A startup can attract signups, earn praise and still have little evidence that customers need its product. Product-market fit metrics are useful when they test what happens after initial curiosity: do the right customers reach the promised value, return when the need arises, and make a meaningful commitment? The answer comes from a pattern of behavior, not a single score. This guide shows how to build that pattern from cohorts, usage and commercial evidence, and how to recognize when apparently impressive growth hides weak demand.
What should a product-market fit metric measure?
Start with the job the customer hired the product to do. A payroll service may create value each pay period; a travel booking service may create value only a few times a year. Daily active users would be a poor common test for both. A Y Combinator growth discussion makes this point explicitly: the relevant retention period and action depend on the product, and bookings can matter more than a generic daily-to-monthly activity ratio for travel. Source: Y Combinator
Before building a dashboard, define the target customer, the action that demonstrates value, and how often the customer needs the solution. A signup is usually an entry event, not the value event. For a team collaboration tool, inviting colleagues may be activation, while completing repeated shared work is stronger evidence. For a marketplace, a transaction matters more than browsing, but repeat transactions and a reliable experience on both sides matter more still.
The Product Validation core page covers the wider process of testing a problem, customer and solution. Here the narrower question is how to interpret the evidence after customers have had a genuine chance to use the offering.
Build cohorts before interpreting retention
Define the starting group and the return event
A cohort is a group of customers who started in the same period or under a comparable condition. Follow each group through later periods instead of mixing newcomers with long-standing customers. For example, if 100 eligible customers first received value in January and 38 completed the relevant action again in March, that cohort’s March return rate is 38%. It is an illustration, not a benchmark. The denominator, eligibility rule, time interval and definition of “return” must be recorded so that the number can be reproduced.
Do not count a background app open or accidental visit as meaningful use if it does not reflect the customer’s outcome. Segment customers with different needs: a weekly operations team should not be pooled with an occasional individual buyer. Stripe Atlas’s interview with founder Tom Blomfield recommends following signup cohorts over subsequent months to see whether users remain or disappear. Source: Stripe Atlas interview
Choose a window that fits the buying cycle
A weekly collaboration product might examine weeks after first use. Annual tax software requires a much longer window; a month without use could be entirely normal. For enterprise products, distinguish the individual user’s activity from the paying organization’s renewal decision. Newly acquired cohorts also need enough elapsed time before comparison. Calling a September cohort “retained at six months” in October would be impossible.
Look at the shape of each cohort over time and whether newer cohorts improve after product changes. A durable group of repeat users is more informative than a large opening spike followed by rapid abandonment. Even a leveling curve needs context: a tiny retained niche may validate a real problem without demonstrating a sufficiently large market. Retention can also be temporarily inflated by contracts, credits or high switching costs. Ask whether customers continue because they receive value.
Read usage and payment together
Activation and repeat value
Activation asks whether a new customer gets to the first useful result. Measure the share of eligible newcomers who complete that event and how long it takes. Then measure repeat value independently. A product can have excellent onboarding and weak retention if the result disappoints; it can also have strong repeat use among the few people who make it through a confusing setup. These are different problems requiring different experiments.
Track relevant frequency, depth or completion only where each reflects the customer’s outcome. Ten clicks are not necessarily better than one successful task. Interview retained and departing customers to explain patterns the event log cannot show. Did they solve the problem elsewhere, lack budget, fail to understand the product, or simply have no need this month? Behavior and explanation should be checked against each other, rather than treating compliments as a substitute for behavior.
Willingness to pay, renewal and expansion
For a commercial product, measure whether the intended buyer pays at a viable price and repeats the commitment. Relevant events may be a second purchase, a paid conversion, a signed contract, renewal or increased usage tied to a bill. Free usage alone cannot establish willingness to pay. Conversely, a few paid pilot contracts may show serious interest without proving that buyers will renew or that the sales process is repeatable.
