents’ generation tied prestige to the institutions you sat inside. Ours tied it to what those institutions paid you. The center of gravity moved from the chair to the comp band.

It worked for a long time. Long enough that most of us stopped questioning it. The institutions kept rewarding the people who fit the template. The template said: pick the right brand, climb the right ladder, post the right milestone updates, do not leave for anything that doesn’t compress all three.

The inflection

Then last week I read the layoff lists from Credit Karma and a few other places. People I knew. People who mentored me. People whose careers fit exactly the template my generation had built — the right firms, the right titles, the right trajectory — and that template did not save them.

When you have built your prestige around what an employer gives you — title, brand, comp — then by definition the employer is the one who decides when you stop being prestigious. That decision can arrive in an email on a Friday afternoon, and there is nothing in the structure of your career that gives you any leverage over it.

That’s the part of the second definition I don’t think most of us were taught to see clearly. We were taught to optimize for what could be taken. We were not taught to ask what couldn’t.

Third definition

Here is what I think the third definition of prestige is becoming.

It isn’t title. It isn’t entirely pay. It’s the part of your career that only you own.

Skills that travel. A network you built, not one you inherited from a brand. The ability to make money without a single employer’s permission. Equity in something — even a small piece of something — that you actually own. Proximity to decision-making regardless of where you sit on the org chart. A reputation that is yours, not your employer’s. A point of view that gets read by people who would still read you if your title disappeared tomorrow.

None of these are easy to point at on LinkedIn. None of them fit cleanly on a resume. They take longer to build than the second-definition things, and the early work doesn’t look like anything from the outside.

But I think that these are the only things that are going to age well in an economy where employers can rewrite the terms of your career inside a single quarter.

I am five months into building my own version of this — a fractional operator practice that, by design, can’t be taken from me by a single employer making a single decision. I wrote about how I’m actually doing that here. The short version: it doesn’t feel like it’s working at first, and that is the part that’s actually working.

The question I have been sitting with — and the one I would ask any operator reading this — is the one underneath all of it: what part of your career, today, would survive losing your job tomorrow?

If this resonated, the practical follow-up — three things I have actually done to build a portfolio life — is here.

I’m Tanya Puri, founder of PurInsights. Fractional operator (finance, ops, strategy) for seed-to-Series B founders prepping for their next round or exit. If you are working through one of these definition shifts — your own or for someone on your team — you know where to find me.

Subscribe now

I was announcing my move from a roughly 15-year brand-name career into fractional work, and asking my network for leads — introductions to founders, VCs looking for an operator in the seat. I’d convinced myself this would be easy. Morgan Stanley, BlackRock, Credit Karma, the right MBA. Surely the introductions would come. I’d be billing multiple clients inside a quarter.

The email took me a week to write. It felt embarrassing, almost like I was exposing myself, which I kind of was.

What came back were dozens of congratulations. And literally zero leads.

So I sent another version a month later. And another. Different framings — sometimes asking for introductions, sometimes asking for thinking partners on a problem I was working through, sometimes just sharing what I had learned that month. Eventually, one of the replies came from someone I hadn’t worked with in nearly twenty years — and on that call he introduced me to The Portfolio Life by Christina Wallace. I’m still reading it. But I realized is that what she describes is what I had been unintentionally building since the January email. I just hadn’t had a name for it.

That realization is what I wrote about in last week’s LinkedIn post on prestige. The question I keep getting since is the same one I had to figure out the hard way: how do you actually build one?

Here is what I have done — including the parts that didn’t feel like they were working.

1. Build the public proof before you need it.

I started writing on LinkedIn and Substack months before any of it could possibly generate a lead. Most people only start posting once they need the work, which is exactly the wrong order. Reputation doesn’t compound under deadline pressure. It compounds slowly, in public, while you’re still doing something else.

The right time to start putting your thinking out is when you don’t need it yet. By the time you need it, the runway is already gone.

This is the lesson that feels least useful when you’re doing it. Posting into a void in front of an audience you don’t have, on topics nobody has asked you about, on a schedule no one will notice. Months of it. And then one day someone shares one of your posts to someone you’ve never met, and you discover that the people who had been quietly reading you for a year were the audience all along.

2. Do the free stuff before the paid stuff.

Earlier this year I collaborated with the incubator that runs AngelCon. Unpaid, no commercial outcome attached.

The first instinct most senior operators have when something is unpaid is to skip it. Your time has a market price; you shouldn’t sell below it. That’s the old reflex. The portfolio reflex is the opposite: free work that gets you into a room with the right people compounds faster than paid work that doesn’t.

AngelCon didn’t pay me a dollar. But it has already expanded my circle in ways no paid engagement could have.

“Free” work isn’t actually free. You are getting paid in surface area, in proximity, in the kind of conversations that turn into the next three years of your practice. The invoice just arrives later, from a different person, in a different form.

3. Make asking a rhythm, not an event.

Most people send their “I’m doing something new” email once. I sent mine again the next month. And the month after that — sometimes asking for intros, sometimes for thinking partners on a problem I was working through, sometimes just sharing what I was learning.

The reason most people stop after the first email is that crickets feel like rejection, and asking twice feels like begging. It’s neither. Your network forgets. They forget you’re available, forget what kind of work you actually want, forget you exist as a working option in their head. Reminding them once a month isn’t desperation. It’s hygiene.

And reach further back than feels reasonable. The single biggest unforced error in portfolio-life work is assuming the people who haven’t talked to you in a decade have nothing to offer. Statistically, they are more likely to introduce you to your next client than the colleague who sat next to you last quarter. The sociologist Mark Granovetter called this the strength of weak ties — your strong ties recirculate the network you already have; your weak ties open new ones.

The argument underneath it all

What Wallace argues — and what I came around to once I started seeing it from the inside — is that the case for a portfolio life is, fundamentally, a financial one.

The old advice to take the stable, full-time job at one company is actually one of the riskiest moves you can make. Concentration risk is concentration risk, whether the asset is your retirement account or your only paycheck. The portfolio life isn’t an aesthetic; it’s the diversification trade applied to your career.

For our generation, this is no longer the optional version. It’s the table-stakes one.

