Not a score and not a moment. Retention, unprompted return and how hard people fight to keep using it.
12 min read · Level 2
Product-market fit is measured in behaviour, not compliments, and it is the one thing that cannot be bought. Acquisition is purchasable, press is purchasable, a funding round is purchasable — people coming back next week without being prompted is not. Founders reliably convince themselves they have it because signups, coverage and enthusiasm can all coexist with nobody actually needing what you built.
Product-market fit means you have built something a specific group genuinely wants, evidenced by their behaviour rather than their compliments.
The honest signals: they come back without being prompted, they tell other people, they complain loudly when it breaks, and they would be genuinely annoyed if you took it away.
Because almost everything else is downstream of it. Hiring, fundraising, marketing spend and scaling all assume PMF. Doing them before it just makes a wrong thing bigger and more expensive.
And because founders very reliably convince themselves they have it. Signups, press, a funding round and enthusiastic feedback can all coexist with nobody actually needing what you built.
Two companies, both at 500 signups.
Company A: 500 signups, 40 weekly active, retention flat at 8% by week 8, no referrals, churn 12% monthly. Users are polite in interviews. This is not PMF, whatever the signup number suggests.
Company B: 180 signups, 95 weekly active, retention flat at 52% by week 8, a third arrived by word of mouth, and two customers emailed within an hour when the service went down.
B has far fewer users and unambiguously has fit. The tell is the retention curve flattening — it means a stable group has genuinely adopted it — and the fact that people noticed and cared when it broke.
The outage is genuinely one of the best signals available. Nobody emails about a product they were indifferent to.
First-time founders often measure signups and press, which are the easiest numbers to move and the least informative. Both can be bought.
The second mistake: treating PMF as binary and permanent. It is a matter of degree, it applies to a specific segment, and it can be lost — through a competitor, a market shift or your own changes.
The third: assuming more features produce PMF. If retention is flat at 8%, adding features usually moves it to 9%. The problem is normally the segment or the core value, not the feature count.
Group users by join week and follow each. You are looking for the curve to flatten rather than fall to zero.
A flattening curve means a stable group adopted it. A curve reaching zero means nobody did, however high it started.
Usage without emails or notifications driving it. Prompted usage measures your marketing, not your product.
'How would you feel if you could no longer use this?' A commonly used threshold is 40% answering 'very disappointed' — treat it as a signal, not a certificate.
Silence during an outage is the clearest negative signal there is.
You may have strong fit with one group and none with another. The average hides the finding.
Continuously from your first cohort onwards, and definitely before scaling spend or hiring.
Do not assess PMF with under about thirty users or fewer than eight weeks of data. Cohort curves need time to flatten, and early enthusiasm is not durable.
The most widely used quantitative signal is a cohort retention curve that flattens above zero rather than decaying to nothing. A product with 500 signups and 8% week-eight retention has a marketing achievement; one with 180 signups and 52% has a smaller number of people who genuinely adopted it.
Plot retention by cohort before believing any other number. A flattening curve is the closest thing to proof this field offers.
One of the clearest signals costs nothing to observe: when the product breaks, does anyone complain? Silence during an outage is the most honest negative feedback a company can receive, and it arrives without a survey.
Note what happens the next time you go down. Angry emails within the hour are a better PMF signal than any deck you could build.
Create a free account to read the rest and apply it to your own startup.