For subscription software, examine both customer retention and revenue retained from the same starting customers. Expansion from some accounts can offset losses from others in revenue totals, so show the underlying customer count and changes in spending as well. Stripe’s SaaS guide distinguishes logo churn from revenue churn and describes how upgrades, downgrades and cancellations affect a customer cohort. Source: Stripe Atlas SaaS guide The SaaS churn article explains the later revenue effect of cancellations; the present framework asks what that behavior says about early product demand.
Revenue is evidence of value exchange, but it is not a complete verdict. A discounted first contract, a founder’s personal network or unusually intensive service may produce initial sales that cannot be repeated economically. Check whether customers independently continue, whether new buyers resemble the target segment, and whether the delivery cost leaves room for a durable business. The Business Model & Unit Economics page addresses that separate economic test.
A worked example: signups rise while value does not
Imagine a workflow startup with 1,000 signups in April, twice March’s total. Its dashboard celebrates growth. Yet only 300 April accounts finish a first real workflow, 90 return to finish another in May, and 25 pay after the trial. These figures are hypothetical. The useful comparisons are 300 of 1,000 reaching first value, 90 of the 300 activated accounts repeating it, and 25 of the original 1,000 paying. Keeping both the full signup base and the activated base visible prevents a favorable denominator from masking the drop-off.
Suppose May’s campaign then brings 2,000 signups but a smaller share completes or repeats a workflow. Total active users could still rise because the company has added a larger batch of new arrivals. That aggregate increase does not show that the product improved. Compare the April and May cohorts at the same age, separately for the same target customer segment and acquisition channel. If paid promotion brought less relevant visitors, the decline may be a targeting issue; if comparable users fail at the same step, investigate the product experience.
Next, interview customers who completed two workflows, those who stopped after one, and buyers who declined to pay. Check whether the value event corresponds to a real task, whether the task recurs often enough, and whether the price fits the buyer’s priorities. A sensible experiment might improve the handoff that blocks a second workflow, then compare later cohorts under the same definition. The decision is not “product-market fit achieved” or “failed” based on one month’s percentages. It is a narrower, testable conclusion about where evidence strengthens or breaks.
Adapt the framework to the business model
Consumer app: Measure the action that solves the recurring need and the proportion of users who return when that need recurs. Advertising-driven attention alone may not show a valuable or defensible product. Segment by the people the product is intended to serve, not just everyone who installs it.
Subscription SaaS: Follow activation, account or user retention, paid conversion, renewals and revenue changes by cohort. Separate seats added within an established account from genuinely new customers. The appropriate interval depends on the product’s actual use cycle and contract length.
Marketplace: A completed transaction requires a suitable match between supply and demand. Examine whether buyers can find available supply, whether suppliers receive useful demand, whether transactions complete successfully, and whether each side returns. Growth on one side alone can conceal a poor experience on the other.
Enterprise product: A successful pilot is an early signal, but procurement, deployment, adoption by the intended team, renewal and expansion may occur at different times. Track each stage and record why deals stop. Small samples and long contracts call for careful qualitative review rather than premature statistical certainty.
No universal percentage establishes product-market fit across these models. Y Combinator describes loyal and ideally paying customers as more persuasive than employee counts or a funding announcement. Its guidance also emphasizes listening to users and iterating on the underlying problem. Source: Y Combinator
What the metrics cannot prove
Metrics cannot by themselves prove market size, competitive advantage, profitability or that demand will survive a price change. A high return rate among ten handpicked customers may be encouraging but uncertain. A poor aggregate rate can hide a valuable segment. Record sample size, acquisition source, product version and incentives alongside the results.
Watch for misleading comparisons. Changing the definition of an active customer between months, excluding failed users from the denominator, or comparing a mature cohort with a new one makes progress appear where none is established. Seasonal demand, sales assistance, free credits and a major outage can also change behavior. State these conditions before drawing a conclusion.
The practical answer is to ask whether comparable target customers repeatedly reach the promised outcome and continue making an appropriate commitment, while the company can deliver it sustainably. Use cohorts to locate the pattern, payment or renewal to test commercial value, and customer conversations to understand exceptions. If one link in that chain is weak, define the next experiment instead of scaling a flattering headline metric. The Private Company Lifecycle hub places that evidence before the later questions of funding, growth and valuation.