The product-market fit survey asks one question: how would you feel if you could no longer use the product? Forty percent or more answering "very disappointed" is the line. The question is easy. Reading the answers honestly is the hard part: segmenting them, refusing to call a result on eight replies, and treating "somewhat disappointed" as information rather than a near miss. This guide covers the question, the reading, and the second check that confirms it.
The 40% test
Sean Ellis proposed the question after running growth at several companies. Ask active users: "How would you feel if you could no longer use the product?" Give four answers.
- Very disappointed
- Somewhat disappointed
- Not disappointed
- I no longer use it
Leave out the people who no longer use it. Count the share of the rest who said very disappointed. Forty percent or more is the line. Ellis observed that companies above it found growth easier than companies below it.
The line is a heuristic, not a law. Its value is that it asks about loss, which people feel more precisely than they feel enthusiasm. "Would you recommend us?" invites politeness. "How would you feel if it was gone?" invites the truth.
Product-market fit matters because its absence is the most common reason software companies stop. CB Insights analysed 385 venture-backed shutdowns and found 43% cited poor product-market fit (CB Insights, 5 Mar 2026). The survey is the cheapest way to find out early.
How to ask it
Keep the survey to the one question, one optional follow-up and an optional email.
Ask the question word for word
"How would you feel if you could no longer use [product]?" Do not soften it or add context. The four answers, in the order above.
Add one open follow-up
"What's the main benefit you get from [product]?" This is where the useful words come from. Keep it optional and short.
Ask for an email, optionally
With an email you can match an answer to a paying account and a segment. Without it you still get the share. Say what you will do with it.
Send it to active, paying users
People who used the product recently and pay for it. Trials and churned users answer a different question. Ask each person at most once in ninety days.
Send it yourself
From your own email, in your own words, with the link. A survey from the founder gets answered. A survey from a tool gets deleted.
Thirty seconds to answer. If your survey takes longer, you have added something that will cost you replies.
Segment it
One blended number hides the result. A product can score 30% overall and 55% among one kind of customer. That 55% is the segment to build for, and the blend was telling you to give up.
Define segments before you read the answers, from facts you hold: plan, company size, role, how they found you. Match each answer to its account by email, then read the share per segment. Each segment needs its own count of answers, which is why segmenting makes the small-number problem worse before it makes the result better.
Read the main-benefit answers by segment too. If two segments both score high but name different benefits, you have two products in one, and your roadmap should know which one it serves this quarter.
Read small numbers honestly
This is where most founders fool themselves. Eight people answered, three said very disappointed, and 37.5% rounds up to "nearly there". It does not. One more answer moves the share by over ten points.
Use thresholds and say them out loud.
- Under ten answers in a segment: "Not enough data." No share, no direction. Send more invitations.
- Ten to twenty-nine: a direction. Show the share with its range and the n beside it, and read it as "probably above" or "probably below", never as a verdict.
- Thirty or more: call it. Forty percent or more holds; under forty does not.
Always show n and a range beside the share. "41% (n = 34, range 29% to 53%)" is a true statement. "41%" alone is a false one, because it hides how much it could move. A tool that gives you a verdict from twelve answers is doing you harm.
One more rule: count each person's latest answer within ninety days, once. A user who answered twice is not two users.
What to do with "somewhat disappointed"
The temptation is to add them to the very-disappointed group and cross the line. Resist it. The group has a different use.
"Somewhat disappointed" users are the closest to loving the product who do not yet. Their main-benefit answers tell you why. Three patterns:
- Same benefit as the very-disappointed group, weaker. They get the thing, less of it. Something in the product is in their way. Find it in their words.
- A different benefit. They use the product for something else. That may be a second segment with its own fit, or a distraction. Check how many there are and whether they pay.
- No clear benefit. They are polite not-disappointed users. Do not build for them.
Treat each main-benefit answer as a piece of feedback and file it with the rest. The "somewhat" group often names the next feature more clearly than the fans do, because the fans have stopped noticing what is missing.
Retention by segment, the second check
The survey is what people say. Retention is what they do. Run both, by segment, and believe them when they agree.
The simplest retention check: of the accounts in a segment that started paying, what share still pays in their third month, and does the curve go flat after that? Sixty percent at month three, holding flat, is a reasonable line for a subscription product. A curve that keeps falling means people try it and leave; the survey score among the ones who stayed will flatter you.
A second check for products past their first year: do the accounts that paid twelve months ago pay at least as much now, in total? New accounts do not count. This is net revenue retention, and 100% or more means expansion beats churn.
Read both checks per main segment, with n and a source beside each number. The same thresholds apply: under a minimum, say "not enough data" rather than a verdict. A segment can hold on the survey and fail on retention. That combination usually means people love it and then stop needing it, which is a pricing or packaging question, not a product one.
Products paid per use
If you sell per use (an election, a report, a job) rather than by subscription, the survey works unchanged. The retention check does not. Subscription analytics will read your business as "MRR $0" and conclude you have no customers.
Read it differently. "Segments stay" becomes repeat purchases: of first-time payers in a cohort, what share paid again within a set number of months. "Expansion beats churn" becomes revenue year on year: did the accounts that paid in the twelve months before pay at least a set share of that in the twelve months since. New accounts still do not count. The thresholds and the n rules stay the same.
If you sell subscriptions and one-off extras, keep them apart: MRR stays MRR, and extras go on their own line. Blending them makes both numbers wrong.
Related guides: The evidence ladder · Customer interview questions · Vibe coding without building the wrong thing
Do this in MindBacklog
Grow reads your own Stripe and your survey, by segment, and says when the numbers are too small. Grow: PMF health by segment, revenue and retention, read-only from your own Stripe. Honest about small numbers: under the threshold it says 'Not enough data' instead of a verdict.
Sell per use, not by subscription? Grow reads your Stripe payments: revenue by month, new and repeat customers, and who hasn't come back. Two read permissions are enough: Customers and Charges. Selling subscriptions and one-off extras? Your MRR stays MRR, and the extras show on their own line.
Grow: PMF health, revenue and retention · People: who paid, who wrote in, who talked with you