What no one mentions

Almost everything you do for the first 6 months (or even the first full year) of building this looks like wasted effort. Posts no one reads. Hours spent on the website and lead magnets. Coffee meetings that don’t go anywhere.

And then one day someone you’d barely thought about in two decades reaches out because they saw something you did months ago, and the entire calculus quietly flips.

We can’t afford to rely on one income stream and one employer to exist anymore. You can either start building the surface area now, or wait for someone else to decide that you’re not the configuration they need going forward. I’d rather be the person who built the option than the one who needed it.

If this resonated, I write here every couple of weeks about fractional work, career structure, and the financial-operator lens on building a company from seed through Series B.

I’m Tanya Puri, founder of PurInsights. Fractional operator (finance, ops, strategy) for seed-to-Series B founders prepping for their next round or exit. If you’re navigating one of these decisions — your own or someone on your team’s — you know where to find me.

Subscribe now

You can forget any economics lesson about a “free market”. Because this isn’t really an IPO - it’s a structural realignment of the global economy. Here is why the world’s biggest institutions effectively cannot let this fail.

1. The “Forced Buyer” Trap

At a $1.75Tn valuation, SpaceX won’t just be “another stock.” It will be an immediate pillar of the Nasdaq-100 and the S&P 500.

• The Math: Because it will likely hit “Fast-Track” index inclusion within weeks, every pension fund, 401k, and passive ETF on the planet will be structurally forced to buy it.

• The Result: Billions of dollars in “blind” capital will flow into $SPCX regardless of whether the company is losing money on AI or not. This creates a synthetic floor that protects early investors, at least in the short term.

  Leave a comment

2. The $75 Billion Underwriting Shield

SpaceX has t(r)apped the largest investment banks and investment managers for their IPO.

• The Fees: This is the biggest payday in the history of investment banking. The fees alone are enough to make a bank’s entire fiscal year.

• The Support: These banks have a massive vested interest in “price discovery” remaining stable. If the stock craters on day one, the IPO window for the rest of 2026 slams shut. They will use every legal stabilization mechanism in the book to keep the ticker green.

3. The 180-Day “Pressure Cooker”

While Musk has agreed to a 366-day lockup, the rest of the company is under a staggered release system.

• The Strategy: Unlike traditional IPOs where everyone sells at once (the “lockup cliff”), SpaceX is letting investors out in waves—20% here, 10% there—based on performance and time.

• The Game: This is designed to prevent a crash, but it also creates a high-stakes game of musical chairs. Institutional investors will be watching the “performance triggers” (like the stock staying 30% above IPO price) more closely than the actual rocket launches.

4. The VC “Recycling” Necessity

The venture capital world has been “cash poor” for years. Billions are locked up in aging private unicorns.

• The Liquidity Event: SpaceX is the great unlock. For firms like Valor Equity that invested in SpaceX back in 2008 and Founders Fund, this IPO is the liquidity they need to pay back their limited partners and start investing in the next generation of startups.

• The Stake: The entire startup ecosystem needs this to succeed to prove that the “Long Private” model actually works.

Subscribe now

The “So What” for You

Don’t look at the $SPCX price on Day 1 and think it represents “value.” It represents momentum and mechanics. We are moving into an era of “Systemic IPOs” where these companies are so large that their failure would be a macro-event.

• The Risk: When a company is “too big to fail,” it often stops being “too efficient to ignore.” Governance takes a backseat to growth, and the public bears the risk if the Mars-bound ambitions eventually hit a financial ceiling.

Bottom line: Wall Street has built a fortress around this IPO. You can bet on the fortress, but don’t forget that eventually, the people inside the walls (the early investors) are going to want to leave.

Hi, I’m Tanya, founder of PurInsights. If you like insights like this, you know where to find me—I’m your partner as you think through what’s happening in the market and what makes sense for your startup.

Subscribe now

Here is what you actually need to know.

1. It’s losing billions on a “Side Quest”

Forget the sleek launches. SpaceX is currently a massive cash incinerator for AI. While Starlink is a revenue beast ($11.4B), the company lost $4.3 billion in Q1 alone. Why? Because it merged with xAI and is now spending $1.25 billion a month just to rent compute power. You aren’t just buying a rocket company; you’re funding a late-entry AI war.

2. Compensation tied to Science Fiction

Elon’s new pay package doesn’t vest based on boring stuff like “dividends.” He gets his billions only if he hits “Mars-or-bust” milestones:

• Establishing a colony of 1 million people on Mars.

• Building non-Earth-based data centers.

• Hitting a $7.5 trillion market cap. It’s the most aggressive “all-or-nothing” bet in corporate history.

3. Total Control (85% of the Vote)

This isn’t a democracy. Thanks to supervoting Class B shares, Musk controls 85% of the voting power. He hand-picks the board, which is already packed with allies from Tesla and his VC circles. If you buy in, you are a passenger, not a co-pilot. You cannot fire him, and you cannot change the mission.

4. The “Incestuous” Musk Economy

SpaceX isn’t an island; it’s part of a circular economy. In 2025, SpaceX bought $131M of Cybertrucks and $506M of Tesla Megapacks. Tesla is mentioned 87 times in the filing. The “Musk-verse” is now so interconnected that if one pillar (Tesla) wobbles, the others feel the vibration.

The “So What” for the IPO Market

What this means for others:

• The AI Premium is Real: To get a trillion-dollar valuation today, you must be an AI company. SpaceX pivoted its narrative to AI to justify the math.

• Narrative beats Numbers: Huge losses are acceptable to Wall Street if your “moat” is literally outer space.

What this doesn’t mean:

• The Floodgates are Open: Don’t expect your average SaaS startup to follow. SpaceX is a “sovereign state” with its own rules. Most companies still need a path to profitability; SpaceX just needs a path to Mars.

Bottom line: If you buy into the SpaceX IPO, you’re not buying just any stock. You’re funding an epoch. Just make sure you’re comfortable with the burn rate.

Hi, I'm Tanya, founder of PurInsights. If you like insights like this, you know where to find me- I'm your partner as you think through what's happening in the market and what makes sense for your startup.

