What cohort analysis and retention curves are
A cohort analysis groups users by a shared starting event — most often the week or month they signed up — and follows each group forward in time. A retention curve takes one cohort and plots the share still active at each period after that start: week 0, week 1, week 2, and on. The shape of that line is the story.
Why cohorts separate retention from growth mix
Your total active-user count blends every cohort together, so a wave of new signups can hide a retention problem underneath. Split the same users into weekly cohorts and each curve stands on its own. If the March cohort holds 40% at week 8 and the April cohort holds 28%, something changed between them — a gap you would never see in the aggregate. This is also why cohorts pair with customer segmentation: compare retention by plan, channel, or activation path to find which groups actually stick — the same splits an engagement platform like Iterable uses to target lifecycle flows.
How to read a retention curve
Read three things: the height of the first drop, where the line bends, and whether it flattens. Almost every product loses a large share between period 0 and period 1 — that is normal. What matters is whether the decline slows and settles on a stable floor, or keeps sliding toward zero. A curve that never flattens means each cohort eventually churns out completely. See how to calculate churn rate for the same story told as one number per period.
A flattening curve as a product-market-fit signal
Growth practitioners treat a retention curve that flattens into a horizontal asymptote as a stickiness signal — Andrew Chen lists flattening cohort curves among his rough markers of product-market fit, and Reforge teaches the same 'smile or flatten' framing. Treat it as a signal, not proof. The floor's height matters as much as its existence: a curve that flattens at 3% describes a tiny loyal core, not a healthy product. And a flattening curve tells you a group retains, not why — that is a correlation to investigate, not a proven cause.
N-day vs unbounded retention
| Method | Counts a user as retained on day N if they… | Best for |
|---|---|---|
| N-day | were active on exactly day N (or a defined bracket around it) | habit products with a daily or weekly loop |
| Unbounded | returned on day N or any day after | irregular-use products; some tools call this rolling |
| Range / bracket | were active at least once inside a moving window | smoothing noise when usage is infrequent |
How to act on what the curve shows
The curve tells you where to look, not what to fix. A steep early drop points at onboarding and activation — the first session isn't delivering value fast enough. A curve that flattens low points at long-term value or the wrong audience. Once you have a hypothesis, change one thing, then compare the next cohort's curve against the last. The cohort is the unit of measurement for almost every retention experiment: ship the change, let a fresh cohort age, and read whether its floor moved.