पाठशाला Pathshala · उत्पाद Utpād, The product · Lesson 04 · Build

What product-market fit looks like in the numbers

Product-market fit is not a feeling. It is a retention curve that flattens above zero and a survey in which forty per cent of users would be very disappointed to lose you. Both can be measured this month.

Pathshala, The Founder Library · 10 October 2026 · 8 min read

Marc Andreessen gave the term its definition in 2007 and its feeling: when you have it, customers are buying as fast as you can make the product, and when you do not, word of mouth is not spreading and deals never quite close. Nineteen years on, founders still diagnose it by feeling. The feeling arrives late and lies early. Two numbers do not.

This lesson gives you both numbers, how to compute them from data you already have, what good looks like for consumer and business products, and a worked account of how one company moved its score from 22 to 58 per cent in three quarters by treating the measurement as an engine rather than a verdict.

Three definitions, one measurement

Andreessen’s The only thing that matters defines product-market fit as being in a good market with a product that can satisfy that market. Michael Seibel of Y Combinator is blunter in a 2018 conversation: product-market fit is what happens after you build the thing customers want, and you know you have built it when they are using it in an explosive and destructive way. He cites his own earlier company, Socialcam, whose users had basically zero chance of using the app ten days after downloading it. Absolutely horrible retention, in his words, and therefore no fit, whatever the download chart said.

Both definitions reduce to the same observable: people who try the product keep using it, and the ones who keep using it would mind losing it. Retention measures the first. A single survey question measures the second. Everything else, growth rate included, is a consequence of these two or a disguise for their absence.

The forty per cent question

Sean Ellis built a customer development survey around one question: How would you feel if you could no longer use this product? with three answers: very disappointed, somewhat disappointed, not disappointed. In his own account, it becomes possible to sustainably grow a product when around forty per cent of the people who try it would be very disappointed without it. Rahul Vohra, writing for First Round Review, reports that Ellis reached the threshold by benchmarking nearly a hundred startups: companies that struggled to find growth almost always had fewer than forty per cent answering very disappointed, and companies with strong traction almost always exceeded it.

The mechanics matter. Survey people who have actually experienced the core product, not everyone who signed up; a user who never finished onboarding has no opinion worth measuring. Ellis suggests at least thirty responses before reading anything into the number and a hundred or more for confidence. Add three more questions, as Vohra did: what type of person would most benefit from the product, what is the main benefit you receive, and how can we improve it for you. The first question gives the score; the other three tell you what to do about it.

Retention curves, and where they flatten

The survey is a sample. Retention is the census. Build it as a cohort grid: one row for each month in which users first signed up, one column for each month since, each cell the share of that cohort still active. Then average down the columns and draw the curve.

Sequoia’s data science team, in their note on retention, describe the two shapes. If a curve flattens, a percentage of the users who sampled the product found value in it and are staying. If it declines continuously, it will eventually reach very few or zero users, and growth is pouring new users into a bucket that empties. The rare third shape, a curve that rises again as the product improves and network effects pull churned users back, belongs to exceptional products in their best years. The question for your grid is only the first one: does it go flat, and at what height?

Two readings from the grid. Month-one retention is about onboarding and the promise made at acquisition; it can be moved with product work in weeks. The height at which the curve settles is about whether the product does a job people need done repeatedly; it moves only when the product changes. A startup that spends a quarter optimising month one while the curve still runs to the floor has been busy and has learned nothing.

What good looks like

Benchmarks vary by category and by how “active” is defined, so treat any single number as a range. The most-cited public set comes from Lenny Rachitsky’s 2020 survey of about twenty experienced growth practitioners, combined with public company data, on what good retention is. For six-month user retention: consumer social, around 25 per cent is good and 45 per cent great; consumer transactional, 30 and 50; consumer SaaS, 40 and 70; SMB and mid-market SaaS, 60 and 80; enterprise SaaS, 70 and 90. The pattern is the useful part: the higher the switching cost and the more the product is part of someone’s job, the higher the floor must be.