All this lead to uncovering that on one of their largest products, the company hadn’t touched its pricing in six years.

When the pricing got built

The pricing they had was the pricing they’d built when they were scrappy. Chasing first customers. Discounting hard to win design partners. Bending on terms to get logos on the wall. The kind of pricing where you’re not really pricing the product — you’re paying to be considered.

That made sense in year one. It even made sense in year two and three. But by year six, none of those conditions still applied. They had product-market fit. They had recurring revenue and reference customers and enough brand recognition to get meetings without three follow-up emails. They had a track record buyers could actually verify. But the pricing hadn’t moved. Every new deal was being signed at terms designed for a company that didn’t need to be that company anymore.

Why this was actually a growth problem

The obvious cost was the money they were leaving on the table — same, if not a more complex product, same or more effort to sell and same effort to deliver, but all set at prices six years ago. The unit economics were quietly degrading even as the business looked more impressive than ever from the outside.

The less obvious cost was the signal. Their pricing didn’t match what the rest of the company was advertising. The product was mature. The website was polished. The customer logos were impressive. And then the price came in low. That reads as uncertainty. A scrappy price on a not-scrappy product makes buyers wonder what the company isn’t telling them.

What drifts when pricing sits untouched

Here’s what I usually find when pricing hasn’t been revisited in a few years:

1. The cost base has shifted and nobody updated for it. You’re absorbing real fixed costs — onboarding, success, infrastructure — that the original pricing model never accounted for. Margin is leaking quietly while the topline keeps growing.

2. The value you’re delivering has grown, but the price hasn’t. The product is more mature. The integrations work. The success team is good. Customers are getting more, at the same price — and the gap between what they’re paying and what you’re delivering keeps widening.

3. Your earliest customers locked in something that’s quietly anchoring every renewal conversation. Even with new logos. The grandfather rate becomes the reference point, and your team starts negotiating against itself before the prospect has even opened the contract.

4. The competitive set you’re benchmarking against has changed. You think you’re competing with cheaper players. Your prospects have already moved you into a more premium bracket — and your pricing is still arguing against the old comp.

None of these are pricing problems in isolation. They’re the consequences of a company that grew up while its pricing didn’t.

Subscribe now

Every category leader you can name has done this

The companies you most admire didn’t pick a price on day one and hold it forever. They picked a price calibrated to win, and they raised it once they no longer needed to.

Netflix’s standard plan was $7.99 a month in 2011. By 2025 it was $17.99 — more than double. Across fourteen years they raised prices roughly every two to three years, every time as their content library grew and switching to a competitor got more painful. They stopped being the cheap alternative to cable because they didn’t need to be anymore.

Uber’s average fares rose about 92% between 2018 and 2021. The early years of cheap, heavily-subsidized rides were a customer acquisition strategy paid for by venture capital. Once they went public in 2019 and had to chase actual profitability, the discounting stopped and prices climbed to match what the rides actually cost to operate. The average service fee alone went from $0.29 in 2014 to $1.50 in 2022 — a 5x increase.

ChatGPT was free in late 2022 because OpenAI needed mass adoption to lock in the category. By December 2024 they’d added a $200/month Pro tier for power users. Anthropic followed in April 2025 with $100 and $200/month Max plans on top of their $20 Pro. Last month — April 2026 — OpenAI slotted in its own $100 tier to fill the gap between Plus and Pro. Both companies are doing the same thing your scrappier clients should be doing: segmenting buyers by willingness to pay, and charging the premium ones what they’re actually worth.

None of these companies were raising prices on existing customers in a vacuum. They were re-pricing to match a different reality. The product had changed. The competitive position had changed. The cost of building, serving, and supporting the product had changed. And the customer base had segmented in ways the original pricing never anticipated.

The pattern is consistent: start cheap to win the market, raise prices once you’ve built the position that lets you.

Leave a comment

What this looks like from the other side of the table

This is also one of the first things a thoughtful Series B investor pokes at in diligence. Pricing that hasn’t moved is a tell. It says either the company doesn’t know what it’s worth, or it knows and is afraid to test the answer. Neither is the story a founder wants told for them on a diligence call.

Going into a raise with a pricing model that hasn’t been pressure-tested in years means showing up with an unflattering data point already on the table — and giving the investor the framing of the conversation before you’ve had a chance to set it yourself.

The fix is usually “raise prices”

When pricing has sat for years while everything else moved, the right move is almost always to raise prices — sometimes meaningfully — and to do it in a way that matches how the company actually delivers value now: by segment, by deal shape, by what’s being delivered versus what was in the original proposal.

That feels uncomfortable for founders who came up scrappy. The original pricing reflected how the company needed to behave when it was new — hungry, flexible, willing to discount to win the room. Raising prices means accepting that the company doesn’t need to behave that way anymore.

Most of them don’t. Most have already outgrown it everywhere except in their pricing.

If you’ve raised in the last twelve months and haven’t revisited pricing since: what version of your company is your pricing still apologizing for?

I’m Tanya, founder of PurInsights. I help post-raise founders figure out where to deploy capital when the growth path isn’t obvious — and find the risks hiding in the plan before the board does. If your pricing is still calibrated for a company you no longer are, you know where to find me.

It’s been published, republished, repackaged, and templated into every deck builder on the internet. Every founder I work with has seen it. Every one of them has tried to build to it.

For most companies, the template works because it follows the order an investor’s brain naturally moves through when evaluating a venture deal. They want to know there’s a problem before they evaluate the solution. They want to see the market shape before they price it. They want to meet the team after they’ve already decided the company is interesting, so the bet feels less like “I’m betting on these specific people” and more like “I’m betting on a real opportunity in capable hands.”

If your company fits a category an investor already understands — a SaaS tool for a known persona, a consumer app for a known behavior, a fintech for a known financial workflow — the template works on autopilot. Don’t fight it.

But that’s not every company.

The case where the template breaks

I’ve been working with a founder building deeply technical infrastructure in the medical device supply chain. The product is genuinely novel, but a category that no one has a mental model for. There’s no comp slide that works for them. There’s no investor in the room who walks in already understanding what the team has built.

