How founders answer what is product-market fit

Product-market fit is sustained evidence that a specific group of customers repeatedly gets enough value from a product to keep using it, paying for it when payment is part of the model, and pulling other people toward it. It is not a mood inside the company. It is not a compliment from an investor, a crowded launch, or a revenue line that one heroic founder dragged upward by hand.

The phrase gets abused because founders want a graduation ceremony. Fit is better treated as a working conclusion with boundaries: this product fits this market segment, for this job, through this buying and usage pattern, under these conditions. Name those boundaries and you can test the conclusion. Leave them vague and nearly any encouraging event can masquerade as proof.

What product-market fit actually means

Product-market fit means a product can satisfy a real market well enough that customer behavior starts carrying part of the company’s weight. Marc Andreessen’s 2007 essay, "The only thing that matters," credits Andy Rachleff with the formulation and defines fit as being in a good market with a product that can satisfy it. That definition contains two separate tests: a real group must urgently need something, and your product must solve enough of that need.

Founders often collapse those tests into "people like the product." A beautifully designed product can delight ten handpicked users in a market too small to support the company. A large market can exist while your product misses its purchase criteria, workflow, price, or trust requirements. Product quality, market demand, and company viability overlap, but they are not interchangeable.

The unit of fit is therefore not "the startup." It is the pairing of a product and a segment. A bookkeeping product might fit independent US retailers with five to twenty employees and fail with venture-backed software companies. The same product might fit the user but not the buyer if employees adopt it while finance refuses the contract. Saying "we have fit" without naming the customer, problem, and repeated value event hides the information your team needs.

Fit also has a time dimension. Customers must reach value more than once, or receive a durable outcome they would deliberately buy again. Repeat use makes sense for payroll software. Renewal, expansion, or continued deployment makes more sense for annual enterprise software. A home-buying service may see the same person only once in years, so successful transactions, referrals, and willingness to use the service for a later move matter more than weekly activity.

This is why no universal number certifies fit. The correct evidence depends on how often the underlying problem occurs and who makes the buying decision. A founder’s first job is to state the claim precisely enough to be wrong: "Operations leads at US dental groups with ten to fifty locations complete payroll reconciliation every two weeks, keep using us for six months, and renew without founder intervention." That sentence can meet data. "Customers love us" cannot.

Demand shows up as costly customer behavior

The strongest signals require customers to give up something scarce: time, money, reputation, workflow stability, or political capital. Praise is cheap. A customer who imports sensitive data, trains colleagues, asks procurement to approve a contract, or introduces you to a peer is taking a cost because the expected value exceeds the hassle.

Look for behavior that continues after the novelty wears off. Good signs include customers returning on the natural cadence of the problem, completing the core action without reminders, renewing when they have a real chance to leave, and complaining urgently when the product fails. Complaints can be stronger evidence than polite satisfaction when they come from people whose work now depends on the product. Silence from an inactive account is not satisfaction.

Pull also changes conversations. Prospects describe the problem before you teach them your vocabulary. They ask how soon they can use the product, whether it handles a specific constraint, and what switching requires. Existing users request improvements around the core job rather than unrelated features. Referrals arrive with a clear sentence about who should use the product and why.

Operational strain is a lagging signal, not a definition. Andreessen described customers buying as fast as a company can serve them, usage growing quickly, and sales and support hiring racing to catch up. That picture is memorable, but it favors visible, fast-growing software companies. A narrow B2B product can have convincing fit before reporters care or inbound volume overwhelms the team. Use his account as a description of market pull, not a checklist of startup theater.

Watch what happens when you remove founder force. If every renewal needs a personal rescue call, every deal needs a custom roadmap, and every user needs manual prompting, the company may have a skilled founder-led service rather than a repeatable product. Early manual work is normal. The warning is that the manual work substitutes for customer value instead of helping customers reach it.

Retention must match the natural use cycle

Retention is the cleanest behavioral test when customers should receive value repeatedly. Y Combinator partner Gustaf Alstromer has argued that cohort retention is the best way to measure fit because actions carry less bias than survey answers. The useful chart groups customers by when they started and then asks whether each group continues the core value action at the interval the product promises.

Choose the event before you draw the chart. "Logged in" is usually weak because a customer can log in while failing to accomplish anything. For invoicing software, the event might be sending a paid invoice. For recruiting software, it might be moving a qualified candidate through a hiring stage. For a marketplace, measure successful transactions on both sides, not visits.

