This assumes the argument of Raising for Consumer Brands 101: you’ve proven the equation, the working capital cycle is navigable, and you’ve decided multiplication is worth permanent equity. What follows is the process — which runs through a narrower investor pool than software, requires heavier diligence, and takes longer than most founders expect — shaped by a market that is nothing like the one the pitch guides describe.

The market you’re entering.

DTC venture investment fell roughly 97 percent between 2021 and 20231 — the clearest measure of a broader contraction in consumer capital. The funds that defined the boom are gone or quiet; the ones still writing checks know each other, share deal flow, and have seen the same pitch about category disruption several hundred times. A software seed runs a wide process — forty investors, narrative momentum, manufactured urgency — because there are thousands of relevant funds. A consumer raise runs a narrow one, because there is a finite, heavily overlapping pool of relevant funds for your stage and category, they have long memories, and burning the pool by approaching them before you’re ready is a recoverable problem only in theory.

The practical consequence: a consumer raise is closer to a credit process than to the narrative sprint the guides describe. Investors already know your category, have probably seen your velocity data in syndicated reports, and will call your retail buyers. The founder who treats it like a story competition loses to the one who knows their numbers cold.

The timeline, and retail’s calendar.

A consumer raise takes longer than a software raise because diligence is heavier. Investors verify velocity claims against syndicated data, call retail buyers, inspect co-packer relationships, and audit deduction history. Six to nine months from first meeting to money in the bank is common; plan accordingly.

The calendar implication: raise ahead of your big operational moments, not into them. The largest retailers run category reviews that determine shelf placement for the following year — most major grocery chains review categories annually, with the buying decisions happening months before product appears on shelf. A brand in the middle of diligence when its most important category review lands is either distracted during the review or delayed during the raise, and usually both.

The raise timeline should also account for what the money needs to do. If the raise funds an inventory build for a major retail launch, the money needs to land before the launch, not with it. Working backward from the operational need is how to set the fundraising calendar; most founders set it from the desire to raise and arrive at the operational moment undercapitalized.

What the data room actually contains.

The software guides say pitch deck, model, cap table. All three apply here too, and none of them is what consumer investors open first.

For a retail-led brand, what they open first is the velocity spreadsheet. Units sold per store per week, by account, by month, for the last twelve months. Not an average — the distribution, so they can see which accounts are working, which are declining, and whether growth is coming from new doors or from selling more in the ones you have. Distribution without velocity is the central trap in retail CPG; a fund that has seen it before will find it in the data before the founder mentions it. For a DTC-led brand, the analogous document is the cohort file: acquisition cost, contribution margin, repeat behavior, and payback by acquisition month. The question is the same in both cases — is there evidence of durable consumer demand? — but the spreadsheet that answers it is different.

The deduction line is the one most founders hide and most investors find. Trade spend, promotional allowances, slotting fees, chargebacks, damaged goods — the gap between gross revenue and net revenue, shown honestly, is what separates a founder who understands their business from one who understands their pitch.

Who to approach, in what order.

Consumer funds are not interchangeable. A DTC brand, a retail-velocity CPG product, and an omnichannel business with wholesale revenue attract different pools; targeting the wrong one produces polite passes and costs you months. The fastest way to build the right list is to start from the portfolio backwards: find the last ten deals in your category and stage, identify who led them, and note who co-invested. That concentration map is your target list.

Rough sequencing: angels and operator investors first, because they become reference checks when the funds call. Funds next, in the order you’d most want to work with them — move fast enough that you’re in multiple conversations simultaneously, slow enough that you’re not asking for a meeting before the data room exists. Strategics last, or in parallel if a distribution relationship is the point; strategic capital is patient and its decision process runs on a corporate calendar, not a fund one.

The warm path matters enormously in consumer because the pool is smaller and the network is denser. A founding partner at CAVU, Forerunner, or VMG gets far more cold approaches than they fund; most go nowhere. An intro from a portfolio founder they backed gets a meeting. The way to build that intro is to have been a useful operator before you needed it — which is the argument for starting conversations six to twelve months before you raise rather than the week you decide to. If you are six months out, it starts now — not with an introduction request, but with a useful conversation.

The meetings themselves.

Consumer investors in a first meeting are testing one thing before anything else: whether the founder knows their velocity numbers without looking at slides. The founder who can say “we’re doing 4.2 units per store per week at Sprouts, 3.1 at Whole Foods, and the Natural Grocers doors we added in Q3 are underperforming at 1.8 and we know why” is a different conversation than the one who opens the deck to find the slide.

Everything else in the meeting follows from that. Category thesis, competitive positioning, brand — investors in this space have formed views on all of it; they’re pressure-testing yours, not learning about the category from you. What they can’t know in advance is whether the numbers are real, whether the founder understands what drives them, and whether the business can actually scale what it has. The velocity conversation is the fastest way to answer all three.

On terms: negotiate sequence, not just price. The first close sets the template for every subsequent close in the same round. Get the preference and participation terms right on the first check, because the sixth investor is going to match whatever the first investor got.

What a no usually means.

“We’d love to see a bit more traction” is not actionable feedback by itself. Ask what evidence would have changed the decision — velocity, repeat, margin, concentration, or scale. Then look for the pattern across passes. If five funds cite the same thing, it’s real. The pattern across no’s is the diligence report you didn’t pay for.

What the instruments do.

Most early consumer rounds use SAFEs or convertible notes because they’re fast and cheap to close. The simplicity is real. The danger is structural: a SAFE is optimized for a company expected to raise a priced round. Consumer brands do not always follow that sequence — a company may become profitable, remain privately held, or sell before the conversion event the founder imagined. Standard SAFEs address liquidity events, but the payout depends on the precise document and the relationship between the purchase amount and the holder’s as-converted proceeds. The risk is not that every unconverted SAFE becomes legally indeterminate. It is that several early instruments — SAFE, convertible note, angel check, each with its own cap and discount — can produce an ownership and preference outcome the founder never modeled. Standard YC forms behave differently from bespoke investor forms, and the difference surfaces at exit, not at signing.

Software seed documents are commonly designed around an expected priced Series A. Many successful consumer companies never raise one.

The worked example: $8 million raised — $2 million on a SAFE and $6 million in a priced round — at a $20 million post-money valuation. Preferred owns 40 percent. Exit at $30 million.

With 1x non-participating preferred, investors choose between their preference and converting to common. At 40 percent of a $30 million exit, conversion pays $12 million — better than the $8 million preference — so they convert. Common receives $18 million.

Add 1.5x participating preferred on the $6 million priced round. The preferred takes $9 million off the top, then participates pro-rata in the remainder. Preferred’s 40 percent of the remaining $21 million is another $8.4 million. Total to preferred: $17.4 million. Total to common: $12.6 million. The participating provision costs common $5.4 million on an exit that looked identical from the outside. The preference stack is the mechanism the 101 named; this is what it does in arithmetic.

The three terms to understand before signing anything: liquidation preference (how much, and at what multiple), participation (whether preferred keeps participating in upside after getting paid out), and redemption rights (whether investors can force a liquidity event after a set period — rare in venture, not absent in consumer bridge capital).

The position to close from.

Run the process so that a no from every fund still leaves you with a business. Proven velocity, a clean data room, and a working-capital cycle that does not require equity buy the founder time. That time is leverage: the ability to reject the wrong investor and the wrong terms without putting the company at risk.

Subscribe now

Responsibility moves before authority does.

A company reaches a certain size and begins producing the same complaint in every function.

Nobody owns anything.

The product team is waiting for the roadmap. Sales is selling around gaps in the product. Marketing cannot finalize the positioning because the founder keeps changing the language. Finance has built three versions of the forecast because nobody will commit to a hiring plan. Decisions happen in meetings and return as questions two days later.

The founder is exhausted. The team is frustrated. The board suggests hiring someone senior. A head of product will own the roadmap. A CRO will bring discipline to revenue. A COO will connect the functions. The job description expands as each unresolved problem is added to it. Eventually the company is looking for a person who can set strategy, drive execution, create accountability, align the team, professionalize operations, preserve the culture, and free the founder to focus on the things only the founder can do.

This person does not exist.

There are excellent executives who can do meaningful parts of that job. What they cannot do is resolve the question the company has avoided: who is allowed to decide what?

An org chart appears to answer this. Product decisions belong to product. Revenue decisions belong to revenue. The CEO sits at the top and resolves whatever crosses the lines. Very few important decisions stay inside the lines. Pricing belongs to product until it changes the sales motion. A new customer segment belongs to revenue until supporting it changes the roadmap. A partnership belongs to business development until it requires engineering resources, changes the positioning, and appears in the forecast. The decisions that matter are important precisely because they alter several parts of the company at once. Assigning them to a function does not resolve the dependencies. It only determines where the argument begins.

In a founder-led company, the argument usually ends with the founder. Sometimes this is explicit. More often, authority is transferred in language and retained in practice. The new executive is told she owns something. She studies the problem, makes a decision, and begins moving the team. Then the founder encounters the consequence somewhere else in the company and reverses it. A pricing change makes the product feel too expensive. A positioning decision excludes a customer the founder still wants. The executive technically owned the decision, but the founder retained the right to redefine the objective after seeing the answer. The next time, the executive asks before acting. Now the founder is frustrated that the person she hired is not proactive enough.

This is the central trap of executive hiring in an early company: responsibility moves before authority does.

The executive becomes accountable for an outcome produced by decisions she cannot reliably make. She can recommend, persuade, coordinate, and follow up, but the consequential choices remain provisional until the founder has personally encountered them. Titles make this harder to see. “Chief” sounds like authority. “Own” appears throughout job descriptions. None of these tells the organization what happens when two legitimate objectives conflict.

So decisions climb. The founder’s calendar fills with choices other people were supposedly hired to make. Leadership meetings become queues of unresolved tradeoffs. Executives spend more time preparing decisions for approval than acting on them. The company has added senior capacity without reducing its dependence on the founder.

This does not always mean the founder is controlling. Sometimes the company has never distinguished between founder judgment that is genuinely irreplaceable and decisions the founder continues making because she always has. Those are different categories. A founder may be the company’s best judge of product quality, brand, technical architecture, or the character of the people trusted with the mission. Preserving that judgment can be rational. The purpose of decision design is not to distribute every choice democratically. It is to identify where her judgment creates disproportionate value and stop requiring it everywhere else.

Without that distinction, the organization cannot learn the boundaries of its authority because the boundaries move with the founder’s attention. The founder does not need to decide less in the abstract. The company needs to decide which kinds of decisions belong to her.

A useful starting point is three questions. How reversible is the decision? Where does the relevant information live? How far do the consequences travel? A decision that is reversible, local, and made close to the relevant information should rarely climb. A decision becomes more senior as its consequences become harder to reverse or spread across more of the company. Some decisions remain the founder’s because they alter the identity or survival of the company: whether to raise, whether to sell, which compromises the company refuses to make.

For each recurring class of decision, the organization should know who decides, whose input is required, which constraints bind the decision, and what would cause it to escalate. The person deciding should understand the objective before the options arrive. Otherwise, the founder still owns the reasoning while the executive merely selects from choices prepared elsewhere.

Decision rights become usable only when the company has made its priorities consequential. Not “grow efficiently” — what happens when growth and efficiency conflict? Not “enterprise is important” — does the company modify the product for an enterprise contract? Not “protect the brand” — who can decline revenue on that basis, and how much? The organization does not need a policy for every situation. It needs enough explicit constraint that two capable people using the same information are likely to make compatible decisions.

This is why decision architecture depends on organizational memory. Authority without context produces arbitrary choices. Context without authority produces informed waiting. The company needs both: a representation of what is true and an agreement about who can act on it.

Without stable authority, learning cannot accumulate. A decision made by an executive and later revised by the founder produces no clean signal about either person’s judgment. The company only learns that ownership lasts until disagreement. This is why replacing the executive rarely solves the underlying problem. The first was too passive. The second was too forceful. The third did not understand the founder. Each may have had real weaknesses. But if the company never established which decisions the role could make, every candidate is being evaluated against an authority structure that exists only in the founder’s head.

The perfect executive is imagined as someone who will independently reach the conclusion the founder would have reached, using context the founder has not transferred, without requiring the founder’s time, and without making a consequential decision the founder dislikes.

The company is not hiring judgment. It is hiring clairvoyance.