The first deck we put together followed the standard order. Problem on slide 3. Solution on slide 4. It landed flat in every mock pitch they ran.

Investors were pushing back because they could not picture what was being sold. By slide 5, they were trying to reverse-engineer the solution slide. By slide 7, they had quietly given up and were waiting for the conversation to end.

I worked with the founder to tear it apart and rebuilt it.

Subscribe now

The new order

— Slide 2: what the product is, in plain English, with one diagram

— Slide 3: how a typical medical device parts order works today, end to end, using a Legos analogy any non-specialist could follow

— Slide 4: the team — every person with years inside the giants this product is built to replace — Then problem, opportunity, market

The Legos frame is worth pausing on. The team realized that ordering a medical device part isn’t ordering a single object — it’s specifying a build. Components with their own suppliers, lead times, and compatibility rules, all of which have to click together for the final assembly to work. Once investors had that picture in their heads, the rest of the slide did its job. Nobody needed a glossary. Nobody needed to be a medical device specialist to feel the coordination problem in their gut. The analogy did weeks of education in thirty seconds.

The shift wasn’t just cosmetic. It re-sequenced the cognitive load.

Here’s what I mean. Every slide in a pitch is a request the founder is making of the investor’s working memory. “Hold this number.” “Trust this framing.” “Believe this team can win.” The standard order assumes the investor can hold all of those simultaneously, because the underlying category isn’t asking for novel comprehension. The deck is rearranging known pieces.

When the product itself is the novel comprehension — when the investor has to first build the mental object before they can do any of the evaluating work — the standard order asks them to evaluate something they don’t yet have. They get stuck. They stop listening.

The team slide moves up for a related but distinct reason. After the investor builds the mental object, the next question is: can anyone actually do this? In a novel category, “the right team” is doing more work than usual — it’s the substitute for the comp slide that doesn’t exist. Putting it after the product explanation, before the market section, gives the investor permission to spend the rest of the deck being curious rather than skeptical.

Three things for anyone building something complex

1. Templates are a starting point, not a script. If your category is unfamiliar, your audience needs comprehension before they can evaluate opportunity. Selling someone on a market they don’t yet understand the shape of is wasted ink.

2. Analogies are deck work, not nice-to-haves. The five minutes you spend finding the right “it’s like X, but for Y” framing saves the next ten slides. If you can’t construct one, that’s a signal — about the deck or about the pitch.

3. Team slides land differently depending on placement. At the end, they’re a credibility check. Up front — right after the product explanation — they’re the proof that the people building this know exactly why the incumbents can’t. Same slide. Completely different job.

The work isn’t the template. It’s the judgment about the template.

Anyone can hand you the standard order. There are free templates on the internet, AI tools that will generate one, and any number of well-meaning advisors who’ll tell you to follow it. None of those things will tell you when it’s wrong for you. That’s a separate skill, and it comes from sitting inside your specific company long enough to understand what an outside investor will actually struggle to hold.

That’s the difference between a consultant who runs your process and a partner who knows your company. The template is the floor. The judgment is the work.

Share

I’m Tanya, founder of PurInsights. Finance and ops support for seed-to-Series B founders prepping for their next round or exit. If this is the kind of thinking you want in your corner before you raise, you know where to find me.

Here’s the version of the story almost nobody is connecting.

What Kalshi is

Kalshi is a federally-regulated US prediction market where users buy and sell yes/no contracts on the outcome of real-world events — elections, Fed rate decisions, GDP prints, sports games — and the live prices function as crowd-sourced probability forecasts. Last year, combined Kalshi/Polymarket volume hit $50B, up from $16B the year before. Kalshi alone has gone from $52B in annualized trading volume to $178B in six months.

Two years of US regulators trying to shut it down

In 2023 the CFTC moved to kill Kalshi’s election contracts as illegal gaming — a federal court overruled it in September 2024, and the CFTC dropped its appeal under the new administration in May 2025. State regulators then picked up the fight: New Jersey, Nevada, New York and others issued cease-and-desists through 2025, with Nevada reinstating an injunction last November. The state-level battle is still ongoing.

Meanwhile, in that same window, Kalshi raised $185M at a $2B valuation in June 2025. By October it was $5B. By December it was $11B. By March 2026 it was $22B — an 11x markup in nine months.

Leave a comment

The Fed flip

In May 2026, The Economist reported on a Federal Reserve research paper by Anthony Diercks and colleagues. The conclusion: Kalshi’s prediction data outperforms conventional financial instruments — like futures and options contracts — for forecasting Fed interest-rate decisions. It matches Bloomberg’s professional analyst consensus on unemployment and core inflation. It beats Bloomberg on headline inflation.

The incoming Fed chair, Kevin Warsh, has gone on record calling traditional economic indicators “evidence of false precision and analytic complacency.” The Diercks team is — “subject to approval” — building a website to make Kalshi data easier for the Fed to consume.

These two storylines are not causally connected. State regulators don’t care that the Fed wants Kalshi’s data. The CFTC dropped its appeal because of administration change, not because the data was good. They’re running in parallel — and that’s exactly the point.

Subscribe now

There is no monolithic “government”

The most useful read of the Kalshi story isn’t “the system bends toward what works, eventually.” Kalshi is still in active state litigation right now. The Fed’s interest hasn’t made any of those lawsuits go away.

What’s actually going on is messier and more useful for founders to understand: there is no monolithic “government.” There are dozens of agencies with non-coordinating mandates. At any given moment, one of them may be trying to shut you down while another wants to consume your output. That’s not a contradiction — that’s the structure.

For founders building anything the existing framework wasn’t designed for, the strategic question isn’t “will the system come around?” It’s:

Which part of the system needs me, and can I survive the part that doesn’t?

Takeaway for founders

Distribution and data are still the moat. Kalshi won not because regulators changed their minds, but because the data became too good to ignore. The CFTC’s “this is illegal gaming” argument collapsed the moment another arm of the federal government had a use for the output. If you’re building something the existing framework wasn’t designed for, the question isn’t whether you’ll get cease-and-desists. It’s whether your output becomes indispensable before they crush you.