Choose the interval from the customer’s problem, not from the analytics tool’s default. Daily retention makes a monthly close product look broken. Monthly retention can conceal the collapse of a daily habit. For annual contracts, look at recurring in-product value during the year and renewal when the contract ends. Contract retention alone can lag dissatisfaction by months because switching is expensive.

A simple cohort readout is enough to prevent several self-deceptions:

• January: 40 started, 26 reached core value, 18 returned at cycle 2, and 15 returned at cycle 4.
• February: 44 started, 31 reached core value, 23 returned at cycle 2, and 21 returned at cycle 4.
• March: 51 started, 39 reached core value, 31 returned at cycle 2, and 29 returned at cycle 4.

The absolute values here are illustrative, not benchmarks. Read the shape. More new customers reach value, later cohorts retain a larger share, and each cohort begins to settle above zero. A curve that keeps falling toward zero says the product has a leaky value proposition, even if acquisition makes the total active-user chart rise.

Segment before declaring victory or failure. A flat overall curve can combine a strong pocket and a weak one. Break cohorts by customer type, use case, acquisition source, plan, and the person who initiated the purchase. Do not slice until every cell tells a flattering story. The purpose is to find a segment with coherent behavior that you can describe and reach again.

Early evidence can be credible without being large

An early team can assess fit before it has statistically comfortable data, but it must expose uncertainty instead of laundering a small sample into certainty. Ten retained design partners may justify another product cycle. They do not justify a claim about an entire market.

Use an evidence ledger for every active customer. Record the segment, triggering problem, promised outcome, first value date, repeated value event, price and discount, founder labor required, current status, and the customer’s own reason for staying or leaving. This artifact is less glamorous than a dashboard and much harder to manipulate.

For each customer, classify the evidence:

• Observed: the customer completed the core action, paid, renewed, expanded, referred someone, or left.
• Reported: the customer described pain, value, alternatives, objections, or disappointment.
• Inferred: the team interpreted behavior without confirmation.
• Induced: a discount, personal favor, investor introduction, or custom work changed normal behavior.

Keep those labels visible. A letter of intent is reported intent until money changes hands. A pilot is induced evidence when it is free and staffed by your team. Five weekly uses are observed behavior, but they prove little if the problem occurs once and the users are merely testing. The ledger forces the team to argue about evidence quality rather than the emotional importance of a logo.

Small samples also make losses unusually informative. Interview people who activated and then stopped, prospects who completed a serious evaluation and chose an alternative, and customers who renewed reluctantly. Ask for the last concrete episode: "What were you trying to finish? What did you do instead? Who cared about the result? What happened after you left?" General opinions produce feature ideas. Specific episodes reveal whether the problem, segment, or product is wrong.

Good-looking numbers can be false positives

Most false positives measure attention, access, or founder effort while pretending to measure durable value. They feel persuasive because the number moves quickly and outsiders recognize it. They fail because the customer has not yet paid a meaningful cost or returned for the outcome.

Treat these signals with suspicion:

• Signups and waitlists mix curiosity, social support, bots, and real demand.
• Launch traffic measures distribution on one day, not repeated value.
• Total active users can grow while every cohort retains worse than the last.
• Pilot logos can reflect a buyer’s appetite for experiments rather than intent to deploy.
• Revenue can come from discounts, services, one large customer, or contracts customers regret.

Net Promoter Score has a similar limitation. It captures a stated likelihood to recommend, shaped by respondent selection and brand feeling. It can help track sentiment within a stable customer group, but it does not replace actual referrals, retention, or renewal. A high score from twenty friendly beta users does not settle a market question.

Growth rate needs decomposition. Write new revenue as new customers plus expansion minus contraction and churn. Then tag how much required founder involvement, custom work, or temporary discounts. Two companies with the same monthly revenue growth can face opposite realities: one compounds retained customers, while the other replaces departing accounts through exhausting sales.

Fundraising is not customer evidence. Investors may finance a strong team, a credible market thesis, technical progress, or the option value of an uncertain bet. Press, awards, social followers, conference invitations, and famous advisors can widen access. None proves that customers repeatedly receive value.

