Product-Market Fit

Product-market fit lives in the flat part of a retention curve. The number at the top of the curve is close to meaningless. A cohort that starts at 38% on day one and settles at 11.5% by day 90 has fit with 11.5% of the people who signed up. The other 88.5% were traffic.

That single reading habit separates teams who know whether they have fit from teams who are still arguing about it.

What does product-market fit mean?

The phrase in its modern form comes from Marc Andreessen's 2007 essay The Only Thing That Matters, where he defined it as being in a good market with a product that can satisfy that market.

Two halves, and teams routinely measure only one. "A product that can satisfy that market" is what surveys and retention curves test. "A good market" is a question about how many people have the problem and whether you can reach them, and no amount of user love answers it. A product that 50 people cannot live without, in a market of 400 people, has satisfied its market and has nowhere to go.

What is the 40% product-market fit test?

Sean Ellis's survey asks existing users one question: how would you feel if you could no longer use this product? The options run from very disappointed to not disappointed. The benchmark he set is that 40% or more answering "very disappointed" indicates product-market fit.

It is the most useful cheap instrument in the set, and it has two conditions that get ignored:

1. Ask people who have actually used the product recently. Someone who signed up and never came back has no opinion worth collecting, and including them makes the number meaningless in both directions.
2. Ask enough of them. At n=30 the confidence interval on a 40% reading is wide enough to cover both a pass and a clear fail. Teams celebrate 43% on 30 responses all the time.

Rahul Vohra's write-up of Superhuman's process in First Round Review added the move that makes the survey actionable: segment the respondents. Split the people who said very disappointed from the rest, look at what they have in common, and build only for that group. An overall score of 25% can hide one segment at 55%, and that segment is the business.

How do you measure product-market fit?

No single signal is enough. Four or five together give an answer you can act on.

A worked example: reading the retention curve

Here is a cohort of 900 signups, with the share still active at each point. The numbers are a worked set chosen so the shape is clear.

The curve flattens at about 11.5%, which is 104 people who keep coming back three months later. That flat line is the fit. It says that for roughly one in nine signups this product became part of how they work, and it says that acquiring 900 more people like these would produce another 104 durable users.

Now the version with no fit: 12% at day 7, 8% at day 30, 4% at day 60, 1.5% at day 90. No floor. Every dollar of acquisition rents users for a few weeks. Growth spend on that curve produces a revenue chart that looks fine for two quarters and then stops, because the base is leaking as fast as the top is filling.

The difference between the two is invisible at day 7. Both look like a product losing most of its signups, which every product does. You have to wait for the floor.

Where do product-market fit frameworks break down?

1. Fit gets treated as binary and permanent. It is per-segment and it expires. A product can have strong fit with early-adopter engineers and none with the mainstream buyer it needs next, and a competitor's launch can remove fit you already had.
2. The 40% test on a friendly sample. Surveying your beta group, your design partners or the users who answer emails will clear 40% for almost any product. The number only means something if the sample looks like the market.
3. Enterprise products have no n. With fourteen customers there is no survey to run. Use renewal rate, seat expansion within an account, and how many of those fourteen would take a reference call.
4. Averaging across segments hides the business. This is the failure Superhuman's segmentation move exists to fix. One number across all users can bury the one group that loves the product.
5. Confusing fit with funding or press. Raising a round and getting written about are evidence that investors and journalists found the story good. Neither is a user coming back in week nine.
6. Measuring satisfaction and skipping market size. The survey measures how much existing users would miss you. It says nothing about how many more of them exist. That is the half of Andreessen's definition that teams drop, and it is the half that decides whether the company can get large.
7. Declaring fit from a single strong cohort. One good month can come from a launch spike of unusually motivated users. Fit shows up when three consecutive cohorts flatten at a similar floor.

How do PM interviews ask about product-market fit?

This turns up in strategy rounds and in the "would you launch" question, and the weak answers all share one feature: they describe fit without saying how they would know.

• "How would you know if this product has product-market fit?" Name two signals and the threshold for each. Survey at 40%, retention curve with a floor. Then say what you would do if the two disagreed.
• "We are at 28% on the Sean Ellis survey. What now?" The segmentation answer. Split the very disappointed group, find what they share, and consider narrowing the product to them before adding anything.
• "Retention flattens at 4%. Do we have fit?" A floor exists, so something is working for a small group. The real question is whether that group is large enough to build a business on, which is a market-size question and not a product one.
• "Would you launch?" Almost always a fit question wearing different clothes. Answer with the signal you would need to see first and the cohort length you would wait for.

Saying "I would wait for three cohorts to flatten before I called it fit" is the kind of sentence that ends this line of questioning, because it shows you have watched a curve rather than read about one.