What to watch

The Nevada injunction. The next state-level court decisions. Whether Polymarket’s CFTC-approved US relaunch broadens or narrows Kalshi’s competitive position. And — most importantly for founders building in regulatorily-ambiguous spaces — which arms of the government start adopting prediction-market data as a normal input. When the Bureau of Labor Statistics or the Census Bureau starts citing Kalshi, that’s the real legitimization moment.

Until then, the parallel-storyline pattern is the model: navigate the part of the system fighting you while you make yourself indispensable to the part that isn’t.

What’s a piece of “non-traditional” data you’ve started trusting more than the official version? Reply and tell me — I’m building a list of the most useful unofficial data signals founders are watching in 2026.

I’m Tanya, founder of PurInsights. Fractional operator (finance, ops, strategy) for seed-to-Series B founders prepping for their next round or exit. If you’re building outside the existing framework and need help mapping which part of the system actually matters for your raise — you know where to find me.

Pix processed roughly 7 billion transactions per month in early 2026, used by 93% of Brazilian adults. As a reference point, its US counterpart, FedNow, processed 2.7 million transactions for all of Q1 2026, with an average transaction size of $99,000.

Same underlying capability on paper. Wildly different reality.

What Pix actually is

Mechanically: money moves from any bank account to any other bank account in Brazil, in under 10 seconds, 24/7/365, free for individuals. Merchants pay an average 0.22% — about one-tenth what cards charge. You send by scanning a QR code or typing a “Pix key” — phone number, email, tax ID. Money lands before you put your phone away.

What runs on it? Splitting dinner. Paying rent. Buying coffee. E-commerce checkout. B2B vendor invoices. Payroll. Gig payouts. Government benefits. The entire money-moving stack of a country, on one rail.

The numbers tell the story. Pix processed 64 billion transactions in 2024, surpassing combined card volumes by 80%. By early 2026 it was approaching 8 billion transactions per month. Cash withdrawals are down 40% from their peak — Pix didn’t just take share from cards, it took share from cash too. Brazil’s GDP has beaten expectations three years running, and the BCB credits Pix-driven productivity gains as part of why, though the causal claim is hard to prove cleanly.

The card networks lost the consumer-to-merchant flow in roughly 24 months. That’s faster than any payments shift in modern memory. And it happened in a country with the same incumbents we have (Visa, Mastercard, a powerful local banking lobby). What Brazil had that we don’t was a central bank willing to mandate a rail.

Pix didn’t win because Brazilians wanted faster payments

Pix won because Banco Central did three things at once on launch day in November 2020: made it free for individuals, mandatory for any bank with 500K+ accounts, and 24/7 from day one. The combination collapsed the cold-start problem that every private real-time payment network has run into.

By the end of 2021, Pix had over 100 million users. The downstream effects — Stone’s working-capital products, PicPay’s gig-payouts business, an entire generation of Brazilian fintechs — are all built on top of one piece of public infrastructure. The cleanest illustration: Nubank had 17 million Brazilian customers in 2019. By 2024 it had 102 million, plus another 12 million in Mexico and Colombia, making it the largest digital bank in the world outside China. That trajectory does not happen without Pix.

And the rail keeps compounding. The BCB is now building Drex — a central bank digital currency — directly on top of Pix’s user database. Recurring payments and contactless via smartphone rolled out in 2025. Each of these would have been a multi-billion-dollar standalone fintech in the US. In Brazil they are incremental updates to the public rail.

None of this is free, in the sense that matters most. Pix concentrates a remarkable amount of power in the Banco Central — every transaction in Brazil flows through encrypted databases the central bank alone controls. As one fintech executive put it in The Economist: “Now we live in a democracy, but imagine if this existed under an autocracy and all your information was available to the government.” That’s a real concern, and one the US version of this story should be designed around. Public infrastructure is also concentrated infrastructure.

So why doesn’t the US have this?

The US has the rail. FedNow does instant settlement, 24/7, at 4.5¢ per transaction. The capability is not the problem. The political economy is. Three reasons:

1. No mandate. Banco Central required every Brazilian bank with 500K+ accounts to support Pix from launch day. The Fed cannot legally do that — FedNow is opt-in, like check clearing. The result: 2.5 years in, FedNow reaches roughly 40% of US deposit accounts, and most participating banks are still receive-only — they can take a payment but can’t initiate one. Even when your bank is on FedNow, you probably can’t push money out via FedNow. That’s why the average transaction is $99K (treasurers between the few banks with send capability), not $20 (coffee). It also doesn’t help that the US has two competing real-time rails (FedNow + RTP, the latter owned by the big banks themselves) splitting attention across ~10,000 institutions. Brazil mandated five megabanks. The equivalent US mandate would touch thousands.

2. Banks have direct financial incentive NOT to push it. US banks earn tens of billions per year on overdraft fees, wire fees, and float on settlement delays. FedNow at 4.5¢ wipes most of that out. Banks aren’t going to migrate customers off rails they make money on, and the Fed can’t make them. The banks moving fastest on FedNow are smaller institutions and credit unions where the fee revenue at risk is low. The big banks — the ones whose participation would actually shift the curve — have been the slowest to enable send.

3. No consumer brand. Pix has “Pix accepted here” QR codes in every Brazilian coffee shop. Brazilians ask merchants if they take Pix and complain when their bank’s Pix experience is bad. FedNow is invisible — even when your bank uses it, the app says “instant transfer,” not “FedNow.” The Fed doesn’t market to consumers; it markets to banks. No brand, no consumer pull, no pressure on banks to enable send or merchants to accept.

Subscribe now

What could actually work?

None of these problems are unsolvable. They just require somebody other than the obvious player to break the logjam. Four paths:

Read more

Take a second on that.

In a country where the median income is roughly 1/15th of the U.S., last-mile logistics have compressed to a point that the U.S. healthcare system, with all its capital and infrastructure, cannot touch. We measure ER throughput in hours. They are racing the clock in minutes.

If your first reaction was “regulatory arbitrage” or “cheap labor,” you read it the same way most of the US reads it. And you’re missing the actual story.

The misreading

The lazy take on India’s quick-commerce miracle is that the unit economics work because rider wages are cheap. That’s part of it. But it’s a small part, and it doesn’t explain why the same companies have figured out how to compress healthcare — a category where the gating factor isn’t logistics, it’s clinical handoff, payment authorization, and regulatory routing.

