The 40% product-market fit test, and what to do at 25%

By Meet Patel · 2026-10-03 · 6 min read

In summer 2017, Superhuman's founder Rahul Vohra ran a short survey and got a score of 22%. Three quarters of deliberate work later the same survey returned 58%. He described the whole sequence in a November 2018 First Round Review piece, How Superhuman Built an Engine to Find Product-Market Fit, and the article is the best practical manual I know for what to do when a team suspects it has not found fit yet.

The survey rests on a benchmark from Sean Ellis. According to the article, Ellis found that the magic number was 40%: companies that struggled to find growth almost always had less than 40% of users respond “very disappointed” to the key question, whereas companies with strong traction almost always exceeded that threshold. This post covers how to run the survey, how to read a result like 25%, and where the number stops being useful.

The four questions

The survey as Vohra describes it has four questions.

1. How would you feel if you could no longer use the product? (Very disappointed, somewhat disappointed, not disappointed.)
2. What type of people do you think would most benefit from the product?
3. What is the main benefit you receive from the product?
4. How can we improve the product for you?

The score is the percentage of respondents who answer “very disappointed” to the first question. Questions two to four are what you act on, and I would argue they are worth more than the score itself. The score tells you how far you are from the line. The other three tell you who is already past it and why.

Who to survey, and how many

Vohra followed Ellis's recommendation to survey users who had recently experienced the core of the product, defined as people who used it at least twice in the last two weeks. He also notes that you start to get directionally correct results at around 40 respondents, which is fewer than most teams assume. At the time Superhuman had between 100 and 200 users to poll.

Small samples still deserve care, and the arithmetic shows why. With 40 respondents, each person moves the score by 2.5 points. With 12 respondents, one person moves it by more than 8 points, so a result of 25% against 33% means nothing. Treat a first survey with fewer than about 40 responses as a prompt to read the free-text answers, and avoid comparing the percentage with the benchmark.

Reading the free-text answers

Most of the value sits in the free text, and reading it takes discipline. Print the answers to the main-benefit question for the very disappointed group only, and tally the words people repeat. If eleven of fifteen people describe the same outcome in different words, that outcome is the product's job in their lives. If no two answers resemble each other, the product has not yet become something specific to anyone, which is a result worth writing down.

Do the same for the improvement question, then separate the requests into two piles: things the very disappointed ask for, which deepen what they already value, and things the somewhat disappointed ask for, which remove the friction between them and a stronger score. The two piles feed the two halves of the roadmap split described below.

What Superhuman did at 22%

The team looked for the segment before trying to raise the number across the board. Per the article, it grouped responses by how disappointed people were, assigned a persona to each respondent, and looked at which personas dominated the “very disappointed” group. That narrowed the market it was building for into a high-expectation customer profile. It then filtered the feedback by the main benefit people named, which for the people who loved the product was speed.

After the first round of segmentation the score stood at 33%, a result consistent with the survey now describing the people the product was built for. The remaining gap closed through the roadmap. Vohra's team politely disregarded the somewhat disappointed users for whom speed was not the main benefit, and paid very close attention to those for whom it was. Half the roadmap went to amplifying what users loved, and half to the friction holding back the near-fit group. His reason, in the article's words: “If you only double down on what users love, your product-market fit score won't increase. If you only address what holds users back, your competition will likely overtake you.”

A worked example at 25%

Take a hypothetical 12-person company that sells invoicing software and surveys 80 recently active users. The figures are invented.

Twenty users answer “very disappointed”, 28 answer “somewhat” and 32 answer “not disappointed”. The score is 20 out of 80, which is 25%, well under the benchmark, and the first instinct is to build more features for everyone. The persona question changes the picture:

• Freelancers: 30 respondents, 15 very disappointed, 10 somewhat, 5 not. A score of 50%.
• Agencies: 50 respondents, 5 very disappointed, 18 somewhat, 27 not. A score of 10%.

The overall 25% combines a segment that is past the line with one that is far from it. Among the 15 freelancers, the most common answer to “main benefit” might be getting paid within days of sending an invoice. Among the 5 agencies, the benefit might be vague. The company now has a sharper sentence for its positioning, a customer to prioritize, and a roadmap question that can be answered: what stops the 10 somewhat disappointed freelancers from becoming very disappointed?

Suppose their improvement answers cluster around two requests: recurring invoices (7 mentions) and bank feeds (2 mentions). Recurring invoices goes to the top of the friction half of the roadmap, and bank feeds waits. Without the survey, those two requests would have sat in a backlog ranked by whoever asked loudest.

What to do at 25%

1. Check the sample. Recent users, around 40 responses or more. Below that, read the answers and hold the percentage loosely.
2. Segment before optimizing. Compute the score by persona, plan or use case. Look for a segment above 40%.
3. Read the main-benefit answers from the very disappointed. Their words are your positioning and the feature set to protect.
4. Sort the somewhat-disappointed group. Follow Vohra's rule: study those who share the main benefit of the very disappointed, and set aside the rest.
5. Split the roadmap. Half to deepening what the very disappointed value, half to removing the top friction named by the near-fit group.
6. Re-run quarterly with the same questions and the same recency rule, so the trend is comparable.

Where the number stops being useful

I treat 40% as a rule of thumb from Ellis's observations of startups, and I would not defend it as a law. Three limits are worth keeping in view.

• It measures sentiment of current users. A segment can score 58% and still be too small to build a company on. The score says nothing about how many such customers exist, which is a separate question I come back to in day-one density.
• It ignores behavior. Pair it with a retention curve. If users say they would be very disappointed and still leave after a month, the two sources disagree and the disagreement deserves an investigation. I wrote about why retention is a product problem.
• It can be inflated by who you ask. Surveying only your most active users raises the score. The recency rule is there to keep the sample honest, and the segment-by-segment view is where the useful reading comes from, a point close to the one in the audience trap.

A score under 40% describes a gap between the product and a market, and the survey tells you which market to close it for. Run the four questions, find the segment already past the line, and spend the next quarter making it larger while you repair what holds the near-fit users back.

Written by Meet Patel — startup operator and growth strategist in Dubai.

Read on themeetpatel.com