Before opening another executive search, the founder should ask which of the decisions returning to her genuinely require her judgment — and which return because two objectives conflict and the company has never ranked them. The answers may still support the hire. But the role should then be designed around a real transfer of decisions, not an accumulation of problems.

The founder should be able to say: these are the outcomes you own, these are the decisions you can make without me, these are the constraints that bind them, these are the circumstances that require escalation, and these are the decisions I am deliberately keeping.

That is an executive role. Everything else is a request for someone to carry responsibility while the company postpones deciding where authority lives.

If your company is adding senior people without transferring the authority that would make them useful, I work with seed–Series B founders through RIN on the decision architecture underneath the org chart. Get in touch.

In the beginning, the founder is the company’s best system.

She remembers who asked for what, why the product works the way it does, which investor needs an update, what was promised to the first customer, and the reason a decision that looks arbitrary was made three months ago. She knows which partnership is real, which is politeness, and which will happen if someone remembers to follow up on Thursday. None of this is written down particularly well. It doesn’t need to be. The founder is in every conversation and has enough context to reconstruct the state of the company whenever someone asks.

Then the company grows. A salesperson joins and begins having customer conversations the founder does not attend. An engineer makes a product decision based on a support issue the salesperson has never seen. Someone takes ownership of partnerships, someone else starts managing the financial model, and the founder is added to more channels so she can stay close to all of it.

The company now has more information than the founder can personally absorb, but its operating design still assumes that she can.

The symptom is rarely described that way. It appears as slow execution, poor communication, lack of ownership, misalignment, or a team that needs to be more proactive. A decision is delayed because the founder has not responded. Work is redone because an earlier decision cannot be found. A customer receives an answer that contradicts what they were told two weeks ago.

The conclusion is usually that the team needs to communicate better. So the company adds meetings. A weekly leadership meeting. A product review. A pipeline review. A planning session. The founder remains the only person who can connect the information across them. She attends all of them, answers questions throughout the day, and becomes increasingly frustrated that nobody else sees the whole company.

Nobody else can. The whole company exists inside her.

Every company has a state whether it represents one or not. What customers have been promised. What the company currently believes about the market. Which decisions are open, who can make them, what the product is optimizing for, which accounts are at risk. In an early company, these facts belong to different functions and move quickly — the CRM knows that an account advanced, Slack knows why, the product tool contains the feature request, the founder remembers that the buyer only advanced because a feature was promised. Each system contains a piece of the truth. The founder supplies the relationships between them.

As the company grows, the number of relationships between facts increases faster than the facts themselves. A new customer affects the roadmap. The roadmap affects hiring. Hiring affects burn. Burn affects the financing timeline. Adding people does not automatically create capacity — every new person also creates information that must be integrated with everything else. If the founder remains the integration layer, adding headcount can increase the demand on her faster than it increases the output of the organization. She experiences this as becoming a bottleneck. That phrase makes the problem sound personal. She needs to delegate. She needs to let go.

But a founder cannot delegate a decision by transferring the task alone. She has to transfer the state required to make it: the objective, relevant history, constraints, dependencies, definition of a good outcome. Without that, the employee receives authority without context. The decision comes back differently than the founder expected, she intervenes, and everyone concludes she has a control problem. The employee learns that ownership is provisional. The founder learns that delegation produces lower-quality decisions. Future work returns to her earlier. Her memory becomes even more central.

This is how a company becomes dependent on a person while insisting the problem is that the person will not step away.

Documentation helps, but most documentation records the wrong things. Companies write process documents because processes are visible and stable enough to describe. The information trapped in the founder’s memory is usually less procedural. It is interpretive. Why this customer matters more than its current revenue suggests. Why the company is tolerating a manual workflow for one segment but refuses to support it for another. Why a metric that appears to be improving is not yet trusted. A checklist cannot preserve judgment. It can preserve the inputs, constraints, and history that allow someone else to exercise judgment without starting from zero.

The company therefore needs more than a knowledge base. It needs a way to represent what is currently true. That representation can be surprisingly small: a list of active company assumptions and the evidence supporting them; a decision log that records what was decided, by whom, why, and what would cause the company to revisit it; an account state that distinguishes what happened from what someone hopes will happen next; a financial forecast whose assumptions have owners and dates. The purpose is not to capture everything. It is to preserve the information whose absence causes the company to reconstruct reality each time it makes a decision.

This is often the moment someone proposes hiring an executive. The imagined executive will absorb context from the founder, hold the functions together, establish accountability, and create enough leverage for the founder to return to product or customers. A good executive may help. But if the company has no external representation of its state, the executive’s real job for months becomes interviewing the founder: what happened here, why did we decide this, is this account actually going to close, who owns this, can I make this decision. The founder may interpret the questions as evidence that the executive is not sufficiently strategic. The executive may interpret the answers as evidence that the company has no strategy. Both are observing the same structural absence from opposite sides.

A company becomes capable of using senior leadership when important context can survive outside the founder long enough for another person to act on it.

That does not require removing the founder from the center. Her taste, conviction, and accumulated understanding are not inefficiencies to eliminate. The objective is to stop making her immediate availability the condition under which everyone else can do good work. The difference is visible in what happens when she leaves for a week. In one company, decisions wait. Meetings become status reports. Work continues where the path is obvious and stops where interpretation is required. In another, the team can see what changed, which decisions fall within existing reasoning, and which genuinely require her judgment. She remains important without being required for the company to remember itself.

The company has outgrown the founder’s memory when the amount of context required to operate exceeds what one person can reliably hold and transmit. Beyond that point, working harder makes the system more fragile. The founder becomes better informed and more exhausted while the organization becomes more dependent on access to her. Growth changes this before anyone notices — the company acquires more customers, more employees, more capital, more systems, while preserving the original assumption that somebody at the center still knows how all of it connects.

Eventually, she does not.

The founder is usually the first person to feel that failure and the only person the organization has no way to relieve. So she works longer, attends more meetings, answers faster, and tries to become a larger memory.

The company does not need her to remember more. It needs to learn how to remember itself.

If your company has outgrown the way it holds context and makes decisions, I work with seed–Series B companies through RIN to build an operating structure that no longer depends on the founder remembering everything. Get in touch.

Every founder raising for a consumer brand has read the same guides, and the guides were written about a different industry. The advice in them is good: run a tight process, create scarcity, get warm introductions, raise enough runway to reach the next financing milestone. All of it assumes a company with low marginal production costs, legible recurring revenue, and a small number of outcomes large enough to carry venture portfolio math. A company that manufactures inventory, sells through intermediaries, and exits into a concentrated pool of strategic buyers obeys a different set of constraints. The mechanics of the advice may survive: process, scarcity, references, narrative. The underwriting logic does not.

A software seed prices a search for a business that doesn’t exist yet. A consumer round prices an equation that has already started to work: margin that survives the retail stack, acquisition that doesn’t lose money, and repeat.

So this rebuilds the answers from companies that actually produced great consumer outcomes. Three archetypes anchor it. RXBar: started, by Peter Rahal’s own telling, with $10,000 and a loan from his father, sold to Kellogg for $600 million having taken essentially no institutional money. 1 Siete: a reported $1 million from Austin angels, then one $90 million minority investment led by Stripes, then a $1.2 billion sale to PepsiCo.2 Liquid Death: raised repeatedly against demonstrated demand, reaching a $1.4 billion valuation before profitability. 3 Bootstrap, one big check, and the venture path.

Should you raise venture at all?

Start with what the person who bootstrapped to $600 million now says as an investor. Peter Rahal spent years after the RXBar sale angel investing in consumer seed deals and concluded the practice didn’t hold up: in his words, there are “not enough billion-dollar outcomes to offset all the losses.”4 His response wasn’t to leave the category but to move down the risk curve — his fund Humble Growth, by his description, writes $10 to $40 million checks into brands already doing $20 million or more in sales.5

That’s not one investor’s quirk; it’s the shape of the whole capital market. Consumer capital concentrates at the growth stage because that’s where the math works, and recent outcomes keep confirming the pattern from the founder side. Hailey Bieber has said she funded much of Rhode herself, and the brand reached a billion-dollar acquisition without a conventional venture financing path.6 AG1 grew on its own cash flow beyond a reported $150 million revenue run rate before taking its first external investment in July 2021, then raised $115 million at a valuation above $1 billion in early 2022, from a position where the round was a choice.7 The institutional seed check that routinely finances a software company’s search is scarcer in consumer, and the firms that write it underwrite a different kind of proof. For consumer brands, “should I raise venture” usually means “should I raise it yet,” and the answer runs through what you can prove.

What do you need to prove?

The software guides say team, market size, and early product, because a software seed prices a search. A consumer round prices an equation with three terms: gross margin that survives the retail stack, acquisition that doesn’t lose money, and repeat. Repeat is the hardest of the three to fake. SaaS retention is contractual and observable. Consumer retention is a habit that gets re-earned every time someone stands in front of the shelf. This is why velocity — units per store per week, reorder rates, repeat cohorts — is what serious consumer investors underwrite. Doors go in the deck. Velocity decides whether the doors meant anything.

When e.l.f.’s CEO explained paying up to a billion dollars for Rhode, the justification was the numbers: “they’ve gone from zero to $212 million in net sales, direct-to-consumer only.”8 The case studies all cleared their proof bar before institutional money appeared. Siete had demonstrated demand in retail for years before its principal institutional round, and Miguel Garza has said the family wasn’t looking for capital when Stripes approached. Poppi’s TikTok moment produced a hundred thousand dollars of Amazon sales overnight (”We did $100,000 while we were sleeping,” as Allison Ellsworth told it), well before its later institutional financing.9 Liquid Death used a viral launch video to create retailer demand before scaling its capital base. In each case, outside capital followed observable evidence that demand could travel.

When should you raise?

Later than the software calendar suggests, and against evidence rather than runway math. In many consumer categories the earliest demand test can be run far more cheaply than the replication of what it discovers: a farmers market, a co-op shelf, a modest e-commerce page. Some categories genuinely can’t test cheap — regulated supplements, refrigerated products, anything where minimum order quantities put six figures between you and a valid experiment — and in those, raising earlier is legitimate because acquiring the evidence is itself capital-intensive. The rule that survives both cases: raise equity when the money can multiply evidence you already have, or when obtaining that evidence is the expensive part.

Native is the clean case of the first kind. Moiz Ali reportedly raised about $500,000, built a deodorant business whose subscription economics visibly worked, and sold to P&G roughly two and a half years after founding at a price reported around $100 million.10 The raise was small because the proof came fast, and the proof came fast because the product repeated. Raising before proof, in a category where proof is cheap, means selling permanent equity to fund discovery you could have financed with patience.

How much money does the business actually need?

This is the section the software guides can’t write, because a software company’s cash needs are mostly salaries. A consumer brand’s cash needs live in the working capital cycle, and the cycle is where most of the real financing decisions happen. Money leaves when you pay the co-packer’s deposit against a minimum order quantity. It sits in inventory, then in a distributor’s receivable on sixty- or ninety-day terms, and what comes back has been shaved by trade spend, promotions, deductions, free fills, and the occasional chargeback you’ll spend a month disputing. Gross margin is not contribution margin, and contribution margin is what you actually finance. Add customer concentration — one retailer as half your volume is also one payment schedule as half your cash flow — and forecast risk, because inventory you guessed wrong on doesn’t just tie up cash, it expires.

Run that cycle and you’ll usually find the number you need is a working capital number with a duration attached, not a war chest. That gap is a dated problem. Equity is a permanent instrument. Don’t match them. But the matching cuts both ways: debt isn’t automatically the safe choice. Early brands often can’t get it cheap, purchase order financing carries real fees, inventory lenders advance only a fraction of eligible stock, and a personal guarantee can make a bad quarter worse than dilution ever would. The right instrument depends on how certain the receivable is, what collateral exists, and how much margin survives the financing cost. The principle isn’t debt good, equity bad. It’s that the instrument’s duration should match the problem’s.

Who should you take money from?