The real story is harder, and more interesting: BlinkIt didn’t build a faster delivery company. It built an operating system for the last mile.

Want the cleanest proof this isn't a labor-cost story? Amazon couldn't compete in India with its existing fulfillment network. It rolled out Amazon NOW from scratch as a separate product to play in 10-minute delivery. The most logistically sophisticated company on earth needed a new operating system, not a faster version of the one it already had. If Amazon's 1-2 day stack can't be retrofitted into a 10-minute stack, neither can anyone else's. That tells you everything about what this actually is.

That question for U.S. founders isn’t “can we copy 10-minute delivery here.” It’s “what does the U.S. last mile look like when somebody finally builds the operating system underneath it.”

Three things I would steal.

Subscribe now

1. Stop optimizing the order. Optimize the network.

The dirty secret of quick-commerce is that you do not get to 10 minutes by improving the rider’s route. You get there by pre-positioning inventory within ~2 km of the customer. BlinkIt operates dark stores — small, hyper-local fulfillment centers — and stocks them by predicted demand at a per-postcode level.

The unit of design isn’t the delivery. It is the geography.

That sounds obvious until you realize how few U.S. healthcare systems think this way. Hospital networks expand by acquiring hospitals. Pharmacy chains expand by acquiring pharmacies. Almost nobody is asking: where would I place 80 micro-clinics, each holding 200 SKUs, if my goal were to be within 10 minutes of 90% of demand?

This is the BlinkIt question for U.S. healthcare. Whoever asks it first wins.

2. The “instant” gap in healthcare is bigger than it ever was in commerce.

Quick-commerce in groceries was a real but bounded innovation — it shaved minutes off something that already happened in hours. Healthcare is different. The clinical reality is that a 90-minute window for stroke, a 60-minute window for sepsis, and a same-day window for antibiotic-resistant infections create a category of demand where time isn’t a feature. It’s the entire product.

The U.S. doesn’t have a 10-minute ambulance problem because it lacks ambulances. We have one of the densest EMS networks on earth. The bottleneck is upstream — dispatch is fragmented across 911, hospital systems, and private providers; clinical routing is decided by paper-thin handoff protocols; and payment authorization happens after the patient is already in a bed.

The first U.S. operator who treats a stroke window the way BlinkIt treats a chai order has a category-defining business. That isn’t hyperbole. It’s the same arbitrage Uber ran on yellow cabs, applied to a market 30x the size.

3. Speed is the wedge. Trust is the moat.

Here’s what most analyses of BlinkIt and Apollo miss: the companies that won India’s quick-commerce wars weren’t the fastest. They were the ones whose customers didn’t get burned. Wrong order, missing dose, expired produce — these things kill quick-commerce companies far faster than slow delivery ever did.

In healthcare, that bar is 100x higher. A wrong dose is a malpractice case. A late ambulance is a death.

This is exactly where U.S. founders have an edge that India’s incumbents don’t, and which gets buried under “they’re so far ahead.” We have stronger clinical infrastructure, better data systems, and (mostly) more rigorous safety culture. The U.S. founder who layers BlinkIt’s network architecture on top of U.S.-grade clinical trust isn’t copying India. They’re building something India can’t.

Upgrade for exclusive content

A fintech sidebar — because of course

A small note for my fintech readers, because this is the part nobody talks about. Real-time fulfillment forces real-time payments. Quick-commerce in India runs on UPI, which clears in seconds and costs nothing. U.S. healthcare runs on prior authorization, claim adjudication, and a payments stack that takes 30 to 90 days to settle.

You cannot build a 10-minute prescription business in the U.S. on top of a 30-day payments rail.

Whoever wins the U.S. version of this story will either bypass insurance — cash-pay, subscription, employer-direct — or rebuild the auth layer to run synchronously. There is a fintech opportunity hiding inside every healthtech opportunity here, and the operators who see it as one stack instead of two are the ones to bet on.

The thesis

The 10-minute economy isn’t a delivery story. It is an operating-system story. The companies that figure this out won’t look like DoorDash-for-X. They’ll look like infrastructure — boring, dense, hard to dislodge, and in five years, impossible to imagine the world without.

The U.S. category I’d compress next isn’t pharmacy or food. It’s diagnostics. Lab work, imaging, urgent care triage — these are the businesses where time, trust, and payments converge in exactly the way India’s quick-commerce winners have already solved.

What’s the category you’d compress?

I'm Tanya, founder of PurInsights. Fractional operator (finance, ops, strategy) for seed-to-Series B founders prepping for their next round or exit. If this is the kind of thinking you want in your corner before you raise, you know where to find me.

It starts with something that sounds much more ordinary. “I haven’t had a strategic conversation in weeks.” Or “I just don’t have time to focus on detailed analyses” or “We’re about to raise and I’m not sure our numbers tell the right story.” Or my personal favorite: “Things are working, but everything feels harder than it should.”

These are not complaints. These are symptoms. And by the time a founder is saying them out loud, the underlying problem has usually been building for months.

I’ve written a lot in this section about what fractional operators do - the systems we build, the bottlenecks we fix, the infrastructure we install. But I haven’t written about something equally important: how to recognize that you need one. Not in theory. In practice. In the actual texture of your week.

So here’s what I’ve learned. The signal is almost never dramatic. It’s not a crisis. It’s a pattern.

Signal 1: You’re the Last Mile on Everything

You hired smart people. You trust them. And yet somehow, nothing ships without you touching it.

The financial close that NEEDS your review. The partnership contract sits in your inbox for three days because you’re the only one who knows the context. You’re not micromanaging - you’re the only person with enough cross-functional visibility to make the final call.

This is the founder bottleneck, and I’ve written about it before. But the timing signal isn’t the bottleneck itself. It’s when you notice that the bottleneck has started to cost you opportunities. A deal slipped because the contract took too long. A hire fell through because the offer approval sat for a week. A board member started asking questions you didn’t have time to prepare answers for.

When being the last mile starts costing you the things that only you can do — fundraising, customer relationships, product vision — that’s the signal.

