Startups & Business › Product-Market Fit

Retention Curve

Share of users still active over time; a curve that flattens is the clearest sign of fit.

Also known as: retention curve, cohort retention, flattening retention

A retention curve plots what share of a cohort is still active at each age: 100% at start, decaying as users leave. Curves that decay toward zero mean nobody stays — no fit, whatever growth says. Curves that flatten at a stable level mean a core of users found lasting value — the clearest quantitative sign of product-market fit there is.

no fit:    100% → 40% → 15% → 5% → 2% → ~0%   (decays to zero)
fit:       100% → 55% → 38% → 32% → 30% → 30% (flattens: a third stays)

Read it by cohort and by segment, never blended: overall curves mix improving new cohorts with decaying old ones and hide both stories. Compare flattening levels across cohorts over time — rising floors mean the product is getting better at keeping people.

The classic mistakes:

• Defining “active” to flatter. Logged-in-but-idle, or any heartbeat event, inflates retention into meaninglessness. Active must mean real value received — define it strictly and never loosen it to improve a chart.
• Blended curves. One number across all cohorts and segments conceals the cohorts that flatten and the ones that die. Always cut by cohort age first (cohort analysis).
• Reading too early. Week-two retention of a week-old cohort is noise with a chart. Wait for cohorts to age past the novelty window before judging the floor.
• Ignoring resurrection. Users returning after months (seasonal products, annual cycles) break simple decay reads. Define re-engagement explicitly rather than letting it smear the curve.
• Optimizing the curve cosmetically. Re-engagement spam and dark patterns lift short-term activity while destroying trust. Retention earned through value compounds; retention extracted does not.

The verdict rule: flat-ish floors, stable or rising across recent cohorts, built on strict activity definitions. That shape — more than any survey or spike — is what “it works” looks like in data.