Consumer money comes in roughly six kinds, each with its own clock. Angels and operators write the earliest checks, often the only formal capital before real traction. Celebrity and cultural capital is its own category. Poppi shows its real mechanics: $400,000 for 25 percent on Shark Tank reads as brutal dilution until you notice the investor was Rohan Oza, whose firm builds beverage brands for a living, and who then led Poppi’s disclosed institutional rounds; Ellsworth has said they deliberately kept the cap table slim.11 Distribution capital priced like equity is expensive; distribution capital that shows up can be the whole company. Rhode demonstrates the inverse: when the founder is the cultural capital, you don’t need to buy it. Incubator equity is beauty’s variant — Kylie Cosmetics traded large ownership to Seed Beauty’s product machine for speed and infrastructure. Strategics and distributors invest for information and lock-in: Liquid Death’s rounds included Live Nation and the Anheuser-Busch distributors who move its cans. Growth equity is where the institutions genuinely live — Stripes into Siete, VMG’s reported minority position in Drunk Elephant before the $845 million Shiseido sale12, Rahal’s Humble Growth — arriving when there’s real revenue to underwrite, often as the first and last outside money. And beneath everything, the working capital stack from the previous section, which is not glamorous and is most often the correct answer.

What should the cap table look like?

Small, boring, and yours for as long as possible. Exit multiples don’t punish cap-table clutter; dilution divides proceeds, it doesn’t shrink the price. What consumer multiples actually do is make dilution harder to hide. Run the recent exits: PepsiCo announced the Poppi acquisition at $1.95 billion, including $300 million of anticipated cash tax benefits for a net purchase price of $1.65 billion, plus a performance-based earnout.13 Siete’s $1.2 billion was roughly 2.4 times reported projected 2024 revenue.14 Rhode’s guaranteed $800 million was about 3.8 times trailing net sales, rising toward 4.7 only if the full earnout pays.15 Those are the great outcomes, and none carries the kind of multiple where a sliver of ownership becomes enormous.

Three separate risks hide inside “dilution.” Straight founder dilution, which compounds round over round. The preference stack — liquidation preferences and participation rights that can turn a good strategic sale into an outcome that works for investors and barely works for common. And valuation overhang: a price so high that only an unusually large exit lets the founder participate meaningfully, which converts every reasonable acquisition offer into a disappointment. Software portfolio math can support terminal outcomes large enough to forgive heavy financing stacks in the winners. Most consumer outcomes leave less room. The founders who preserved the most ownership were the ones who took little capital, took it late, or sold only a minority position: RXBar, Siete, Rhode, Drunk Elephant. They weren’t necessarily better negotiators. They kept more because the business produced options before the financing did.

What are the mistakes?

Raising software-style: growth bought below contribution margin on the theory that scale fixes economics. Casper was reportedly valued at $1.1 billion in its last private round and went public at an implied valuation around $470 million, because the public market performed the arithmetic the private one had deferred.16 Glossier is the beauty rhyme: a reported $266 million raised at a peak $1.8 billion valuation, followed by layoffs and a strategy rebuilt around the retail economics that were true all along.17 Treating placement as traction: doors in the deck, no velocity behind them, an opening order converted into a valuation the reorders then have to justify. Raising for a working-capital gap: the ninety-day problem financed with the forever instrument. And the one nobody prices until it happens: your exit market is a handful of strategics under antitrust review. Harry’s built a real business, agreed to a $1.37 billion sale to Edgewell, and the deal died after the FTC moved to block it.18 A financing strategy that only works if the exit clears is carrying a risk the deck never mentions.

The exceptions.

Venture-scale raising has worked in consumer. Dollar Shave Club raised real venture money and sold to Unilever at a price reported around $1 billion by attacking distribution itself.19 Liquid Death raised round after round, reached $263 million in retail scanned sales across 113,000 doors, and was, by Cessario’s own account, still short of profitability at its $1.4 billion valuation. It’s working because Mike Cessario understood exactly what he was building: “the product itself has to be insanely shareable,” at a scale that made marketing the moat.20 The pattern in the exceptions is that the venture-sized bet was on distribution or attention as the product, not on capital fixing a consumable’s economics.

The selection-bias objection is real: a list of companies that raised little and sold big will always make raising little look optimal. The lesson is not that these companies won because they raised less. They were able to raise less because their economics and their demand generated options, and preserving those options compounded the founders’ outcomes. In these companies, the raising was downstream of the proof.

The sequence.

Prove the equation small. Finance the gaps with instruments whose duration matches the problem’s. Take durable money from people who bring distribution or pattern recognition, keep the cap table legible, and let the institutions find you at the stage where consumer capital actually concentrates. A great consumer business barely needs a fundraise, and the surest sign you’re ready to raise is discovering you no longer have to.

Subscribe now

The plan is old, and every few years it comes back sounding like the best of both worlds. Raise one seed round, use it to reach profitability, then never raise again. Keep the team tiny, let software do the coordination, grow at whatever rate the cash flow supports, and own most of the company forever. Seedstrapping. One drink of venture, then sobriety.

The plan treats the seed round as a small, self-contained transaction. It isn’t one. A seed check is priced as the first payment on a much longer path, and the investor writing it is not buying your company as it exists. They’re buying an option on the company you’d have to become.

Whether venture fits a given business is a separate question, and some excellent companies should answer no. Seedstrapping is stranger than that. It’s the plan of a founder who has already answered no and takes the money anyway, on the theory that one small round doesn’t count. And the pitch contains its own problem: we’ll be so efficient we’ll never need you again. The founder says that with pride. The investor hears the ambition the check was priced on being quietly withdrawn. The follow-on round, the markup, the reserves waiting for the winners: those are machinery, and a fund can live without any one piece of it if the destination survives. A company that reaches an enormous outcome without ever raising again is a spectacular trade. Seedstrapping cancels more than the next round; it cancels the outcome a next round would have existed to chase. Capital efficiency in service of that outcome is a gift. Capital efficiency as the destination is the disqualifying feature. Venture doesn’t fund need; plenty of businesses need money and no fund wants them. It funds appetite, the ability to convert more capital into more speed toward something large. A company that has decided to stop converting has, from the fund’s side of the table, stopped being interesting.

Look at how the money actually works. A seed fund runs on power law: most positions go to zero or nowhere, and the fund’s return comes from a small number of investments that grow enormous. For that math to function, every check has to be written into a company that could plausibly be one of the enormous ones. Not will be. Could be. The investor knows most won’t. What they need is for each one to be trying, and trying doesn’t mean serially raising. It means continuing to pursue an outcome large enough to matter to the fund, at whatever speed capital allows.

That’s why the round happened at all. The valuation, the speed, the willingness to fund a company with no revenue and a partial product, all of it prices in the assumption that the company will keep choosing the largest version of itself, a path that almost always runs through an A, then growth, then an exit large enough to matter. Take that out and the same check makes no sense. No fund prices a seed round of a company planning to be a $6 million profitable business, because at seed-fund economics that outcome rounds to zero.

Which is the part the seedstrapping plan skips. The founder thinks they raised a small amount of money. The investor thinks they bought a ticket to everything after it. Both leave the closing dinner saying alignment, and the word means opposite things at the two ends of the table.

The misalignment takes five or six years to surface. By then the company worked. It’s profitable, growing 25 or 30 percent a year, throwing off cash, run by a small team that likes their lives. By every operating measure it’s a success. On the investor’s books it’s a problem. There’s been no follow-on round, so there’s no markup; the position has been carried at cost for half a decade. There’s no exit in sight, so there’s no distribution. The fund that wrote the check has a ten-year life and LPs who measure it on money returned, and this position, one of the fund’s rare working companies, is illiquid indefinitely. A fast failure at least resolved. This just sits.

So the pressure starts, and it arrives politely. Have you thought about what a growth round could unlock. We know a banker who’s great with founder-friendly processes. Would you consider a secondary. The investor isn’t being cynical; they’re doing their actual job, which was never patient minority ownership of a private cash-flowing business. The founder now spends real time managing the expectations of someone whose model the company quietly opted out of years ago, and the options on the table are all versions of the same thing: re-enter the path, buy the investor out at a price with no market to set it, or live with a permanently disappointed shareholder who holds information rights and a board seat.

None of that is a scandal. It’s the deal working as designed, applied to a company that didn’t want the design.

The strange part is that the founder usually knew. The seedstrapping plan is chosen precisely because the founder doesn’t want the venture trajectory, and then it gets financed with the one instrument that exists to enforce that trajectory. The reasoning is that a seed is small enough not to count. But the obligation was never proportional to the check size. It’s structural. A fund that owns 15 percent of you needs your outcome to be fund-relevant whether they paid $1 million for it or $10 million.

If the plan is genuinely one round and out, the honest financing list looks different. Your own capital, if the ambition is sized to it. Debt or revenue-based financing once there’s revenue to underwrite. Angels who are investing their own money, know the plan, and price it, because an individual can be delighted with a dividend stream or a small secondary in a way a fund structurally cannot. And there’s a small set of investors who’ve built explicitly for this shape, with redemption rights or profit-share structures that give them a way out that doesn’t require your company to be sold. Fewer of them exist than blog posts about them, but the structure is real: the return path is written into the instrument instead of assumed from the trajectory.

Or raise the seed and mean it, which is the other honest version. The point was never that venture is a trap. It’s a specific product with a specific job, and the seed round is that product’s smallest unit, not a different product that happens to share a cap table.

Wanting a profitable company you mostly own is a fine ambition. Financing it with a promise to become something else is how founders end up, six years in, negotiating their independence back from someone they invited in for what seemed like a rounding error.

I keep circling one question: what’s the smallest a real company can be? Not a freelancer with an LLC, but an actual company—with products, distribution, customers, and almost nobody in meetings. I think the answer has changed in the last two years. Next issue, I’m going to try to put a number on it.

On the build:

Yodo is officially sold out of almost every SKU. Najo was the most popular and made a beautiful espresso; we ordered 200 pounds of green and flew through it. Mabu was the crowd-pleaser as a pour-over, blood orange and cacao. Miele was our top seller at Happier Grocery. The names all come from the beans themselves: Najo for Naranjo, the region. Miele because it’s honey processed. Taba for the obata varietal.

The catch with single-origin is that when a coffee sells out, it’s gone. In most cases, there’s no reordering, just next year’s harvest. Great for the coffee, terrible for scale, and grocery buyers don’t love SKUs and UPC codes that keep changing. That’s why Mesa exists. Mesa means table in Spanish. It’s the house coffee, something we love as much as our top seller Najo but can keep on rotation forever without the name or the code changing, while the specialty lots come and go around it. Whatever we can get of those is what it is.

There’s no world where a grocery giant wants a SKU with a max capacity of 100 bags.

Where I went:

Reforming Pilates — new studio that opened in Miami Beach. It’s not as hard as Forma but has been a solid workout for the past week.

Nami Nori — a classic in New York thats been open in Miami for about a year or so? I oscillate between here and Pari Pari for my handroll fix.

On the wishlist:

_{Some links below are affiliate links — small commission to me, no change to the price.}

Miu Miu bag: LINK, Crown Affair finishing gel: LINK, The Row leather coat: LINK, The Row slides: LINK, Adyela dress shirt: LINK

Open loop:

A question for the bootstrappers and agency operators reading this: what's the last “role” you eliminated, and what replaced it — software, a process, or nothing?

For me, retailer outreach and CRM tracking were among the first functions to disappear into software. Claude Cowork now handles nearly all of it. If it would be useful, I can share the entire system in a future post.

— Lou

The first customers feel disproportionately important because they’re the first evidence the company might work. Someone who isn’t your friend gives you money, then another person does, a few logos appear on the website, revenue stops being zero. So you do the obvious thing: study how those customers arrived and try to do more of it.

This is where the trouble starts. The first ten customers are ten observations collected under unusually favorable conditions. The founder sold them personally. The product may have been configured around their needs. They may have come through investors, former colleagues, or people who’d followed the founder for years. They tolerated missing features because they believed in the team or had a problem acute enough to overlook everything that stops a normal buyer. None of that makes them bad customers. It makes them dangerous evidence. They tell you someone wants what you have. They don’t tell you why, how many more exist, what it costs to find them, or whether anyone other than the founder can convince them to buy.

I’ve seen companies treat the path to their first customers as a repeatable motion when what they’d actually discovered was that the founder had a network. I’ve seen consumer brands read a handful of enthusiastic retailers as wholesale demand when every placement came from someone the founder already knew. I’ve seen software companies hire a salesperson because the founders had closed a few large accounts, then watch the salesperson fail to reproduce whatever happened in those rooms. The salesperson gets blamed. Sometimes the salesperson is bad. Sometimes there was never a sales process to inherit, just a founder.

The distinction is easy to miss because founder-led sales works so well. A founder can sell around an unfinished product because they know what it will become. They can make commitments in the room, change the roadmap after the meeting, price creatively, and carry the credibility of being the person building the thing. They hold the entire context of the company in their head. A salesperson inherits a deck.

