What is user retention?

Retention is not one number, it is the same measure taken at different distances from signup. Each horizon reads a different part of the user's life: the early ones read onboarding and habit, the later ones read whether the value was durable.

All of these come from grouping users into cohorts by when they joined, then measuring who is still active at each horizon. Retention and cohort analysis are the same machine viewed from two angles.

A retention curve plots one cohort over time: start with everyone who joined in a given week, then chart the fraction still active on day 1, day 7, day 30, and beyond. It almost always starts high and falls, the question is what shape it settles into.

A curve that decays to zero is the bad case: users tried the product and drifted away, and pouring in more signups just refills a leaky bucket at the top while the bottom keeps draining. Growth on a decaying curve is an illusion that ends the moment you stop spending on acquisition.

A curve that decays and then flattens into a plateau is the good case: some fraction of every cohort found lasting value and stopped leaving. That flat tail is the closest thing to a measurable read on product-market fit, because it means each new cohort stacks on top of a retained base instead of replacing the one before it.

Plotting several cohorts side by side (a cohort grid, one row per signup week) shows whether the shape is improving. If cohorts after a change hold a higher plateau than the ones before, the change worked. That is the whole point of measuring retention as a curve rather than a single percentage.

Acquisition gets a user in the door once. Retention decides whether they stay, and only users who stay compound into revenue, word of mouth, and durable growth. A product with great acquisition and no retention is a treadmill: you have to spend more every month just to stand still.

This is why retention is the metric investors and operators reach for first. It cannot be bought, gamed by a launch spike, or faked with a discount, a user either found enough value to come back or they did not. Fixing retention also fixes everything downstream: better retention lifts lifetime value, lowers the payback period on every acquisition dollar, and turns paid growth from a leaky bucket into a compounding one.

smolanalytics gives you a retention report, the curve plus a cohort grid, alongside web and product analytics from one snippet or one endpoint. What makes it different is not that it has retention, every tool does, but four choices about how you get the answer:

• 1The report, then an investigator. You get the retention curve and the cohort grid to look at, and beyond that, the investigator flags a retention break the day the curve slips, on the desk and in the morning brief, so a decaying cohort reaches you as a finding instead of a chart you had to remember to open. If the curve recovers on its own, the finding retires as recovered, never as fixed, because nobody proved causality.
• 2Ask in plain English. Type "is retention holding?" or "did day-7 improve since the new onboarding?" into a dashboard bar, or into your own Cursor / Claude Code over MCP (97 tools, 15 prompts), using your own AI model so the AI part is free.
• 3Computed, never guessed. Every retention answer comes from the same deterministic report the dashboard renders, not from an LLM writing numbers.
• 4One binary. Your project runs as its own single Go binary, stdlib only, roughly 7 bytes per event, no Kafka, ClickHouse, or Postgres to run, and your data exports in one file any time.

It now ships feature flags, experiments, heatmaps, surveys, and a session inspector too, without a suite you have to operate or a per-seat bill. The one thing it deliberately skips is pixel-perfect DOM session recording; the session inspector replays journeys from events, not video. It is for teams who want a straight, owned, cheap answer on whether users are sticking. See every feature, the related cohort analysis, or how it compares vs Mixpanel.