For Indian consumer apps the public data is thinner and harsher. AppsFlyer’s India Festive Report 2025, covering the nine weeks around Diwali 2024 on iOS, found shopping apps averaging about 25 per cent day-one retention, with the best in the category holding 31 per cent, and even best-in-category shopping apps retaining 6.5 per cent of new users at day thirty; finance apps did better at 13.2 per cent. Those are day-thirty figures for the best performers in a festive window with heavy paid acquisition, which is the hardest possible setting, but they make the point: an Indian consumer app that holds ten per cent of a cohort at month three has done something most of the market has not.

For Indian SMB software there is no public benchmark worth quoting; the industry reports that exist name churn as the top challenge without publishing a rate. Use the global SMB figure as the bar and expect to earn it more slowly: a kirana or a small manufacturer that pays monthly on UPI can leave with less friction than an American business on an annual card, so the curve for a product that works tends to flatten later and the month-one drop tends to be steeper. If your SMB curve has not flattened by month six, assume it will not.

The Superhuman engine

The most useful public account of using these numbers to build, rather than to judge, is Vohra’s. Superhuman, an email client, surveyed its early users with Ellis’s question and 22 per cent said very disappointed. Below the line. Vohra’s response was to treat the survey as a map rather than a grade.

First, segment. He grouped respondents by their answer, tagged each with a persona from the “who would benefit most” question, and found the very-disappointed group clustered around founders, managers, executives and business development. Narrowing the analysis to those personas moved the score to 33 per cent without changing a line of code: the product had fit with a smaller market than the one it was measuring itself against. Second, listen asymmetrically. Among the somewhat-disappointed, he split those who named speed as the main benefit, the thing very-disappointed users loved, from those who did not, and opted to politely disregard the second group, since even building everything they asked for was unlikely to make them love the product. Third, split the roadmap in half: half the effort doubling down on what the very-disappointed already loved, half fixing what held back the somewhat-disappointed who shared that love. Within three quarters the score nearly doubled to 58 per cent.

Product-market fit is not a line you cross. It is a percentage you can move, once you know who it is measuring.

What is not product-market fit

Seibel’s warning, in his YC Startup Library talk on the real product-market fit, is that founders often believe they have found it when they have not, and that the belief is expensive: they hire, increase burn and optimise before they know what to build. The common false signals are worth naming. Downloads and signups are acquisition, not fit; Socialcam had plenty of them. Revenue growth with a leaking cohort is a sales team outrunning a product, and it stops the week the sales team does. Press and awards measure a story. A waitlist measures curiosity. None of these appear in the grid or the survey, and that is why the grid and the survey are the measurement.

Seibel adds a second test for a company that believes it has fit: is the growth killing you, and is it profitable? Growth that the team can comfortably handle, with unit economics that are not yet known, is not the explosive and destructive kind he means.

A monthly ritual, and a quarterly one

Monthly, on the first working day: update the cohort grid with last month’s column for every cohort, redraw the average curve, and write down two numbers, month-one retention and the height at which the curve is settling, or “not yet” if it is not. If the second number has not moved in three months while the first has, stop working on onboarding.

Quarterly: send Ellis’s four questions to every user who has used the core product at least three times in the past month. Read the very-disappointed percentage, then segment it by persona and by main benefit exactly as Vohra did. Pick the half of the roadmap that doubles down and the half that fixes. Put the percentage on the wall next to the retention curve. When both are moving in the right direction, you may hire. When the survey crosses forty and the curve is flat above the benchmark for your category, you have what Andreessen described, and the problem changes from finding fit to not losing it.

The benchmarks here are ranges reported by others, not rules; your category, your definition of active and your market will move them. The sources are below; Vohra’s article is the one to read in full.

Sources

1. Marc Andreessen, The only thing that matters (Pmarca Guide to Startups, part 4), June 2007
2. Sean Ellis, Using Product/Market Fit to Drive Sustainable Growth, GrowthHackers
3. Rahul Vohra, How Superhuman Built an Engine to Find Product Market Fit, First Round Review, November 2018
4. Sequoia Capital Data Science Team, Retention
5. Michael Seibel on starting a startup, finding product market fit, and fundraising, Y Combinator, November 2018 — See also his YC Startup Library talk, The real product-market fit (ycombinator.com/library/5z-the-real-product-market-fit).
6. Lenny Rachitsky, What is good retention?, June 2020
7. AppsFlyer and Meta, India Festive Report 2025 (iOS retention, October–December 2024)