When the first sales hire fails, the conclusion is usually that the company hired the wrong person, and the more useful question is whether there was anything reproducible to hand them. A GTM strategy begins when you can explain why the customer bought without relying on the biography of the person who sold them: who has the problem, what makes them go looking, how they describe it before they know you exist, where they look, who owns the budget, what they’re comparing you against, why now instead of in six months, and what has to happen between first contact and money arriving. Basic questions. The first ten customers let you avoid answering every one of them. If demand arrives through your network, you never learn where strangers discover companies like yours. If customers trust you personally, you never learn which proof points create trust. If every deal is different, you never learn what the sales process is. Revenue can hide an extraordinary amount of uncertainty.

Which is why companies keep trying to scale a motion before they’ve found one. Five customers close and sales gets hired. Ten stores stock the product and a wholesale rep appears. Paid spend goes up because early spend seemed to work. Everyone is trying to turn the thing that happened into more of the thing, and scale is multiplication, so it’s worth checking what’s being multiplied. If every customer required a different story, more salespeople multiply confusion. If customers bought on founder relationships, more outreach doesn’t produce more founder relationships. If paid acquisition worked while you were spending very little, a bigger budget may reveal you never had a channel, just a pocket of cheap demand.

Wholesale is where I’ve watched this mistake get expensive fastest, because retail placement looks so much like progress. Getting into stores feels like distribution. A placement is permission to attempt a sale; the customer is still the person standing in front of the shelf. The opening order tells you a buyer was willing to take a chance. The reorder tells you something moved. Repeated reorders begin to tell you there’s a business. Brands routinely raise money and hire against the first signal, and the first signal is the one most contaminated by relationships, charm, and a buyer taking a flyer. Doors is the number that goes in the deck. Velocity is the number that decides whether any of it was real.

Founders also tend to study only the customers who bought. The favorite one, who understood immediately, bought quickly, uses the product constantly, and never asks for anything unreasonable, is real evidence: that’s where the pull is strongest, and enough of them start to outline the center of the market. Wonderful customer, incomplete teacher. The center can’t show you where the system breaks. For that, the informative customers sit at the edges: the one who almost bought, the one who took three months, the one who loved the demo and vanished, the retailer that placed an opening order and never reordered, the person who should obviously be a customer and isn’t. The edges carry their own trap, since some of those people were never going to buy and chasing them warps the product toward a market that doesn’t exist. But the places the motion stops working are the map.

None of this means refusing your advantages. Early companies should exploit every unfair one they have. Founder charisma, investor introductions, a launch, a viral post, one buyer at a major retailer who happens to love you, white-glove onboarding for every account. Use all of it. Just don’t confuse the advantage with the system you’ll eventually need, because the advantages are precisely the things that don’t multiply.

The pressure to confuse them peaks right around customer ten. There’s revenue now. Investors want to know how it grows. The founder is tired of selling and the pipeline needs managing, so it feels like time to hire someone to own GTM. Before making that hire, look hard at what they’re being asked to own. If the answer is figure out who wants this, why, how to find them, what to say, what to charge, why customers churn, and then scale all of it, the job on offer is discovering the business, not running one. Discovery is founder work. Handing it to an employee doesn’t delegate it, it just adds a salary and a scapegoat.

The first ten customers exist to give you contact with reality, and the measure of whether you’re learning from them is that the magic requirement declines. Customer eleven should teach you something customer one couldn’t. By customer fifty, some part of the path should be boring. Boring means somebody other than the founder can do it, and that the company has turned an exceptional event into a repeatable one.

The first ten are exciting precisely because none of that exists yet. Ten people finding their way to you is not the same as knowing how to find the next thousand.

The first purchase order felt like the finish line. A buyer said yes, sent the PO, we shipped the box, and my coffee appeared on a shelf I’d wanted it on. I took the photo. I added the retailer’s name to the site. For about five minutes this felt like distribution.

Then I realized what I’d actually acquired was a new problem. Someone still had to take the product off the shelf, and nothing about the purchase order guaranteed anyone would.

Wholesale separates the person who decides to carry your product from the person who decides whether it earns the right to stay. The buyer gives you access. The consumer gives you demand. Those decisions can happen weeks apart, and early brands tend to measure the first one because it’s so much easier to see. A retailer’s order produces an email, an invoice, a shipment, revenue. What happens afterward often produces nothing at all. Maybe the product sells out in four days. Maybe three units sell. Maybe the buyer put it somewhere nobody looks, or customers pick it up and put it back, or the store loves it but already has six weeks of inventory. The silence after the opening order contains a remarkable amount of information, and most brands have no system for hearing it.

So they keep acquiring placements. Twenty doors becomes fifty, fifty becomes a hundred, and the number looks wonderful in a deck because doors are so countable. “We’re in 100 stores” sounds like demand. It may only describe availability. There is a large difference between being somewhere and working there, and nearly every incentive in early consumer pushes founders toward the first. New accounts are exciting and announceable; reorders are boring. The founder can post the shelf. Nobody posts the third purchase order from the same independent grocery store. But if I wanted to know whether a young brand was actually working, the third purchase order is what I’d ask to see. The first order says someone believed the product might sell. The third says they were probably right.

A door, in other words, is not an outcome. It’s a test. You’ve been given some amount of shelf space, for some amount of time, in front of some number of people, and now something has to happen. If it doesn’t, “get more doors” is a dangerous response, because you can scale the appearance of traction much faster than the demand underneath it.

The reason is mechanical: wholesale is two funnels stacked on top of each other. The buyer funnel is the one founders control and obsess over. Find the account, get the meeting, send the sample, follow up, get the order. The consumer funnel is the one that decides whether any of it repeats. The product arrives, gets merchandised, someone sees it, understands it, accepts the price, chooses it over what’s beside it, and ideally comes back. A company can spend months optimizing the first funnel while knowing almost nothing about the second, and this is exactly where placement counts start lying. A brand in 200 stores with weak sell-through and a constant need to replace accounts that quietly stop buying looks bigger than a brand in 40 stores that reorders every few weeks. The second brand has found the more valuable thing: a unit of distribution that works. That’s what you multiply. Distribution isn’t the number of places willing to try you. It’s the number of places where product, customer, price, and environment produce enough demand to keep you there.

The revenue can hide this for a while. Opening orders are revenue, and if you add accounts fast enough, the top line grows even while the cohort underneath deteriorates, new placements replacing the accounts that aren’t reordering. From far enough away the business looks like it’s growing. Underneath, it’s running on a treadmill. Software companies call this churn and organize themselves around measuring it. Wholesale churn is quieter because nobody cancels. There is no event called lost account. There is simply an absence. The purchase order that should have arrived doesn’t. The account that took twelve cases takes six. The store that carried four SKUs carries two. The buyer who always responded stops responding. The account doesn’t go from healthy to dead overnight; it becomes slightly quieter, and by the time anyone notices, the explanation on offer is “they stopped ordering,” which is an outcome, not an explanation. The useful information existed earlier, in events that never occurred, and most companies record events beautifully while noticing absences not at all.

This is why the most important wholesale question isn’t how many doors are we in. It’s what happens after the first order. How long until the second. Which SKUs reorder and which accounts never do. Whether velocity changes with placement in the store, whether staff recommendation matters, whether a particular retailer makes other buyers say yes faster, what happens at a different price. Asked this way, a store stops being a logo and becomes information, and the differences between stores become more useful than the total. Your first twenty accounts are your cheapest laboratory. They can tell you which kind of retailer works, which product earns its space, which price survives contact with the shelf, which parts of the wholesale pitch are promises the product actually keeps. Sometimes they tell you the problem isn’t distribution at all: you got the placement and the product simply isn’t moving. That is painful information, and it is radically cheaper at twenty doors than at two thousand.

Not every placement is worth having, which sounds absurd when you have three accounts and the fourth feels definitionally good. But every account creates obligation: someone has to onboard it, fulfill it, invoice it, chase the payment, answer the questions, and eventually figure out whether it will ever reorder. A $300 opening order can generate an impressive amount of work. If it never becomes a second order, you haven’t acquired distribution. You’ve acquired a task.

The lesson runs past retail. Growth gets narrated through addition: a new lead, customer, account, market. But the useful question comes afterward. Did it stay? Did it repeat? Did it work without heroic intervention? The first yes gets you onto the shelf. The second suggests you belong there. The third begins to prove it.

The placement is still worth celebrating. Someone took a chance on you, and that mattered. Just don’t put too much meaning into the photograph. The shelf isn’t proof that distribution worked. It’s where the test begins.

A few months into building my last company, an investor asked me what I did at the company. Not unkindly. There were four of us, there was no product yet, and my title was Chief Operating Officer, which implied the existence of operations. He was asking a reasonable question. What is there to operate?

I gave him some answer about wearing many hats, which was true and also the least accurate thing I could have said. It made the job sound like coverage, like I was a substitute teacher moving between classrooms. That is the standard understanding of the early operator: the person who picks up whatever work doesn’t clearly belong to someone else. It gets the job exactly backwards, and it took me a long time to find the language for why.

Operating a mature company is largely an optimization problem. The organization already knows what it sells, roughly who buys it, how the product gets made, how customers find it, where decisions get made, and what a good month looks like. There are exceptions everywhere, but the underlying system exists. The operator’s job is to make it work better.

At an early company, most of those things are still variables. The customer might be wrong. The product might be wrong. The way you’re selling it might be wrong. The person who appears to own a decision might not have the information required to make it. A problem that looks like execution might be strategy, a problem that looks like hiring might be product, and a problem that looks like marketing might be that nobody can explain what the thing is. If you optimize too early, you get very efficient at producing the wrong outcome.

So the first operating job is not efficiency. It is making the company coherent: figuring out what kind of company you’re actually building while the company is busy insisting it already knows.

In an early company, almost everything takes shape in relation to everything else. Product, positioning, fundraising, hiring, launch, pricing, customer conversations. It would have been easy to divide the work into functions and treat each one as its own problem. They weren’t separable. A positioning decision changed which customers we pursued. Those customers changed what the product needed to do. The product changed who we needed to hire. The hiring plan changed the burn, which changed the financing timeline. And the financing story depended on being able to explain all of it as though it had been obvious from the beginning.

This is what early operating work actually feels like. You pull on one thread and discover it was holding five things together. The skill is not knowing how to run each function better than the person who eventually will; a good early operator should end up surrounded by people who are better at every individual discipline. The skill is seeing the dependencies before the organization is large enough to make them obvious.

This is why the most important operating decisions rarely look operational. Deciding what not to build is an operating decision. So is deciding which customer not to serve, changing the pricing model because the current one creates the wrong behavior, or realizing the company doesn’t need another salesperson because the existing ones are being asked to sell something nobody has made legible. So is noticing that every meaningful decision returns to the same person and redesigning the decision instead of hiring another person to wait for it.

None of these problems are solved by a better checklist. They require understanding the mechanism producing the problem, which is a different kind of attention than managing the problem.

The deeper reason this work gets missed is that operations is usually treated as the place where strategy goes after the interesting decisions have been made. The founders decide what the company is doing; operations figures out how. That division makes sense much later than people think. In the beginning, the how is constantly producing information about the what.

You decide to sell to a certain customer and discover the sales cycle makes the economics impossible. You launch and discover people consistently use one minor capability while ignoring the thing you built the company around. You hire and find that every decision still climbs back to a founder because nobody understands the reasoning behind the roadmap. Execution is producing information about the strategy. The operating job is to make sure the company notices.

That is also why early operations gets mistaken for administration. The artifacts look almost identical: a meeting, a spreadsheet, a hiring plan, a roadmap, a weekly review. The difference is the question the artifact answers. Administration asks, “How do we keep track of this?” Operations asks, “Why does this keep happening, and what would have to change so it stops requiring someone’s attention?”

The spreadsheet isn’t the work; deciding what needs to be visible is the work. The artifact’s real function is often to force clarity the company has been avoiding. Writing down why a feature exists is not documentation. It is the moment the company finds out whether it knows why the feature exists. Often, the answer was less settled than it had seemed, and discovering that was the work. Companies reward forward motion. Someone has to be rewarded for making that motion checkable.

There is a failure mode here, and I know it from the inside. The early operator can become a handler instead.

A handler absorbs problems. They take the mess into themselves so everyone else can keep moving, and the company feels smooth, and everyone is grateful. It feels like service and it looks like competence. What it actually does is prevent the company from seeing its own structure. Every hole I filled personally was a hole the company never learned it had. The generalist who compensates well enough, for long enough, becomes the reason the design flaws survive.

