How Founders Actually Measure Product-Market Fit

Product-market fit has one famous number attached to it, and the number is a survey result, not a financial metric. The 40% test is useful, cheap and widely misread. Retention curves are slower, harder to fake and the only evidence that actually settles the question. Here is where each one came from, what it does and does not prove, and which leading indicators are worth watching while you wait for the curve to resolve.

The 40% test, verified

The test belongs to Sean Ellis, the growth lead behind Dropbox, LogMeIn and Eventbrite. The question, as Ellis states it in his own April 2019 write-up, is:

“How would you feel if you could no longer use [ProductName]?” with answers a) Very disappointed, b) Somewhat disappointed, c) Not disappointed, d) N/A I no longer use [ProductName].

Ellis’s claim is narrow: “it becomes possible to sustainably grow a product when it reaches around 40% of users who try it that would be ‘very disappointed’ if they could no longer use it.” That is a floor for sustainable growth, not a certificate of fit.

The 40% figure is a pattern observed across startups Ellis surveyed, not a statistically derived threshold. The account in First Round Review’s profile of Superhuman’s PMF engine describes it this way: companies that struggled to find growth almost always scored under 40% “very disappointed”, while companies with strong traction almost always cleared it. Rahul Vohra’s team ran the survey on Superhuman and scored 22% in the summer of 2017. After filtering the results to the personas that dominated the “very disappointed” group, the score for that segment was 33%. Three quarters of focused product work later it reached 58%.

Four ways founders get the survey wrong

1. Surveying the wrong population. Ask people who have actually experienced the core value of the product, not everyone who signed up. A cohort padded with tyre-kickers gives you a low score that tells you nothing.
2. Not segmenting. Superhuman’s whole method was to ignore the aggregate and look at who the disappointed users were. A 25% aggregate that is 55% inside one persona is a stronger signal than a 42% aggregate spread evenly across five.
3. Running it once. The score is only useful as a time series against a fixed question and a fixed cohort definition.
4. Treating it as sufficient. Tristan Kromer’s critique of the test uses his own startup, StartupSquare, as a documented false positive: it scored above 40% while demonstrably lacking product-market fit, because respondents were rating the promise of a solution rather than a product that worked. Kromer notes that Ellis himself acknowledged the test was not enough on its own.

The asymmetry is the practical takeaway. A score well below 40% is strong evidence against fit. A score above it is weak evidence for fit.

The retention curve is the real test

Brian Balfour laid out the argument in The Never Ending Road To Product Market Fit, published in December 2013. Plot the percentage of each signup cohort still active at month 1, 2, 3 and onward. If the curve declines and then flattens, some group of people has found the product indispensable. If it heads to zero, nobody has.

Balfour’s own phrasing is deliberately conditional: “IF it flattens off at some point, you have probably found product market fit for some market or audience.” His broader framing is that fit is not a point you cross but “a series of tests and check points that increase in difficulty,” and that “with out retention, accelerating growth is meaningless.”

Two mechanics matter when you build the chart. Define active as the action that delivers the product’s value, not as a login. And plot cohorts separately rather than blending them, because a blended curve mixing improving cohorts with old ones can flatten for reasons that have nothing to do with the product.

Cohort analysis: what to actually look at

• The flattening level, not the slope. A curve that settles at 8% and a curve that settles at 45% both flattened. Only one is a business.
• Cohort-over-cohort improvement. Is the January cohort’s month-3 retention higher than October’s? That is the signal that your product work is landing.
• Retention by acquisition source. Paid social cohorts routinely retain at a fraction of organic ones. If your flattening comes entirely from one channel, your fit is with that channel’s audience.
• Revenue retention separately from user retention. In B2B they diverge constantly, because seat expansion inside retained accounts can mask logo churn.

Signals versus vanity metrics

The test is whether the metric can go up while the business gets worse. Total registered users, cumulative downloads, page views, press mentions, funding raised, LinkedIn follower count and total revenue booked all pass that test easily, which is why they are vanity metrics. Cohort retention, paid conversion by cohort, organic share of new accounts, net revenue retention and payback period do not.

Two current data points show why the distinction has teeth. ChartMogul’s 2025 retention report, covering roughly 3,500 software companies with data through September 2025, found median net revenue retention of 82% for B2B SaaS against 48% for AI-native companies, with median gross revenue retention of 40% for AI-native products. A company in that group can post spectacular signup and revenue growth while every cohort quietly evaporates. The same report found gross revenue retention of 70% for products above $250 per month against 23% below $50, which means a low-priced product can clear 40% on the Ellis survey and still have no durable business underneath it.

Leading indicators by business type

Retention data takes months to accumulate, and longer in B2B where annual contracts hide the truth for a year. These are the indicators that move first.

The benchmark column comes from Lenny Rachitsky’s June 2020 retention study, compiled with Casey Winters from roughly 20 growth practitioners and investors. Treat the figures as order-of-magnitude reference points, and note that they predate the current wave of AI-native products, where the ChartMogul data suggests the achievable numbers are currently much lower.

One more indicator cuts across all of these. In Rachitsky’s January 2020 survey of founders and investors on recognising fit, the recurring quantitative markers were retention flattening, an Ellis score above 40%, enterprise sales yield above 1.0, and roughly half or more of new accounts arriving through direct and organic channels. That last one is the cheapest to check and the hardest to fake: if you stop spending on acquisition and signups keep arriving, something is pulling.

The instrumentation to put in place before you need it

Three things, in this order. Define one activation event that represents the product delivering its value, and log it from today. Build a cohort retention table by signup month against that event, with a second version split by acquisition channel. And run the Ellis survey quarterly on users who have completed the activation event, keeping the wording and the cohort rule constant so the series is comparable.

Then hold both results at once. If the survey says 45% and the curve is still heading to zero at month six, believe the curve. If the survey says 25% but one persona inside it says 55% and that persona’s curve is flat, you have found your market and you are currently selling to the wrong people.

Sources

• Sean Ellis / GrowthHackers — Using Product/Market Fit to Drive Sustainable Growth
• First Round Review — How Superhuman Built an Engine to Find Product Market Fit
• Kromatic (Tristan Kromer) — False Positives and Product-Market Fit
• Brian Balfour — The Never Ending Road To Product Market Fit
• Lenny’s Newsletter — What is good retention?
• Lenny’s Newsletter — How to know if you’ve got product-market fit
• ChartMogul — The SaaS Retention Report: The AI Churn Wave (2025)