The most dangerous false positive is a handful of passionate customers who all want different products. Their enthusiasm makes the team feel close. Their requests pull the roadmap in incompatible directions, and each account needs a different sales story. That is bespoke demand. Fit starts to appear when customers with similar attributes hire the product for the same job and accept substantially the same product.

Surveys diagnose fit but do not award it

The Sean Ellis survey is useful when it helps a team identify who cares and why, not when a founder treats 40% as a certificate. Ask active users how they would feel if they could no longer use the product, with "very disappointed," "somewhat disappointed," and "not disappointed" as the main choices. Then ask who benefits most, what main benefit they receive, and what would improve the product.

Rahul Vohra’s account of applying this method at Superhuman made the process concrete. He reports that Ellis formed the 40% rule after comparing nearly one hundred startups: companies above that share of "very disappointed" users tended to have strong traction, while those below it often struggled. Vohra then segmented respondents, studied what the most enthusiastic users valued, and used objections from the next group to guide the roadmap.

The method has sharp limits. The 40% threshold came from an empirical comparison, not a law of markets. Results change with whom you invite, who responds, whether they reached the core value, and how recently they used the product. Surveying every signup understates fit because many never experienced the product. Surveying only friends, power users, or customers you personally rescued overstates it.

Run the survey with a written eligibility rule. Include users who had a fair chance to receive the core value and used the product recently enough to remember it. Report the eligible population, response count, response rate, customer segments, and exact wording beside the result. If the sample is small, publish the count with the percentage. "8 of 18 eligible respondents were very disappointed" tells the team more than a triumphant "44% PMF score."

Read the open answers before debating the score. The "very disappointed" group tells you the benefit worth protecting and the segment that feels it. The "somewhat disappointed" group can reveal fixable barriers, but do not build every requested feature. Look for changes that deepen the same core benefit for the target segment. Requests that serve another job or buyer belong in a different hypothesis.

Pair the survey with behavior. If users say they would be very disappointed but rarely complete the value event, find out whether your event is wrong, the product is aspirational, or respondents are being kind. If retention is strong but survey sentiment is muted, switching costs or obligation may be trapping customers. Neither mismatch deserves a victory label.

Revenue proves willingness to pay, not the whole fit

Revenue is necessary evidence for a product that expects customers to pay, but its quality matters more than its mere existence. A dollar can prove that someone crossed a purchase threshold. It cannot tell you whether the purchase repeats, whether delivery costs swallow the margin, or whether the buyer would choose the product without a founder’s relationship.

Inspect revenue by cohort and source. Separate recurring product revenue from setup fees, consulting, reimbursed costs, and custom development. Track renewal and expansion at the account level. Record discounts and unusual contract terms. If one customer supplies most of the revenue, you have evidence of one customer’s demand and a serious concentration risk, not broad fit.

Paid pilots are stronger than free pilots because the buyer must secure budget, but payment alone can still purchase learning or executive attention. Define the conversion event before the pilot starts: deployment to the intended users, completion of the core workflow, and a follow-on contract under standard terms. Otherwise a team can call the pilot successful because everyone enjoyed the meetings.

Price tests belong before you feel ready. Ask for money when the customer can understand the promised outcome, then watch the objection. "Too expensive" can mean the problem is minor, the buyer lacks authority, the value arrives too late, or your product loses to a cheaper alternative. Dropping the price immediately erases that diagnostic. Ask what budget would fund the purchase, what the customer compares it with, and what result would make the price easy to defend.

Some products monetize indirectly or much later. In those cases, do not invent a fake willingness-to-pay test. Identify the scarce commitment that matches the model, such as repeated contribution, completed transactions, or attention that survives novelty. Still write down how that behavior could support a viable business. User love and a workable company model are different questions, and a founder needs honest answers to both.

Find which side of the pairing is failing

When fit is absent, diagnose the market, product, and path to value separately before rebuilding everything. A weak retention number tells you that value does not persist. It does not tell you why.

Start with the market hypothesis. Interview customers around real past behavior, not a pitch for your idea. Did the problem happen often enough? Did it cause a consequence someone cared about? Did a buyer have budget and authority? What alternative did they use? If the answer is "we would like this someday," the pain may be mild, infrequent, or owned by nobody. A better interface will not create urgency.

Next inspect the product promise. Customers may have the problem but reject your approach because it misses a required workflow, creates risk, arrives too late, or demands a behavior change larger than the benefit. Compare retained customers with churned ones. Find the earliest point where their paths diverge and listen for a shared reason. One missing capability can block a coherent segment; ten unrelated requests usually signal a segment problem.