The better version of the job is nearly the opposite: not absorbing the problem but locating it, naming which structural decision produced it, and making that decision visible enough to be remade.

If you do this well, something strange eventually happens. The company becomes operable. There is a sales motion worth repeating, a product direction people understand, a hiring bar that doesn’t get reinvented for every candidate, decisions that used to climb back to the founders getting made correctly without them. At that point, operations begins to resemble what everyone thought it was all along. There are systems to improve and margins to manage. The machine exists, and now someone can make it run better.

But somebody had to build the machine.

This is where early companies misunderstand the role. They hire for efficiency when they need synthesis. They look for someone who ran a function at scale when the problem is that nobody has figured out how the functions fit together. Or they wait until the chaos is visible. By then, the earliest design decisions have hardened into the company. The wrong customer became the sales strategy. The workaround became the process. The founder’s judgment became the decision system, and everyone got very good at navigating a structure nobody consciously chose.

The investor’s question had an answer. There was nothing to operate yet.

There was a machine to design, and it was being designed either way, by every small decision nobody experienced as a decision. Before there is anything to operate, the job is architecture. The only variable is whether anyone is doing it on purpose.

Spent an unreasonable amount of this week building Yodo’s new website.

It started with the fairly normal goal of making the site better and somehow ended with me building a coffee company inside a fake computer from 1999.

On the build:

There’s a desktop. Windows. A hard drive. A catalog that looks like an inventory utility. Right click works. Things drag. There are system dialogs and menus and enough unnecessary detail that I’ve developed opinions about how large a logo should appear during a fake boot sequence.

Most consumer sites have converged on the same thing: giant image, three lines of copy, product grid, newsletter signup. There are good reasons for that. I just don’t think every website needs to behave like a conversion template.

Yodo is a weird little coffee company. I wanted the site to feel less like visiting a coffee brand and more like finding somebody’s computer and looking around.

It took about a week. I’m not a software engineer. A few years ago something this interactive would have meant convincing several other people that all these tiny details were worth building; now I can spend a night obsessing over them and have the thing exist in the morning.

Where I went:

Lion's Den, on the future of functional beverages — A conversation hosted by Grotto Baths. I come at this from coffee, but the beverage shelf keeps getting stranger and more interesting. Alcohol has occupied an enormous amount of social territory for a very long time and it's increasingly obvious that whatever comes after it isn't going to be one replacement category. People want to feel awake, calm, social, focused, relaxed, or just to have something interesting at dinner that isn't wine. Coffee, tea, hemp, adaptogens, functional drinks, and a bunch of things that don't have a clean category yet. I left thinking there's a lot of room on the shelf.

Elsewhere:

"My Best Memory," Mira Ló — On repeat while building the site.

Taste businesses — The more I operate Yodo, the more interested I get in businesses where the important part is still somebody deciding what's good. Coffee obviously, but also food, hospitality, furniture, stores, hotels. The software does more and more of the boring work around that decision and can't make the decision. Very small organizations, unusually concentrated judgment, unusually automated coordination. I've started calling them taste businesses.

Open loop:

The thing I keep coming back to is how much of building a company is continuing after the interesting part is over.

Ideas are fun. Designing is fun. Naming is fun. Launching is fun. Then somebody owes you money, a buyer stops answering, an order needs following up, and something that worked last month stops working without announcing it.

Nothing is wrong until the sequence breaks.

— Lou

There’s a particular kind of product company that is always shipping and somehow never gets anywhere.

Something new appears. It’s thoughtful, polished, surprisingly complete. It ships. People use it less than expected. A month later it’s gone, or buried, or quietly replaced by another idea that is also thoughtful, polished, and surprisingly complete.

From inside the company, this feels like speed. You’re learning. You’re iterating. You aren’t attached to what you build.

But iteration requires something to persist between versions.

If the product keeps changing and your understanding of the customer doesn’t, you’re not iterating. You’re restarting.

Every product team ships things that don’t work. That’s normal. What matters is what the failure leaves behind. You thought people had a problem frequently enough to build around it and discovered they didn’t. You thought a behavior would become habitual and it didn’t. You thought people would pay for something they said they wanted and they wouldn’t. The feature can disappear. The knowledge shouldn’t.

The next bet should be more constrained because of the last one.

I think designers are particularly vulnerable to getting this wrong. Not bad designers. Very good ones.

A good designer can take a loose idea and make it convincing remarkably quickly. Give them something half-formed and a few days later it has hierarchy and interaction and language. The edge cases have been considered. The weird parts have been resolved. You can click through it and understand how it would work.

The idea has acquired coherence. The problem is that the coherence may belong entirely to the design.

This is one reason I’m skeptical of the assumption that great designers naturally make great product founders — I say that as someone whose way of thinking was shaped by design. There's obviously overlap. Both require taste, curiosity, an ability to imagine something that doesn’t exist yet. But there’s a difference that matters.

Designers are trained to resolve ambiguity by making something concrete.

Founders sometimes need to leave the ambiguity unresolved.

An idea in a conversation is still obviously an idea. There are holes everywhere. You can ask whether the customer actually has this problem, whether it matters, whether there’s another explanation for what you’re seeing.

Then somebody mocks it up.

Now there’s a screen. Then another screen. The interaction gets better. Someone solves the edge case. The prototype starts to feel good. People leave comments about the copy. Engineering starts thinking about implementation.

At some point the conversation moves downstream. Should we build this? becomes How should this work? Then: Where does this live? What happens when you click here? How long will this take?

Nobody necessarily decided that the thing deserved to exist. The artifact made it feel decided.

The mocks became the roadmap.

This is where very good design can become dangerous. Concreteness carries authority. A polished artifact implies that some amount of upstream thinking must have happened before it arrived. Someone must know who this is for. Someone must know what behavior it’s supposed to change. Someone must have decided this problem matters more than the other twenty problems the company could be solving.

Often the reasoning is much thinner than the artifact.

Design can also make an idea more convincing to the person designing it. Spend enough time with a product and you start resolving its contradictions. You find the language that makes the concept easier to explain. You simplify the interaction. You connect one piece to another until the whole thing begins to feel inevitable.

The product gets more convincing. You get more convinced by it too.

Internal coherence can feel a lot like truth. They aren’t the same. A product can make perfect sense once you’re inside it and still have no compelling reason for someone to enter.

Then it ships.

People don’t understand it, so you redesign onboarding. They aren’t finding the feature, so you improve discoverability. They aren’t returning, so you work on engagement. They don’t know what the product is for, so you rewrite the positioning.

Sometimes those are exactly the right things to do. Sometimes people just don’t want it enough.

Eventually the evidence gets difficult enough to ignore and the thing gets killed.

This is usually celebrated as good product discipline. And sometimes it is. Keeping something alive because you spent six months building it is obviously worse.

What interests me is what happens immediately afterward.

Failure creates an uncomfortable period where there is nothing obvious to work on.

Why didn’t people use it? Was the problem wrong? Was the customer wrong? Did people like it but not enough? Were we solving something that happens twice a year as though it happens twice a day? Did we mistake people saying something sounded useful for evidence that they would change their behavior? Was the product bad, or was the premise bad?

There’s nothing to screenshot here. No prototype to click through. For a while, all you have is evidence that something you believed was wrong and several possible explanations for why.

A very good designer can escape that period unusually quickly.

Sometimes the next product does not incorporate the lesson of the previous failure. It simply replaces the discomfort of it.

By Friday there’s another coherent product on the screen. People have opinions again. Engineering has something to estimate. The company feels like it has momentum.

The ambiguity is gone. So is the brief period when reality had the company’s full attention.

That ability is a superpower when you learned something from the last product. It’s dangerous when you didn’t.

Because a company can become extraordinarily good at replacing failed hypotheses without becoming any better at forming them.

Product A doesn’t work. Product B appears. Then C. Each one is beautifully resolved. Each one looks plausible enough to restore conviction.

I think a lot of companies that describe themselves as fast-moving product organizations are actually doing some version of this. They ship constantly. They experiment. They kill things without sentimentality. The product looks completely different every six months.

All of those things signal speed. None of them necessarily signal learning.

Shipping is not learning. Changing direction is not learning. Killing something is not learning.

Learning should leave you with fewer plausible explanations than you had before. You thought A, B or C might explain what customers were doing. You built something. Reality gave you an answer. C is no longer plausible.

Good. Something has been eliminated. The possibility space got smaller. That’s iteration.

The evidence that a company is learning isn’t how much the product changes. It’s how much more constrained the next decision becomes.

This is why I’ve become interested in a very boring question after something doesn’t work:

Not what are we building next. Not what did users request. Not what’s the new direction.

What do we actually know?

If the answer is fuzzy, I’m not sure the company has earned the next mock yet.

This is also where design judgment and product judgment separate.

Design judgment can tell you that an interaction is confusing, that the hierarchy is wrong, that something has too many steps, that the product lacks coherence.

Product judgment has to operate further upstream. It has to decide which problem is worth organizing the company around. Which behavior matters. Whether a request represents a need or a preference. Whether people find something impressive or actually depend on it. Whether the evidence supports another iteration or should kill the premise entirely.

Sometimes that means leaving Figma closed for another day.

I think the best designer-founders eventually become a little suspicious of their own ability to make things convincing. They know they can make almost any reasonable idea feel plausible, so plausibility stops being evidence. They get harder on the idea before becoming obsessive about its expression.

That distinction is about to matter much more because the ability to generate plausibility is no longer limited to designers.

Now everyone has it.

A vague product thought can already become a plausible interface in minutes. A prototype can exist before the meeting that produced it is over. Code follows increasingly quickly.

Feature didn’t work? Make another. Customers are confused? Generate another interface. Positioning is weak? Rewrite it. Product isn’t landing? Prototype a different direction by lunch.

For most of software history, execution imposed some friction on bad ideas. Not nearly enough; companies have always spent years building things nobody wanted. But making something still cost enough that the decision to make it carried weight.

That friction is disappearing.

Soon there will be almost no distance between maybe we should and look, it works.

This sounds like an enormous improvement in how quickly companies can learn. It probably will be for companies that are actually learning. For everyone else, it just makes restarting cheaper.

Making execution cheaper also makes being wrong cheaper.

That’s the part I think we’re underestimating.

AI gives every founder some version of the designer’s superpower: the ability to turn an unresolved thought into an extremely convincing artifact before the thought itself has earned much conviction.

We’re going to be able to produce answers much faster than we learn which questions matter.

So the discipline moves upstream.

Can you distinguish a bad execution from a bad premise? Can you remember what you believed before you shipped something? Can you say which part turned out to be wrong? Can you let that answer rule something out? Can you sit with the failure for long enough to understand it before relieving the discomfort by making something else?

The future of product development is going to contain an absurd amount of output. Output won’t be scarce. Understanding might be.

Great design can make a company extraordinarily fast at turning an idea into something real. AI will make everyone faster at it.

But speed only compounds when knowledge does.

Otherwise you can ship twelve products, kill twelve products, and end the year with roughly the same understanding of your customer you had when you started.

You weren’t iterating. You were producing artifacts faster than you were producing understanding.

After I wrote that my coffee company has no moat, a few people asked some version of the obvious question.

What stops someone else from doing the same thing?

Nothing. They should go do it.

I mean that seriously. There is no law preventing someone from sourcing excellent coffee from Costa Rica, putting it in a well designed bag and selling it to specialty grocery stores. They could probably figure out our roaster. They can see what we charge. They can walk into the same stores we sell to and pitch the buyer something similar.

What I have considerably less faith in is their willingness to spend the next few years doing all the annoying things required to make that sentence into a business.

I think founders wildly overestimate how many people are actually going to build the thing.

We imagine competition as though everyone else is sitting around with unlimited energy, waiting for a sufficiently good idea to appear. The moment you prove something works, this hypothetical person quits their job, raises some money, finds the supplier, builds the website, calls the buyers and spends the next several years trying to take it from you.

Real people are much less threatening.

They have jobs. They have other ideas. They get distracted. They decide they want to start a company and spend six months deciding which one. They buy the domain and never build the website. They build the website and never email anyone. They email ten people, hear nothing back and decide the market isn’t ready.

Mostly, they never start. And the ones who start mostly don’t keep going.

I don’t think this is because people are uniquely lazy. Building a business is just an unusually effective machine for turning enthusiasm into tedium.

