The product market fit survey and the Sean Ellis test, explained

The one question that matters

Ask the people using your product:

How would you feel if you could no longer use this product?

Give exactly three options: very disappointed, somewhat disappointed, not disappointed. That is the whole survey, and it is known as the Sean Ellis test after the growth lead who introduced it in The Startup Pyramid.

The signal is the share who answer very disappointed. Ellis's recommendation is not to start optimizing growth until at least 40 percent of surveyed users give that answer, and 40 percent has been the working bar ever since.

Two things about that number are worth saying plainly, because most write-ups skip both.

It is a rule of thumb from one practitioner's experience, not a law derived from a large controlled study. You will see the sample behind it quoted confidently and inconsistently across the internet, sometimes as a handful of companies he took to market, sometimes as around a hundred he later benchmarked. Treat 40 as a useful line in the sand rather than a constant of nature.

And it works because of what the question avoids asking. It never asks whether people like your product. Almost everyone likes almost everything when asked directly. Removal is the only framing that makes a person weigh what they would actually have to go back to, and that weighing is what separates a must-have from a nice-to-have.

Who to ask, and who to ignore

The fastest way to get a meaningless number is to survey everyone who ever signed up.

Survey only people who have used the core of the product at least twice in the last two weeks. Someone who signed up and never returned is telling you something real, but it is about your onboarding, not about fit, and mixing the two produces a score that measures neither.

• At least 40 to 50 responses from real users before you read the percentage. Below that, a couple of answers swing the result by several points.
• Recent users only. A person who loved it six months ago is remembering a different product.
• One open question alongside it: what is the main benefit you get from this product?

That open question is worth as much as the score. The people who answer very disappointed will describe the benefit more sharply than your own marketing does, in words you should probably steal outright. It is the cheapest positioning research available to you.

Read it by segment, never in aggregate

A single blended number across all users is the most common way this survey gets wasted. 25 percent overall can easily be 55 percent among one kind of user and near zero among everyone else, and those two businesses need opposite decisions.

So split the very disappointed group out and ask what they have in common: role, company size, how they arrived, what job they were doing when they used it. That group is your real target customer, and it is usually narrower than the one you set out to serve. Finding that you have strong fit with a segment you were treating as incidental is one of the most valuable outcomes this survey produces.

Then read the somewhat disappointed group, because that is where the roadmap is. Ellis's framing is to improve toward the must-have: find the one change that would move a somewhat into a very, and build that.

Deliberately ignore the not disappointed group. Their feedback is often the most detailed and the most tempting, and building for them pulls the product away from the people who already value it. They were never going to love it, and trying to win them is how a product with real fit in one segment becomes a product with none anywhere.

What to do at each score

Below 40 percent: do not scale acquisition. This is the expensive mistake the survey exists to prevent. Paying to send more people into a product they can take or leave does not fix fit, it just spends money faster and burns the audience you will want later. Keep improving toward the must-have and rerun the survey.

Around 40 percent: look at the segments before you decide. A blended 40 built from one strong segment and one weak one means you have fit and a targeting problem, which is good news. A flat 40 across every segment usually means the product is broadly adequate and specifically loved by nobody, which is harder.

Above 40 percent: shift the question from fit to growth. You have evidence people would miss it. What you do not yet have is evidence you can reach more of them affordably, which is a channel question rather than a product one.

At every score, rerun it periodically rather than once. The number moves as you change the product and as the mix of who uses it changes, and the trend tells you more than any single reading.

Three traps that make the number lie

1. Surveying your friendliest users. If you send it to your beta community or the people who reply to your emails, you have sampled the enthusiasts and the score is meaningless. Survey a random slice of qualifying users.
2. Reading a small sample as precision. With 20 responses, 40 percent and 30 percent are the same answer. Report a range, or collect more.
3. Asking too early. Someone who has used the product twice this week still has a real opinion. Someone who signed up an hour ago does not, and including them drags the score down for reasons that have nothing to do with fit.

The survey is cheap enough that all three are avoidable. What makes it valuable is that the result can genuinely tell you to stop, and that only works if you have not quietly arranged the sample so it cannot.

You can run a version of this before you have a product

At the idea stage you cannot ask how people would feel if the product disappeared, because it has not appeared. But you can measure the thing the survey is reaching for.

Publish a real page describing the product and the result it produces, send a small amount of honest traffic from where that audience already is, and measure how many of exactly those people put down an email or a card. A strong rate from a precisely targeted audience is the earliest version of the same signal: these people would be sorry not to have this.

The mechanics are in how to run a fake-door test, and the wider framing, including how to pick which assumption to test first, is in how to validate a business idea. Both are cheaper than building, and both tell you who to build for, which is the question this survey answers once you have something to survey about.