The 40% rule needs around 100 responses. Here is what to measure instead at each stage, and the three signals that work at any size.
The standard product-market fit framework is the Sean Ellis test: survey your users, ask how they would feel if the product disappeared, and look for 40% answering "very disappointed". It is a good test. It also needs around 100 responses, which most solo founders do not have.
A PMF score built on 14 responses is not weak evidence. It is noise with a decimal point.
This page covers what to use instead at each stage, and the three signals that work at any size.
Sean Ellis established the 40% benchmark after studying a large number of startups, and the guidance is to survey users who have experienced the core of the product recently and repeatedly.
With 14 customers, four "very disappointed" answers produce a 29% score and five produce 36%. One person's mood is moving your headline metric by seven points. The number looks like data and behaves like a coin flip.
The failure is not that the score is inaccurate. It is that founders act on it — pivoting a product that was working, or scaling one that was not.
Signal 1 — The retention curve flattens. Plot the percentage of each monthly cohort still active over time. If the line falls toward zero, you do not have fit. If it drops and then flattens — at 25%, at 40%, anywhere — the people on the flat part are your fit.
This works with small numbers because you are looking at a shape, not a threshold. Three cohorts of eight customers each will show you whether the curve flattens.
Signal 2 — Unprompted specific praise. Not "great product". A customer describing exactly which part of their week changed. Unprompted specificity is difficult to fake and correlates well with dependency.
Signal 3 — The replacement question. Ask: what would you use instead if this shut down tomorrow? Answers cluster into three groups, and the distribution tells you more than any score.
| Answer | Reading | What to do |
|---|---|---|
| "I'd go back to the spreadsheet, and I'd hate it" | Dependency — this is fit | Find more people like them |
| Names a specific competitor | Substitutable — fit is fragile | Sharpen positioning |
| "I'd probably just stop" | No fit — it was a nice-to-have | Interview them properly |
| Customers | Method | Decision it supports |
|---|---|---|
| 0–20 | Replacement question + interviews | Keep going, or change the segment |
| 20–100 | Retention curve by cohort | Scale acquisition, or fix retention first |
| 100+ | Sean Ellis survey, segmented | Which segment to concentrate on |
Segmenting matters more than the headline number. An overall score of 30% that is 55% among one specific customer type is not a failing product — it is a positioning instruction. Rahul Vohra's well-known work at Superhuman took this approach: concentrate on the segment that already depends on you, and ignore feedback from people outside it.
Three things that look like product-market fit and are not.
A launch spike. Four hundred Product Hunt upvotes and eighty signups in a week measures interest in a launch, not fit. The signal arrives in week six, when you see how many of those eighty are still active. If you came from a launch, what happens after the launch is the more useful question.
Revenue without retention. Growing MRR with high churn is a treadmill that feels like traction. Check whether the customers paying you this month are the same ones who paid last month.
Praise from people who do not pay. Free users and peers in founder communities are generous. The only opinion that carries information is one attached to a card.
Product-market fit is not a binary you achieve once. It is fit between a specific product and a specific segment, and it decays when either changes. Founders who treat it as a permanent milestone stop checking exactly when it starts to slip.
The useful response is rarely a full pivot. In most cases the product is fine and the segment is wrong — which is a positioning problem, not a rebuild.
Start with the customers who did stay. Interview five of them properly, using a structure that avoids leading the witness. Then narrow everything toward whatever they have in common.
If nobody stayed, the question moves further upstream to whether the problem was worth solving. And if customers stay but never convert from trial, the issue is in trial conversion rather than fit.
The best-known framework is the Sean Ellis test: survey users asking how they would feel if they could no longer use the product, and look for 40% or more answering 'very disappointed'. It typically needs around 100 responses to be reliable.
Below roughly twenty customers, skip the score. Use the replacement question instead — ask what they would use if the product shut down tomorrow — and check whether your retention curve flattens rather than falling toward zero.
Not at small scale. With 14 responses, a single person changes the score by around seven points, so the number behaves like noise. The rule becomes useful once you can survey 100 users who have used the product recently and repeatedly.
Launch spikes, revenue without retention, and praise from people who do not pay. A Product Hunt spike measures launch interest, not fit; the real signal arrives weeks later when you see how many of those signups are still active.
Usually the product is fine and the segment is wrong, which is a positioning problem rather than a rebuild. Interview the customers who did stay, find what they have in common, and narrow toward that group before changing the product.
Bring your retention numbers and what customers said they would use instead. You get a specific read on whether to scale or narrow.
Try GhostCoach free →14-day free trial · cancel anytime · 30-day money-back on Lifetime