The idea is fun. The first version is fun. The logo is fun. Telling your friends what you’re working on is fun.

Then you need a case pack. Then you discover the margins don’t work at the price you chose. Then the supplier who said six weeks means eleven. Then you send samples to fifteen buyers and twelve never respond.

One says no. One says maybe. One says yes, and three months later still hasn’t sent the purchase order.

Then you finally get into the store and discover that getting onto the shelf was not the accomplishment you thought it was. Now the product has to leave the shelf.

So you follow up again.

This is where a surprising amount of competition disappears.

Not at the idea.

At the fourth follow-up.

I think about this whenever someone says, “But wouldn’t someone else just do it?” Maybe.

But “someone could do this” is an almost useless statement. Someone could open a restaurant across the street from yours. Someone could make another olive oil. Someone could start an agency that offers exactly the same service. Someone could build software that does what yours does. Someone could make a hotel with better sheets and nicer soap and charge the same price.

Graza is probably the funniest example of this. They did not invent olive oil. They did not invent the squeeze bottle. They put good olive oil in a format that was more convenient and built a very good brand around it.

Could someone copy that? Yes. Did they? Immediately.

There are now enough olive oils in squeeze bottles that you could be forgiven for thinking the squeeze bottle grows naturally on an olive tree.

And Graza is still here. Even funnier, Graza eventually turned around and started selling olive oil in a glass bottle — the exact format every olive oil on the shelf came in before Graza arrived.

So apparently the squeeze bottle was neither proprietary nor permanent. It was just a good idea.

The bottle was easy to copy. The brand, distribution, retailer relationships, customer recognition, sourcing, operating knowledge and years spent becoming synonymous with the bottle were not.

By the time everyone else had copied the squeeze bottle, Graza had already gotten the more valuable thing: people knew who Graza was.

And once you have that, apparently you can put the olive oil back in the same glass bottle everyone started with.

The existence of a possible competitor tells you very little about the probability of one appearing, and even less about whether they’ll be good.

The competitor in your head has some remarkable qualities.

They never procrastinate. They never get discouraged. They have enough money. They learn quickly. They have good taste. They answer every email. They can recruit. They understand distribution. They don’t lose interest when the exciting part ends.

They are apparently willing to reorganize their entire life around something you mentioned at dinner.

I would be terrified of this person too. I just haven’t met many of them.

There are obviously businesses where copyability is a serious problem. If your entire advantage is a feature someone with much better distribution can reproduce next week, you should probably care. If the market is winner-take-most, speed matters. If a competitor can spend ten times what you can and the customer has no reason to distinguish between you, that’s a real problem.

But this describes fewer businesses than startup culture would have you believe.

For a lot of physical products, services, hospitality businesses and niche software companies, the work itself is a surprisingly effective filter.

Someone can copy the visible part of what you built.

They can’t instantly copy the hundreds of boring things underneath it.

They don’t have the supplier relationship you developed after the first one failed. They don’t know which retailer looks perfect for the brand and never reorders. They don’t know which product everyone compliments and nobody buys. They haven’t learned where the margin disappears, which email gets answered, what breaks at fifty accounts instead of five.

None of this sounds particularly defensible.

That’s because none of it is, individually. Collectively, it’s a business.

And this is the part I think gets lost in the way people talk about moats. A moat sounds like something you’re supposed to have before you begin. Some structural advantage you can point to in a deck: proprietary technology, network effects, patents, exclusive access, switching costs.

Those are real advantages.

But a lot of businesses build their advantage in the opposite direction. They don’t begin with a moat. They accumulate one.

You kept going.

You spent three years getting slightly better at something while everyone else was still deciding whether it was worth doing.

You know the buyers now. You know the suppliers. You know what customers actually come back for. You know where the business breaks. You’ve made a thousand small decisions, most of which are invisible in the finished product and annoying to relearn from scratch.

None of those things happened because you had a moat. They happened because you stayed around long enough to acquire them.

Persistence isn’t the moat. It’s how you end up with one.

Time has started doing some of the defending for you.

“Wouldn’t someone else just do it?” reminds me a little of “is it innovative enough?” You can use either question to kill a perfectly good idea from your desk.

The idea is obvious, so it isn’t worth doing. Or the idea is so good that someone better positioned will inevitably do it. Either way, you don’t have to find out.

There’s a strange assumption underneath the fear of being copied: that recognizing an opportunity and pursuing one are almost the same act. They aren’t.

Knowing that a restaurant would work in a neighborhood is not opening the restaurant. Knowing there’s room for a better product in a category is not getting it manufactured. Knowing a certain kind of software should exist is not spending two years getting people to use it.

The distance between that’s a good idea and I’m going to organize a meaningful portion of my life around making it work is enormous.

Most ideas die somewhere inside that distance.

Starting isn’t even much of a filter. Plenty of people start. The bigger separation comes six months later, when nobody particularly cares yet and the work has become repetitive. You have to send the twentieth email after nineteen people ignored you, fix the packaging problem, follow up again, and keep doing the same basic thing long enough to become unusually good at it.

A lot of businesses don’t require a genius insight. They require someone to care about a fairly ordinary thing for an unreasonable amount of time.

That’s hard to put in a pitch deck because at the beginning it isn’t an advantage. It’s just you doing something anyone else could do.

Then a few years pass. Most of them didn’t.

And you’re still here…

Someone told me recently that I was going to spend an unbelievable amount of time on building a coffee company to make almost no money.

They weren’t trying to be rude. They were looking at the economics of a small coffee company and making a fairly reasonable observation.

CPG has inventory. Wholesale margins. Shipping. Retailers who pay late. Samples that go nowhere. Bags that have to get packed and orders that have to go out. You can build a nice business and still accidentally create a job for yourself that pays worse than the one you left.

I knew all of this when I started.

The funny part is that I had spent the previous two years building software designed around exactly this problem.

Mainframe was built on a fairly simple belief: small teams should be able to do things that previously required much larger organizations. A person should be able to give software an objective and have it work across the tools where the work actually happened. Find the information. Move it somewhere else. Draft the thing. Update the system. Follow up. Notice what changed. Handle the steps between intention and outcome that normally accumulate into someone’s job.

We were building an operating system for that future.

Then I left and started selling coffee.

I can understand why this looks strange from the outside.

There is a very legible version of what someone with my background is supposed to do next. Start another software company. Join one. Raise another round. Find a sufficiently large market and apply everything I learned the first time.

Instead I am emailing grocery buyers about case packs.

But I haven’t become less interested in AI. I think I’ve become less interested in AI as the thing being sold.

Building Yodo has made that distinction much clearer to me.

There are parts of the company I want to spend an unreasonable amount of time on. I care where the coffee comes from—a part of Costa Rica I’ve been connected to for fifteen years. I care about the farms. I care about the bag and the type and which stores carry it. I care about what sits next to it on the shelf. I care about whether something we make feels inevitable once it exists.

Then there is everything required to keep a small wholesale business alive.

I have much less romantic attachment to any of that. Historically, though, the two have come bundled together.

You start a company because you care about the first category and eventually discover that much of your time belongs to the second. Then the company grows and the second category grows faster. More customers create more communication. More accounts create more follow-up. More products create more inventory decisions. More revenue creates more reconciliation, reporting, scheduling and exceptions.

Eventually you hire people to manage the work created by having more work.

This is how a ten-person company becomes a fifty-person company.

It’s also why the person warning me about Yodo had a point. Small physical businesses can produce an absurd amount of coordination relative to their size. If I intended to operate one exactly the way it would have been operated ten years ago, I would be less interested in doing it.

I don't intend to.

One of the more interesting questions AI raises has very little to do with chatbots, agents, or which model scores highest on a benchmark.

I don’t think we know yet.

A lot of the work that causes organizations to grow isn’t particularly creative. It exists because information has to move between systems and people. Someone has to notice that something happened, remember what should happen next, find the relevant context and do something about it.

Companies have traditionally solved that problem with humans.

Increasingly, they won’t have to.

This matters to me because the part I want to preserve is the human part.

I don’t want software deciding what Yodo should taste like. I don’t want it deciding which farm is interesting or whether the packaging is good or which retailer feels right for the brand. I don’t particularly want an algorithm deciding what the next company should be either.

Those are the reasons I want to build things.

I would be delighted, however, if software noticed that a retailer hadn’t reordered in forty-five days, checked what they bought last time, looked at the previous correspondence and put the right follow-up in front of me.

Or noticed an invoice was overdue before I did. Or kept track of every sample we sent and surfaced the ones nobody followed up on. Or told me on Monday morning which accounts needed attention without requiring someone to spend Friday afternoon assembling a report.

There is an enormous amount of work between those two categories.

That’s the territory I’m interested in.

For most of the history of business, increasing economic output eventually required increasing organizational size. Not perfectly, and not in every industry, but enough that we came to treat the relationship as natural. A larger business needed more people because there were more things to coordinate.

If AI absorbs enough of that coordination, the interesting outcome isn’t only that existing companies become more efficient.

Different companies become possible.

A tiny team might be able to support a business that once required twenty people. The same operating layer might eventually support several brands. The systems for orders, invoices, inventory, wholesale follow-up and reporting don’t particularly care whether the thing inside the box is coffee or olive oil.

The judgment doesn’t transfer so easily.

That’s where I want to spend my time.

I like businesses where judgment matters disproportionately. Coffee. Food. Objects. Hospitality. Things where somebody has to decide what belongs and what doesn’t, where to stop, what to make, how it should feel. Businesses where a lot of the value comes from someone caring excessively about details that don’t fit neatly into a spreadsheet.

I call them taste businesses.

Historically, owning several businesses like that sounds exhausting to me. Each one comes with its own invoices, vendors, customers to follow up with, inventory to track, information to reconcile, people required to keep all of it moving.

I’m less sure that has to remain true.

I didn’t start Yodo to prove any of this. I wanted to make coffee.

But once I was inside the business, it became difficult not to notice how much of the work surrounding the thing I cared about looked exactly like the work I’d spent the previous two years trying to automate.

That has changed how I think about bootstrapping too.

Yodo doesn’t need venture capital. Taking it would give us money to solve problems with headcount before we’ve figured out whether they require headcount at all. I’d rather find out how far we can get before organizational complexity becomes the automatic price of growth.

How much revenue can stay in a company when the organization stays small? How much ownership can the founders keep? How many operational problems can be solved before the answer becomes another salary? How much of the work can disappear without making the thing itself worse?

I don’t know… but ownership changes the equation.

A founder who owns most of several profitable companies does not need any one of them to become enormous. The businesses can grow at the rate that makes sense for them. They can stay specific. They can throw off cash. They can share infrastructure without sharing identity.

You don’t need to turn the coffee company into a technology company.

You use technology so the coffee company gets to remain a coffee company.

Once you see it that way, it’s difficult not to wonder what else becomes possible.

Maybe it’s a ten-room hotel.

Maybe it’s a specialty food brand that would have been too operationally messy for one person to bother with before.

Maybe it’s five of them.

I don’t know yet.

What I do know is that I have very little interest in spending my life making companies that are legible as ambitious to other people.

I want to make things I care about. I want to own meaningful pieces of them. I want them to make money. And I want to see how much of the machinery that traditionally sits between those things can disappear.

So yes, I may spend an unreasonable amount of time on the CPG company.

That’s partly the point.

For the last several years, ambitious people have been asking what AI company they should build. I understand the impulse. I did it.

I’m more interested now in what AI lets me build that has nothing to do with AI.

Mainframe asked whether software could make a small company as capable as a much larger one.

I still believe it can… and now I have a small company.

A lot of people come to me because they want to raise money. Usually they already have a business. Sometimes it’s doing real revenue: a physical product, a restaurant, a service people love, a few retail accounts, a beautiful brand. They’ve gotten far enough that something is clearly working, and now they want to know how much to raise and what the deck should say.

Pretty often I end up asking a different question.

Founders tend to hear that as a judgment on the business. That I don’t think it’s big enough, or good enough, or ambitious enough. What I mean is much narrower: I don’t yet understand why this particular company should take money that expects it to get very large very quickly and eventually produce an exit big enough to matter to an institutional fund.

Those are specific requirements. They are not the definition of a good company.

Say you make an exceptional olive oil from one estate in Greece. Production is genuinely limited. The bottle sells for $70, the margins are great, restaurants want it, people buy it as a gift. You can sell a few million dollars of it a year and make a lot of money.

Then someone asks how it becomes a $100 million company.