Signal 2: Your Numbers Are Technically Correct but Strategically Useless

You have a financial model. It might even be a good one. Your bookkeeper closes the books monthly. You have dashboards.

But when your lead investor asks “what are your unit economics at scale?” you hedge. When someone on your team asks “can we afford this hire?” you say “errr yes because revenue is growing??” and then spend an hour reverse-engineering an answer from three different spreadsheets. When you sit down to plan the next quarter, you realize the model you raised on doesn’t reflect how the business actually works anymore.

The numbers aren’t wrong. They’re just not useful. They don’t connect to the decisions you’re making every week. And the gap between what the data says and what you actually know about your business is widening, slowly, in a way that’s hard to see until someone asks the question you can’t answer cleanly.

I rebuilt a founder’s entire financial model last year in about a week. Her reaction was: “I didn’t realize how much I was guessing.” She wasn’t guessing because she’s not sharp. She was guessing because the model had frozen in time while the business kept evolving. That drift is invisible until it’s not.

Signal 3: You Keep Saying “We’ll Get to That Next Quarter”

There’s a list. Every founder has it. The things that aren’t on fire but aren’t getting done either.

Fix the reporting infrastructure. Document the processes that live in your head. Build a real onboarding system. Set up a hiring plan that connects to the financial model. Clean up the cap table. Figure out the compliance calendar before someone asks about it.

These aren’t urgent. They’re just important. And because they’re not urgent, they keep getting pushed. Next month. Next quarter. After the raise. After the launch.

The problem is that “important but not urgent” has a shelf life. At some point, the reporting infrastructure you didn’t fix becomes the reason your board deck takes a week. The processes you didn’t document become the reason your new hire is still asking you questions two months in. The compliance calendar you never built becomes the thing a partner finds in diligence.

When you realize that the list of “next quarter” items has been the same list for three quarters running, that’s not a prioritization issue. That’s a capacity issue. And it’s the clearest signal I know that the company has outgrown the founder’s bandwidth.

Leave a comment

Signal 4: You’re Making Strategic Decisions Without a Framework

Should you expand into a new market or deepen the current one? Should you build in-house or partner? Should you take that strategic investment even though the terms aren’t ideal? Should you hire a VP of Sales or double down on the channel that’s already working?

Every founder faces these decisions. The question isn’t whether you’re making them — it’s how.

If the honest answer is “intuition plus whatever data I can pull together in an afternoon,” you’re not alone. Most founders at this stage operate the same way. And a lot of the time, the intuition is good.

But the process isn’t structured. The financial implications aren’t modeled. The second-order effects aren’t mapped. And when the board asks “why did you make that call?” three months later, the reasoning isn’t documented.

This one is subtle because the decisions might be right. The signal isn’t bad outcomes — it’s the feeling that every big decision takes more energy than it should, that you’re carrying the weight of choices you can’t fully defend, and that you don’t have a thinking partner who can pressure-test your instincts with data.

The Pattern

If you noticed, these signals share something: none of them feel like emergencies. They feel like friction. Like things are just a little harder than they should be, a little slower, a little less clear.

That’s why founders wait. The company isn’t broken. It’s just running on the founder’s nervous system instead of on systems. And the founder’s nervous system is remarkably resilient — right up until the moment it isn’t.

The founders I work with almost always say some version of the same thing in our first real conversation: “I should have done this six months ago.” Not because anything catastrophic happened in those six months. But because six months of compounding friction — slow decisions, fuzzy numbers, deferred infrastructure, strategic calls made without frameworks — adds up to a company that’s harder to run, harder to raise for, and harder to exit than it needs to be.

Subscribe now

The Question to Ask Yourself

If you’re reading this and nodding, here’s the diagnostic I’d run:

Can you go on vacation for a week without everything stalling?

Not “will the company survive” — it will. But will decisions get made? Will the numbers be accurate? Will the team know what to prioritize without asking you?

If the answer is no, the company isn’t ready to scale. It might be ready to grow — those are different things. Growth is adding revenue. Scaling is building the infrastructure that lets growth compound without breaking everything underneath it.

And if the answer is “I don’t know” — that’s the same as no.

The fix isn’t dramatic. It’s not a reorg. It’s not a full-time executive hire you can’t afford yet. It’s the foundational work — the financial model that reflects reality, the operating rhythm that creates alignment, the decision frameworks that let you move faster with more confidence, the documentation that gets the business out of your head and into a system.

That’s the work I do. Through PurInsights, I’ve taken companies from messy books to exit-ready: closing financials for lender and acquirer diligence, rebuilding margin analysis that reshaped the profitability story, securing lines of credit, and building the infrastructure founders actually raise on.

If any of these signals sound like your week, let’s talk. A 30-minute conversation is usually enough to know whether there’s a fit.

Subscribe if you haven’t. I write about what it actually takes to build and scale in complex industries — no templates, no theory, just the real work.

I want to talk about what that work actually looks like — because I think the invisibility of it is part of why founders underestimate it, misprice it, and sometimes don’t hire for it until they’re already in pain.

The Company at Day Zero

Here’s what I typically walk into. The founder is smart, the product works, there’s real traction. And yet everything feels harder than it should.

Board decks take a week to assemble because data lives in six different places and nobody agrees on the numbers. The financial model was built for the last fundraise and hasn’t been touched since. Compliance is getting done — sort of — but it’s running on institutional memory and a shared Google Doc that three people update inconsistently. The CEO is spending a third of their time on things that aren’t building product or talking to customers, and they know it, but they can’t figure out how to stop.

The company isn’t broken. It’s just running on the founder’s nervous system instead of on systems. And that works until it doesn’t.

What Changes by Day 30

The first month is mostly about building the infrastructure that should have existed before the company started scaling. It’s not glamorous. Nobody writes a case study about it.

The financial model starts telling the truth. Not just “the numbers are correct” — most founders’ numbers are correct. I mean the model actually reflects how the business works. The unit economics connect to the operating plan. The assumptions are documented and defensible. When someone asks “what happens if we lose that partner?” or “what does 3x volume look like?”, there’s an answer that doesn’t require the founder to build a new spreadsheet from scratch. This matters more than founders realize. I’ve watched companies nearly stall a fundraise because the financial model told a different story than the pitch deck. Not because anyone was lying — because nobody had forced the two to talk to each other.

