Retention

What is Retention?

Retention is the percentage of users who come back and keep using a product over a given period, and it is the single clearest signal that a product has found product-market fit.

Most teams track it as a curve rather than a single number: plot the percentage of a signup cohort still active on day 1, day 7, day 30, and day 90, and you get a line that drops fast at first and then, if the product has staying power, levels out. Casey Winters, in Lenny Rachitsky's retention research piece, puts it in plainer terms: great retention is the scalable way to grow a product, and the best indicator of product-market fit a team has (Lenny Rachitsky, "What is good retention?", Lenny's Newsletter, 2020). Andreessen Horowitz frames the same mechanic from the metrics side. In their startup metrics research, they look for retention to stabilize within each cohort after a period such as six or twelve months, which means a business is building a progressively larger base of recurring usage rather than replacing users it is quietly losing (Andreessen Horowitz, "16 More Startup Metrics").

Where the field gets less settled is which single number best represents retention, a fixed-day mark or the shape of the whole curve over months. Produck's position is that the curve matters more than any one day. A flattening curve tells you people who tried the product found a reason to stay, which a single retention percentage on its own cannot.

Why it matters for product-market fit

Inside Produck's Listen, Diagnose, Decide, Ship loop, retention is the scoreboard for Ship. You can run a tight feedback loop and still ship fast, but none of it counts for much if the people you shipped for don't come back. Retention is what tells you the loop is actually working, not just spinning.

Retention also disambiguates two things founders routinely mix up. Activation tells you whether someone got to first value. Retention tells you whether that value was real enough to repeat. A product can have great activation and terrible retention, which usually means the first-use moment was impressive but the ongoing habit never formed. Watching churn alongside retention closes the loop, since churn is retention's mirror, the users the curve lost.

This is also where a lot of teams go wrong chasing a scattered stack of disconnected tools instead of a loop that connects feedback to shipped changes and back to a retention number, a mismatch our team wrote about in the PMF stack is broken. Produck exists to close that gap. Feedback comes in, gets diagnosed and prioritized, ships as a real change, and the retention curve is the readout on whether it worked.

When it works, and when it doesn't

It works when:

1. You track retention by cohort over time, not as one company-wide average that hides how new users behave differently from old ones.
2. You pair the curve with the reasons behind it. A flattening curve plus real user feedback tells you what to protect when you ship new features.

It falls short when:

1. You measure retention too early to mean anything. A week-one number for a product used quarterly is close to meaningless.
2. You optimize the number itself instead of the behavior underneath it. Re-engagement notifications that nudge a login without any real use inflate the curve without fixing the product.

How to apply it

1. Pick the cohort window that matches how often your product should naturally get used. A daily tool needs day-1 and day-7 retention. A quarterly finance tool needs month-3 and month-6.
2. Build the retention curve for your last few signup cohorts and look for where it starts to flatten, not just what the day-30 number says.
3. For every user who drops off the curve, pull their feedback and support history. Patterns here are your diagnosis, not guesses.
4. Ship the fix that addresses the most common reason people stop coming back, not the loudest one-off request.
5. Re-run the same cohort analysis after the change ships and compare the new curve to the old one, same time window, same segment.
6. Repeat monthly. Retention curves move slowly, so judge trend over two or three cycles before declaring a fix did or didn't work.