So you source from more farms. Add six products. Enter mass retail. Lower the price. Expand distribution until the thing is everywhere.

Maybe that works. But at some point it’s worth noticing that you’re engineering away the thing people were paying for. Scarcity looked like the constraint. It may have been the product.

That company could be exceptional. It could also be a terrible venture investment.

The word scalable has done a lot of damage here. People use it as though it means good, when the useful questions are much more literal. What repeats? What gets cheaper with volume? What requires another human every time revenue goes up? What is actually constrained? Does money remove that constraint? Does another location make the brand stronger or just make another location?

Consider a salon built around a colorist with an extraordinary reputation. She charges $600 a session, books six weeks out, has actors and editors in the chair, could probably open another location tomorrow.

But someone still has to do the color.

More revenue means more chairs, more colorists, more training, more scheduling, more real estate. Her reputation is part of what customers are buying. The salon scales the way services scale.

Now say the salon sells products.

That’s a different business hiding inside the first one. A bottle of gloss doesn’t need another chair every time it sells. It can move through ecommerce, wholesale, hotels, other salons, retailers. The salon may actually be an unusually good acquisition and product development engine for it: customers sit there for two hours, trust the person doing the work, and leave with something in a bag.

So which company are you financing?

If 70 percent of revenue comes from color and 30 percent from products, you can’t give the entire company the growth characteristics of the product business because it makes the deck look better. Maybe services fund the product company and the flagship salon becomes the showroom. Maybe products eventually dwarf services and *that* becomes the venture business.

Or maybe the founder ends up owning a very profitable salon and a $4 million haircare line and makes a lot of money every year.

Not exactly a tragedy.

Hospitality has the same problem. You build a bagel shop people line up for on Sundays. One location works. Three work. The deck, inevitably, has twenty locations, then fifty, then a hundred.

The company has done $1.5 million in revenue. Then the forecast goes to three, seven, fifteen, thirty, sixty.

What physically changes between seven and thirty?

Wholesale expands. Marketing scales. New locations open. International launches. More SKUs.

Okay. How?

If a luxury product becomes a $60 million company, what happened to supply? If a services company does it, what happened to labor? If it’s a restaurant group, who paid for the locations? If wholesale grew tenfold, where did the working capital come from? How much inventory did you carry? How concentrated are the retailers?

At some point the cells in the spreadsheet have to become things that happen in the world.

A spreadsheet is a description of what would have to become true. It keeps getting mistaken for evidence that it will.

Sometimes you work through those questions and find a genuinely interesting venture business underneath.

Other times you find a very good company underneath a silly forecast.

I would rather keep the company.

A founder who owns 70 percent of an unglamorous $20 million company can easily end up richer than one holding 7 percent of the exciting $200 million one. Scale is one input. Ownership is another. So are dilution, time, profit, and what your actual life looks like while you’re building the thing.

Which is why “lifestyle business” is such a strange insult.

There are restaurant groups built over twenty years, specialist firms with absurd margins, luxury companies that stayed narrow on purpose, consumer brands that grew through wholesale without raising giant rounds. Some throw off enough cash for the founder to go buy other companies.

Often the person calling these lifestyle businesses has less money than the person running one.

None of this is an argument against ambition. Venture has just been allowed to stand in for ambition for so long that people forget they’re different things.

So my fundraising advice varies more than founders expect.

Sometimes: raise the round.

Sometimes: raise much less.

Use debt. Finance the inventory instead of the company. Franchise the locations. Keep the services business profitable and let it fund the product business until the product economics prove themselves.

And often enough: don’t raise. You have a good business. Keep it.

There is nothing small about building something profitable that you substantially own.

Most founders assume they have to earn the capital. On a pre-product raise, that's mostly true. But once a company works, the arithmetic flips. Venture capital has to earn the company too.

There is a bag of my coffee on a grocery shelf in Florida right now, and there is nothing defensible about it. No proprietary process. No ingredient nobody’s heard of. Nothing I could put in a patent or a pitch deck. It’s coffee from a place I’ve had a real connection to for years, made well, in a bag people want to keep on the counter.

If you asked me whether it was innovative, the honest answer is no. And it wouldn’t help me if it were.

I think about this every time I watch a founder talk themselves out of a good business by asking whether it’s new enough. They have a product that works, that people want, that they could start selling more or less now, and instead of selling it they’re in a corner worrying about whether there’s a moat. A wedge. A defensible insight nobody else has had. The worry feels responsible. Usually it’s the thing keeping them from doing the actual work.

I understand where the worry comes from. Software rewarded innovation so disproportionately that many founders began treating it as the universal answer. Sometimes it is. A new capability, a technical breakthrough, a thing that genuinely wasn’t possible a year ago can become the entire business. The problem is that this way of thinking leaked into categories where novelty rarely determines the outcome, and founders in completely different industries absorbed the same question as the first thing to ask about any product.

Most companies aren’t software. And most products, physical ones especially, don’t win on novelty. Look at what actually succeeds in physical goods. A better version of something that already exists. The same thing in a format that fits a life it didn’t fit before: smaller, subscribable, packaged so you’d want it visible in your kitchen, sold in a channel where the right person finally sees it. A nicer olive oil is not a breakthrough. It’s a good product someone can find.

Being novel would not make one more person likely to discover my coffee, try it, or buy it a second time. What matters much more is whether the right people come across it and, once they have, whether they come back on their own.

I also think the innovation question is attractive because it lets you stay with the product. You can spend another month refining the concept, sharpening the positioning, figuring out exactly what makes it different. All of that feels like work because it is work.

Meanwhile somebody has to get the thing into a store.

That part is much less romantic. You email buyers who don’t respond. You send samples. You follow up. You figure out case packs and wholesale margins and whether you can afford free freight. You get the account and then discover getting onto the shelf was only the beginning. Now the product has to move.

We decided fairly early that our path was specialty grocery. That decision changed the product. The price a buyer will accept sets our margins. Our margins set what we can spend on sourcing. The shelf sets the bag, because the bag has about two seconds to earn a second look from someone who walked in for something else. None of that came from refining the idea of the coffee. It came from deciding, early, exactly how it would reach a person. Design the perfect product first and then go hunting for a way to sell it, and you end up with something lovely and no affordable route to a single customer.

There are brilliant physical products that never get enough distribution to find out whether people wanted them. There are also completely ordinary products doing enormous volume because they’re available everywhere their customer already shops. Availability sounds boring until you’re the person trying to create it.

Building Yodo has made me much more interested in distribution. The product can be excellent and the brand can be beautiful and none of that solves the problem of getting a second order from a store.

When founders ask me whether something is innovative enough, I usually want to know three much more boring things. Who buys it? How do you reach them without spending more than they're worth? What makes them buy it again?

When the buyer at the grocery store reviewed us, she never asked whether the coffee was innovative. She asked what it cost, where it would sit, and whether it would still be moving in six months. Those turned out to be much better questions.

I once heard my own work described as admin.

The jolt of it stayed with me, because admin was the one thing it wasn’t. What I’d actually been doing was deciding what the company built and why, in what order we hired, how information moved between people who could no longer all fit in one room, and which of a hundred small structures was producing the behavior everyone was frustrated by. But I understood why the word reached for me. Done well, from the outside, it does look like admin. That is the defining problem of operations, and the misread is expensive enough to take apart slowly.

Start with what the work actually is, because most people picture the wrong thing. Say “operations” and they see logistics — the person who keeps the trains running, sends the doc, owns the checklist. Those are outputs. They are not the work. The work is finding the invisible structure generating the behavior you can see, and changing the structure instead of nagging the behavior. A team is slow. A product isn’t landing. The same mistake keeps recurring. The amateur move is to push on the symptom with more urgency and more process. The operator treats the symptom as information and asks what setup makes this the natural outcome: which incentive is rewarding the thing we say we don’t want, which decision has no owner so it never gets made, which priority was never written down so people fill the gap with a guess.

When engineers were building things without knowing why, that wasn’t a motivation problem you fix with a speech; the reasoning lived in people’s heads and never got written down, so I built the thing that could hold the reasoning without me — roadmaps where every item had to say why it existed, and ownership maps so each decision had someone whose job it was to make it. When users couldn’t tell what our product even was, that wasn’t a capability problem you fix by shipping more; the product presented as three things at once, so the fix was to narrow it until one clear thing came through. The visible artifact at the end was almost boring each time — a doc, a map, a smaller surface. The work was the diagnosis underneath, and diagnosis leaves no evidence anyone can point at.

Most of it isn’t even diagnosis. Most of it is maintenance, and maintenance is the least narratable work there is. It’s the follow-up that keeps a decision from evaporating between the meeting and Monday. The second message that turns “someone should own this” into “you own this.” The catch, three weeks early, that two people are already building the same thing. None of these are events. Nobody remembers them, because a thing that was prevented never happened, and you cannot form a memory of an absence. It is grind in the most literal sense — the same small acts of tightening and connecting, repeated daily, none of them worth mentioning on its own, and the whole company standing on them anyway.

That is why smooth operations are nearly invisible even to the people they’re carrying. When the work is done well, the company simply runs. There is no fire, no save, no visible heroics, because the crisis that would have happened doesn’t, and you can’t applaud a disaster that failed to occur. The person who prevents it looks like they did nothing; the disaster is precisely the thing that isn’t there. The person who fights a fire they helped start looks like a hero, because everyone can see the flames and see them running toward it. So the discipline that most determines whether a company works is the one most easily mistaken, from the outside, for paperwork.

Which means the appreciation, when it finally comes, almost always comes through absence. Good operations are connective tissue, and connective tissue is invisible until it’s cut. You don’t notice the glue while it holds; you notice the gap when it lets go. The truest measure of what an operator was doing is what stops working the moment they stop doing it. Decisions that used to land start hanging unresolved. Priorities that used to be shared scatter back into private heads. Handoffs that used to be seamless start dropping things on the floor. The calm curdles into a low-grade chaos nobody can quite name, and people reach for explanations — we grew too fast, the market shifted, morale dipped — because the real explanation was never visible to begin with. The glue only becomes legible in the shape of the hole it leaves.

There is something structurally unfair in that. The better the operator, the more invisible the work; the more invisible the work, the later the recognition arrives, until it arrives too late to be worth anything to the person who earned it. You get credited in the past tense, by absence, after you’re already gone.

The cost of the misread compounds the whole time everyone’s looking elsewhere. Companies under-invest in the structural work until it’s on fire, then hire for it in a panic and call the rescue a turnaround. They reward motion over work, because motion is visible and work often isn’t. They read the calm of a well-run team as an easy problem rather than the output of someone doing the hard, invisible thing that keeps it calm. And they describe the person holding it all together as doing admin, because the better that person is, the less it looks like anything is being held.

The doc and the map are just where it gets stored. What an operator actually produces is judgment that keeps working after they’ve stopped being in the room to supply it — a company that still decides well when the founders are traveling, that still knows why it’s building what it’s building when the person who set the direction is asleep. That persistence is the product. And it’s the one product that, working perfectly, looks like nothing happened at all, because you can’t point at the chaos that didn’t occur and take credit for its absence.

The work that looks like admin is usually the work holding everything else up. The unfair part is that you rarely find out which was which until it stops.

We closed $6.75M across pre-seed and seed before we had anything to show. No product. No metrics. No data room. There was nothing to diligence except three founders and an idea.

“We raised on the idea” is the version founders tell each other at dinner, and it isn’t true. Nobody wrote a check for the idea. The idea was a paragraph. What got underwritten was whether the three of us could turn that paragraph into a company, and that is a completely different thing to sell, because there’s no slide for it. Having someone on the team who can obviously build matters more than people admit. On a pre-product round, a team that can visibly ship is the closest thing an investor has to evidence.

When you have a product, the raise is partly an evidence problem. You point at the metric, the metric argues on your behalf, and your job is to frame it. When you have nothing, the evidence can’t do any of the work, so all of it comes back to you. Every question that would normally get answered by a chart gets answered by your judgment, in real time, in the room. Before there’s a product, investors have very little to diligence besides the founding team. They’re evaluating whether you’re the kind of operator who reduces their uncertainty or adds to it.

So the raise itself became the only product we had. We treated it like one.

That meant it had a strategy, not just a pitch. A target list of exactly who we wanted and why each name was on it — not a spray of every fund with a website, but a specific set of people chosen for a specific reason, whether that was conviction in the space, tolerance for pre-product risk, or the signal their name would send to the next person on the list. And it had a sequence: who we talked to first, who we deliberately held until later, what had to be true before a given conversation opened. On a round with no metrics, the sequence is the metric. The order the yeses arrive in is the only momentum you get to manufacture. It isn’t logistics. It is the strategy.