There’s a weekly rhythm. A real one. Not a standing meeting that everyone half-attends. I mean a cadence where the team knows what they’re accountable for this week, what’s on track, and what’s not. The founder stops being the router for every question and every decision. The 11pm Slack fires don’t disappear entirely — it’s still a startup — but they become the exception rather than the operating model.

The compliance mess gets a structure. In regulated or operationally complex businesses, compliance isn’t a side task. It’s infrastructure. By day 30, the question shifts from “are we doing this?” to “here’s the system for how we do this, here’s who owns it, here’s how we know it’s working.” This is one of those things that’s completely invisible from the outside but transforms how confidently the company can operate, partner, and fundraise.

Leave a comment

What Changes by Day 60

This is where the work starts compounding.

The founder gets time back — and actually uses it. Not “I theoretically have more bandwidth.” I mean the CEO is back in customer conversations. Back in product. Back doing the things only they can do, because the operational and financial infrastructure is no longer running through them as the single point of failure. One founder I worked with told me she realized she hadn’t had a real product conversation in six weeks before we started working together. Six weeks. She was spending all her time on board prep, partner management, and operational firefighting. By day 60, that ratio had flipped.

The board and investor conversations change tone. This is subtle but powerful. When a founder shows up to a board meeting with clean numbers, a clear operating cadence, and a financial narrative that connects to the strategic thesis — the whole energy shifts. The board stops asking “do you know what’s going on?” and starts asking “how can we help you go faster?” That shift doesn’t happen because of a single deliverable. It happens because the underlying infrastructure is solid enough that the story tells itself.

Decisions get faster and better. Should we invest more in this partnership or diversify? Can we afford to hire for this role now or do we wait? Is this product line worth the operational complexity? These questions used to sit in the founder’s head for weeks because evaluating them required pulling data from three systems, talking to four people, and building a model from scratch. Now there’s a framework. Not a rigid process — a framework. The information exists, it’s organized, and there’s someone who can pressure-test the options alongside the founder.

What Changes by Day 90

Here’s the thing about day 90: the most important change is what doesn’t happen.

The founder goes on vacation and the company doesn’t stall. I realize that sounds small. It’s not. At most early-stage companies, if the CEO disappears for a week, everything slows down. Decisions wait. Information gets stuck. People work around the gap rather than through it. By day 90, if we’ve done our job, the operating rhythm holds. The reporting happens. The decisions that can be made without the founder get made. The ones that can’t are queued cleanly, not lost in a Slack thread.

The company is fundraise-ready — not in a “we polished the deck” way, but in a “we can withstand diligence” way. The model is tight. The operating plan connects to the financials. The compliance infrastructure is documented and defensible. The founder can articulate the story clearly because it’s grounded in a business that’s actually organized. I’ve seen too many companies rush fundraise prep in the final weeks before going out. That’s cosmetic. What we’re building is structural.

The founder’s role has actually changed. This is the one that matters most and gets talked about least. At day zero, the founder was the integrator — the person who held all the context, made all the connections, bridged every gap. By day 90, they’re the leader. They’re making strategic decisions instead of operational ones. They’re spending their time on the highest-leverage activities instead of the most urgent ones. That shift doesn’t show up in any metric. But it’s the whole point.

Subscribe now

Why This Matters

I write about this because I think the market for fractional operators is still largely misunderstood. Founders think they’re buying a deliverable — a financial model, an operating plan, a board deck. And yes, those things get built. But the real value is the transformation underneath: a business that runs on systems instead of on the founder’s memory, relationships instead of heroics, and infrastructure instead of improvisation.

That transformation is invisible to everyone except the people inside the company. And it’s worth more than any single deliverable.

If you’re a founder who’s feeling the weight of being the operating system for your entire company — and you’re starting to suspect there’s a better way — I’d love to talk. Book a free 30-minute strategy call here.

If this resonated, share it with a founder who needs to hear it. And subscribe if you haven’t — I write about the operational and strategic realities of building in complex industries.

This is the insurance version of a conversation I’ve been having in fintech for months. Same lens, different industry, same answer: in regulated, capital-intensive businesses, structure is strategy.

I’ve spent the last several weeks going back through the scoreboard of the 2020–2021 insurtech IPO class. I wanted to understand why some of these companies are clawing back and some aren’t — and whether the pattern matches what I’ve been seeing in fintech.

It does. And honestly, it’s more stark here.

The companies on this scoreboard launched into the same wave of enthusiasm. Same rhetoric about legacy insurers being disrupted. Same pitch about loss ratios improving with AI, data, and digital-native distribution. Same access to capital when markets rewarded growth over discipline.

What’s different now is what’s underneath them. And that structure — not the brand, not the app, not the NPS score — is determining who’s being taken seriously for the next cycle and who’s fighting for survival.

The Scoreboard

Lemonade: The comeback the market didn’t expect.

Lemonade IPO’d at $29 in July 2020, ran to an all-time high above $180 in early 2021, and then collapsed — trading below $15 for most of 2023 and 2024. By early 2026, it’s clawed back into the $30s. Revenue crossed $600 million for the trailing twelve months. The gross loss ratio — the metric that matters more than almost any other in this business — finally dropped below 75% for the first time in its history.

What changed? Not any claims made around AI marketing but in reality - the book.

Lemonade deliberately slowed new home and auto growth, sunset unprofitable geographies, and forced the mix toward products where it actually had underwriting data advantage. It also stopped treating reinsurance as a cost line and started treating it as a strategic lever — ceding more premium in volatile lines, holding more in stable ones. Ops got boring. Boring worked.

The lesson isn’t that Lemonade figured out AI. It’s that Lemonade finally built a book.

Root: Still fighting gravity.

Root IPO’d at $27 in October 2020 and has traded below $15 for most of the last three years. The core auto insurance business is structurally difficult — one of the most competitive personal lines in the industry, with entrenched incumbents who have decades of telematics data and capital advantages Root can’t match.

Read more