Then inspect the path to value. A product can solve the right problem and still lose customers before they experience it. Measure the steps between commitment and first meaningful outcome. Watch customers attempt them without coaching. Remove setup that exists for your internal convenience, clarify the promise, and shorten time to value. Do not call every onboarding failure a product-market-fit failure, but do not excuse it either: value customers cannot reach is not delivered value.

Distribution comes last in this diagnosis because acquisition can hide weak fit. If qualified prospects understand the promise, activate, retain, and pay, but too few encounter the product, you may have a channel problem. If unqualified traffic converts poorly, buying more traffic will make the dashboard busier without answering the product question. Y Combinator’s growth guidance is blunt on this point: heavy acquisition before retention can burn cash while customers keep leaving.

Write a one-page diagnosis with four lines: target segment, recurring problem, core value event, and largest observed break. Attach evidence for each line and mark assumptions. Teams often discover that they have been arguing about features while holding different definitions of the customer.

Change one major assumption at a time

The fastest route toward fit is a sequence of explicit bets, each designed to distinguish among causes. Random feature shipping feels productive but destroys the evidence trail. Choose the narrowest change that could alter customer behavior, define the expected signal and decision date, and keep the affected customer group visible.

A useful experiment record looks like this:

1. Claim: finance leads at multi-location clinics abandon setup because importing historical payroll data takes too long.
2. Change: offer a guided importer that completes the job in one session without custom engineering.
3. Evidence: compare qualified customers who reach the first completed reconciliation and return for the next pay cycle.
4. Decision: keep the importer if activation and second-cycle use improve without increasing manual hours per account.

This is not a universal template for analytics. It prevents a common postmortem: the team shipped six features, changed price, switched channels, and recruited a new segment, then could not explain why the numbers moved.

Change the segment when one group has a frequent, costly problem and another does not. Change the promise when customers care about a different outcome than the one you sell. Change the product when the target group wants the promised outcome but your approach fails a repeated requirement. Change pricing or buyer when users receive value but nobody with authority can justify the purchase. Change distribution when retained, paying customers share recognizable traits but your current channel rarely reaches them.

A pivot does not need theatrical reinvention. It can narrow the market, move to the adjacent user who already pulls the product into work, or remove a broad use case that dilutes the core. Preserve what the evidence supports. Throwing away retained customers because a new idea sounds larger is not rigor; it is boredom.

Set stopping conditions as well as success conditions. Decide how many qualified attempts, over how many natural usage cycles, you will fund before revisiting the premise. The exact count depends on sales cycle and runway, but the decision must arrive before exhaustion makes it for you. Continuing because the team has already spent a year is sunk-cost reasoning, not persistence.

Scale after the evidence survives less founder effort

Scale when a defined segment repeatedly reaches value, retention settles at a viable level for the category, payment behavior supports the model, and acquisition can increase without the founder recreating every result. Fit does not require a finished product or effortless sales. It requires enough repeatability that adding customers tests a known engine rather than pours traffic into an unexplained leak.

Reduce founder involvement in stages and watch what breaks. Let another person run discovery from the written segment definition. Standardize onboarding for a small cohort. Ask customers to renew through the normal process. Test whether referrals describe the same benefit you think you sell. Each handoff reveals knowledge that lived in the founder’s instincts instead of the product and operating system.

Do not wait for complete certainty. Markets move, competitors respond, and customer expectations change. A team can lose fit after finding it, especially when it expands into a new segment or replaces a beloved workflow. Keep cohort retention, renewal quality, time to value, and the core value event on the operating review after growth begins. Aggregate growth can conceal decay in newer cohorts for a long time.

Peer review helps because founders are unusually good at explaining away their own evidence. In Sisters, a founder can ask women who have built and sold products to challenge a segment definition, interview notes, pricing test, or retention chart. The useful question is not "Do you like my idea?" It is "Which claim in this evidence would you refuse to bet the next six months on?"

Fit rarely arrives as one cinematic moment. It becomes the least strained explanation for a consistent set of behaviors: the same kind of customer reaches the same value, returns on the right cadence, pays or commits in the way the model requires, and brings in others without needing a new story each time. Until those facts agree, keep the claim narrow and keep changing the assumption that the evidence actually contradicts.