Warm capital is the best capital there is. People who’ve already backed you have already run the experiment of trusting you and lived with the outcome. We had some of that, and it mattered. But relationships get you into the room. They don’t get you out of it with a term sheet. The campaign still has to be designed.

We got a lot of nos. That part never survives into the dinner-party version. Not soft nos, not “keep us posted” — actual zeros, over and over, from smart people, for the reasonable reason that there was no product and no metrics and therefore nothing for them to hide behind if they were wrong. A no on a pre-product round isn’t a verdict on the idea. It’s a person deciding they can’t underwrite you specifically with that little to go on, and most people, correctly, can’t.

Which is why the sequence mattered so much. When you’re going to collect a pile of zeros no matter how good you are, the only thing that keeps the raise alive is making sure they don’t compound. One no in isolation is information. Five nos in a row, in front of the wrong next meeting, is a narrative, and the narrative kills you before the idea ever gets a fair hearing. The sequencing was, in large part, a system for keeping the zeros from turning into a story. You put the people most likely to say yes where a yes does the most work, and you keep the conversations that are going to be hard away from the conversations you can’t afford to lose.

We pulled it off. $6.75M, three founders, a paragraph, and a lot of doors that stayed shut.

If you’re pre-product and about to raise, the lesson I’d take from it is this: stop treating the round as a pitch you deliver and start treating it as a thing you design. The deck isn’t the product; the raise is — the target list and the reasons behind it, the order, the way you manage the flow of information so that the inevitable nos don’t become the story. Investors on a pre-product round are primarily diligencing the founding team. The most convincing evidence you can give them that you’ll build a functioning company is that the way you ran your raise already looks like one.

Last year, someone from Apple came up to me after a talk I gave in Miami. That turned into a demo at their offices. Later, they spent time with our beta, and the feedback that came back from it changed how I thought about the whole product.

The feedback wasn’t about what the product could do. It was about what people thought it was. People couldn’t answer a simple question when they used it: what is this? Was it ChatGPT? Was it Shortcuts? A to-do app? An email client? Users kept reaching for a category to slot us into and couldn’t find one, so they made one up, and every person made up a different one.

The product was Cobot, and we’d built it as a platform for MCPs on the thesis that agents are the new apps. So the Library, looked like an app store: a shelf of agents you could browse and install. Except there was also a text chat sitting in the same product, which reads as ChatGPT. And a task surface with Auto-Do, which reads as a to-do app. Three different product categories stacked in one window, each pulling the user toward a different story about what they were looking at. We thought we were showing people a platform. They were looking at an app store, a chatbot, and a to-do list at the same time, trying to work out which one we actually were.

I didn’t come away from that thinking they liked us. I came away thinking they’d handed us a diagnosis.

The problem wasn’t that the product couldn’t do enough. The problem was that it violated the mental model people brought to it. And once I saw it that way, I couldn’t unsee what it was costing us. Every demo needed a preamble before it could produce an aha. I had to explain what the product was before I could show what it did. Onboarding carried the same tax: before a user could learn how to use the product, they first had to decide what kind of product they believed it was. And every feature we’d added, each one individually good, was competing with all the others for attention, because there was no organizing principle telling anyone what the product was for.

We had built capability and starved it of legibility.

The obvious move when a product feels unclear is to add something. A tutorial, a tooltip, one more feature that finally makes the value obvious. My instinct was the opposite. Subtract until it’s legible. Pick a wedge narrow enough that a person could understand the product in one sentence, and let everything else wait. For me, that wedge was Mail and Calendar, and a shift toward the product being proactive rather than a general-purpose surface you had to drive. Not because Mail and Calendar were the whole vision. Because they were a category people already understood, a door they already knew how to walk through.

They were also where the work actually was. Most useful agent work starts in your inbox and calendar. That’s where the requests arrive, where the commitments get made, where the follow-through lives. So it wasn’t a trade between legibility and capability. It was the one place both were true at once.

I remember where I did the real thinking on this. London, jetlagged, awake at hours that belonged to neither time zone. I’d taken a pen to the back of a receipt, and what came out was Spaces: Mail and Calendar living next to a screen where you could watch your agents work, generating the interface as they went. Nobody was doing that at the time.

The receipt became a doc, and the doc became the argument.

The problem wasn’t writing it. The problem was what it replaced. Spaces meant the to-do list wasn’t the front door anymore — it was a thing that happened inside one. And a to-do list had been the answer for long enough that arguing against it felt less like updating a view than like breaking something. I’d write the Notion doc, see what it cost, close the laptop.

A vision starts as a hypothesis, but the longer it exists, and the more publicly you’ve staked yourself on it, the less it behaves like a hypothesis and the more like an identity. We hadn’t just planned to build a platform for agents. Agents are the new apps was the banner over the front door. Evidence updates a hypothesis and threatens an identity, so the honest move, letting the feedback narrow the product, arrived wearing the costume of a failure of nerve. It wasn’t. The thing that felt like abandoning the vision was actually the thing the vision required.

The harder part was external. There was a disagreement about the direction, and writing the direction down turned out to be the easy part next to arguing for it. I thought I was responding to evidence. It took me a long time to see that we weren’t arguing about product strategy at all.

The hardest product disagreements rarely announce what they actually are. They arrive dressed as feature fights, or prioritization fights, or “we’re moving too slowly.” Underneath, people are answering two different questions and each assumes the other is answering the same one. One is about execution: how do we build the thing we decided on? The other is about identity: is this still the thing we should be building, given what we now know? Argue across those two layers long enough and it starts to feel like a values problem when it’s really a question-mismatch problem.

The failure worth fearing was never incompetence. It’s that a company can become extremely good at answering a question the market has stopped asking.

Thinking about splitting time more evenly between Miami and New York. Also entertaining a fantasy where Paris enters the rotation.

Otherwise: a lot of thinking about what to build next. A conversation with a friend last night has me turning over a software idea I didn’t expect to be turning over.

On the build:

The blocker has always been that all coffee is naturally non-GMO, but “naturally” doesn’t get you on a shelf. The paperwork does, and we didn’t have it.

What people don’t tell you is that it isn’t one certification — it’s the whole chain. Farm, roaster, packing facility, every link certified or the whole thing doesn’t count. Which is a lot when you’re small enough that your options at each step are already thin. You’re not choosing the best certified roaster. You’re finding out whether there’s one who’ll take your volume at all.

Now we have it. That opens Erewhon, Nude, Laurel Supply — the anchor accounts we want alongside the boutique grocery that’s been good to us.

Also: Happier Grocery, my favorite grocery store in the world, placed their second order. A first order means they liked the pitch. A second order means the product moved.

Not doing drip bags. Still thinking hard about RTD, and every single person I’ve talked to has told me not to. Which, unhelpfully, makes me want to do it more. Toxic trait.

On the wishlist:

_{Affiliate links below — small commission, same price.}

Crown Affair radiance oil — The one product this year I've repurchased without thinking about it. LINK

Crown Affair clips — Bought for function, kept on the desk because they look good there. LINK

Toteme flats — LINK

Open loop:

Here’s the tension I haven’t resolved.

Organic costs more to make. I’ve been trying to move our prices the other direction, because I want this to be accessible. The certification opens doors we want to be in and simultaneously makes the coffee more expensive, which is the opposite of the brand I’m trying to build.

Four options, none good. Absorb it into margin. Raise everything and normalize. Keep the conventional lot as the accessible entry and let organic sit above it. Or find a cheaper certified link somewhere in the chain.

Leaning on the third. No clean answer.

If you’ve priced a certified product against an uncertified version of the same thing — what did you actually do?

— Lou

Every difficult situation offers you the same flattering explanation: the problem is out there. The team isn’t executing. The market isn’t ready. The co-founder’s defensive, the board’s short-sighted. Sometimes that explanation is correct. The hard part is that it feels identical from the inside whether it’s correct or not.

So the useful question isn’t am I surrounded by the wrong people. It’s how would I know if I’d become the problem I keep diagnosing in everyone else. Here are the tests that actually discriminate, because they’re the ones a person who is the problem can’t easily pass just by wanting to.

The first is the disconfirmation test. When was the last time you changed your mind about something that mattered, because of something someone told you? Not a small tactical update — a real reversal, on a question where you’d been committed. If you can name a recent one, specifically, you’re probably still reachable by reality. If you have to reach back years, or you find yourself explaining why the people around you simply haven’t been right lately, that’s the signal. The person who is the problem is almost never the person who is wrong. It’s the person who has stopped being able to be shown wrong.

The second is the departure test. Look at who has left — the team, the cap table, the friendships, the advisors. Not why they said they left; people are polite on the way out. Look at the pattern in who leaves. If the people leaving are disproportionately the most capable, the most direct, the ones who used to tell you things you didn’t want to hear, that pattern is information about the environment — and you built the environment. Mediocre people leave every company. But when the ones walking out are your highest reality-tolerance people, they aren’t a retention problem. They’re the early-warning system, leaving.

The third is the translation test. Notice whether the people around you have started managing you. Softening things before they say them. Testing the temperature before raising a problem. Bringing you conclusions instead of open questions. It’s easy to miss, because it looks like respect and feels like a calm, aligned team. What it actually means is that people have learned accuracy is expensive in your presence, and started rationing it. The calmer the room has gotten, the more worth checking this is — because sometimes a quiet room isn’t aligned, it’s just done trying.

The fourth is the friction test. When someone disagrees with you now, what’s your first internal response — before the reasonable one you perform a second later? Curiosity, or the fast flicker of threat? You can’t control the flicker, but you can notice it, and its frequency is a gauge. When your relationship with reality is intact, disagreement lands as information arriving. When you’ve become the problem, it lands as an attack to be managed. The tell isn’t whether you eventually engage well. It’s what happens in the first half-second, where you can’t perform.

None of this is rare, and the people who most need these tests are often the ones who love reading things like this — they finish each one relieved, having recognized a boss, a co-founder, an ex in every line. Recognizing the pattern in other people is the comfortable version of the same activity. It’s how the tests get consumed without ever being applied.

The fifth is the calibration test, and it discriminates better than the other four, because you can’t pass it by wanting to.

Not: what opinion have you changed. Opinions are cheap and revisable and easy to narrate. The question is sharper. What were you confidently, specifically wrong about — recently — where you’d made an actual prediction? Not a preference. A forecast. This would work. This hire would land. This would take two months. The market would move that way. Something with a right answer that arrived.

If you’re still in contact with reality, you keep a rough ledger of these, because you were paying attention to whether you were right — so you can produce the misses without much prompting. The specificity is the tell: the exact call, what you expected, what actually happened, what it cost you to have been wrong. Someone who has drifted can’t produce them. Not because they’ve never been wrong, but because they stopped tracking. The forecasts got vaguer, the scoring quietly stopped, and what’s left is a general sense of having mostly been right — which is what being wrong feels like from the inside once the ledger closes.

The second half of the test is even more diagnostic: how often are you surprised? The best operators are surprised constantly, because they’re making real predictions against a world that keeps not cooperating, and each surprise is the system working — reality arriving to correct the map. Lose contact and you become progressively harder to surprise. Not because you see more, but because you’ve stopped forecasting anything precise enough to be wrong about. Nothing surprises you because nothing is being tested. An unsurprising world isn’t a sign of mastery. It’s usually a sign the instruments have been turned off.

None of these tests will tell you you’re the problem if you’re determined not to hear it. Nothing can, and that determination is itself the answer. But run honestly, they beat any amount of introspection, because they don’t route through your own account of yourself. They route through things that happened whether you narrate them or not: who changed your mind, who left, who stopped being direct, what you predicted, what actually came true.

That’s the whole point of moving the question outside your own head. Self-awareness can’t be measured by how much you think about yourself. The people who think about themselves the most are often the most insulated — they’ve just built a richer, more flattering account, and every new piece of evidence gets folded into it as being about someone else. Introspection is narration, and narration is the thing that drifts.

Reality doesn’t. The environment around you has already registered whether you’re still reachable — in who stays, who tells you the truth, how often you’re wrong in ways you can name. You’re the last one to get that information, because you’re the only one standing inside your own story. Everyone else can already see the answer.

The question was never whether you think you’re the problem. It’s whether the room would agree with you.