The observation
Ask a founder how they know they have product-market fit and the honest answer is often a shrug dressed up as confidence. The wiki's own Finding Product-Market Fit theme records Eric Ries's version of the same shrug turned into a rule: if you are asking whether you have it, you do not have it. That is a fine test after the fact. It is useless the week you actually need an answer.
Sean Ellis needed an answer earlier than that. Before he coined 'growth hacking,' before the ICE framework, he was trying to get honest feedback out of senior managers at Xobni who were reliably too harsh on ordinary satisfaction questions. He changed the question rather than the audience: instead of asking how satisfied people were, he asked how they would feel if the product simply disappeared. He tried it again at Dropbox, then shared it with enough Silicon Valley companies that a pattern emerged — companies where 40% or more of users answered 'very disappointed' tended to have found real product-market fit; companies below that line tended not to have.
The counterintuitive core
The trick is not the survey. It is the direction the question points. 'How satisfied are you?' invites politeness — people round up, especially about something they use every day and have no strong reason to criticise. 'How would you feel if you could no longer use this?' invites loss aversion instead: people are far more honest about what they would miss than about what currently pleases them. Ellis built the entire test on that asymmetry — ask about the gap the product would leave, not the glow it currently gives.
The wording matters as much as the arithmetic. Respondents choose 'very disappointed,' 'somewhat disappointed,' or 'not disappointed,' and the whole method rests on refusing to average the three. As Ellis puts it: 'Just ignore the people who say they'd be somewhat disappointed. They're telling you it's a nice to have. If you start paying attention to what your somewhat-disappointed users are telling you, maybe you're going to dilute it for your must-have users.' The 40% is not a grade on a curve. It is the size of the group that has already revealed, honestly, that losing the product would cost them something.
What it means in practice
- Survey the right users, not the easy ones. Ellis restricts the sample to people who have activated — used the product at least twice — and used it within the last one to two weeks. Homepage visitors, demo watchers, and long-churned accounts all dilute the signal rather than sharpening it.
- Segment instead of averaging. Rahul Vohra's PMF Engine turns the raw score into a roadmap: among the 'somewhat disappointed,' separate those for whom the product's main benefit already resonates from those for whom it doesn't — and ignore the 'not disappointed' group entirely, since their feedback points away from fit, not toward it.
- Build for the fans and the near-converts, half and half. Ask the 'very disappointed' what they love most and double down on it; ask the resonating 'somewhat disappointed' what's still missing and fix that. Both moves raise the score by construction — nobody's roadmap is built around pleasing the people who were never going to be won over.
- When the score is borderline, chase the peak, not the average. Jag Duggal's bullseye cohort method: find the small sub-segment already scoring 70%+, work out exactly what it shares, and expand it. Nubank's Payments Assistant launched with a borderline score in the hundreds of thousands of monthly actives; the bullseye turned out to be users with four or more bill commitments across two or more payment rails, and building for that pattern took the product past ten million monthly actives over roughly fifteen months.
- Treat the number as a gate, not a report card. Nubank raised its own bar to 50% — adjusted upward for a cultural tendency toward optimistic survey answers — and does not scale a product until it clears that line. The score stops being something you check and becomes something that decides what you're allowed to do next.
The honest limit
Uri Levine, co-founder of Waze, cuts underneath the whole method: 'Product market fit have one metric. One metric. That's it. Retention. That's really simple. If you create value, they will come back. If they're not coming back, that means that you are not creating value.' On his account, every other number — downloads, survey scores, revenue — is a leading indicator of intent, not proof of fit.
Ellis's own material concedes the same point from the inside. A high 'very disappointed' score driven by switching costs — the friction of leaving, not the value of staying — is structurally different from one driven by genuine utility, even though the survey cannot tell the two apart on its own. A stated feeling about an imagined loss is not the same evidence as a person who actually came back this week. The survey is still worth running: it is faster to collect than retention data, it segments a user base that raw retention curves treat as one lump, and it tells you which users to build for before you have enough behavioural data to know for certain. But it is a demand-side proxy for fit, not the fit itself — use it to steer the roadmap, and let retention be the thing that actually declares victory.
Go deeper
The single best source for seeing the score turned into a working system is Rahul Vohra on Lenny's Podcast — watch it here, or read the episode.