Cohort

A cohort is a set of users who share something material — typically the week they signed up, the channel they came through, or the pricing plan they started on. Grouping users into cohorts lets you ask questions like "did the November signups behave differently from the October signups?" — questions that are invisible if you look at the user base in aggregate, because new and old users always blend into a misleading average.

For experiments, cohort analysis is the difference between "the metric went up" and "the metric went up because of what we shipped." A retention chart on aggregate users will look better any time you ramp acquisition (more young users in the mix) and worse any time you slow acquisition. Cohort retention curves separate the two: each cohort is its own trajectory, and a real product improvement is visible as later cohorts retaining better than earlier ones at the same week-N.

When to use it

Use cohorts any time you are measuring retention, activation, or any metric that takes time to mature. Aggregate views hide compositional shifts; cohort views show whether the change you shipped actually changed behavior.

What this looks like in practice

The most common cohort cut is signup week. Retention at week N for the week-K signup cohort is the share of week-K signups still active at week K+N. The classic cohort-retention table — rows are signup weeks, columns are weeks-since-signup, cells are retention rates — shows whether your product is leaking in the first week or in the long tail, and whether changes you shipped are bending the curve. No other view shows it as cleanly.

Cohorting by acquisition channel is the second-most-useful cut. Paid-search signups, organic signups, and referred signups behave differently. A retention "improvement" that is actually a shift in the channel mix is invisible in aggregate. Splitting cohorts by channel reveals whether the product changed or the user changed. Most experiment verdicts on retention get clearer when you cut by both signup week AND channel.

Cohort sizes get small quickly when you cut multiple ways. A weekly cohort of 200 split into three channel buckets gives you ~70 users per cell — too small to read confidently. The fix is to widen the time window (monthly cohorts), narrow the splits (one channel cut at a time), or accept directional reads instead of statistical certainty. Forcing cohorts smaller than they should be is a faster route to wrong conclusions than not cohorting at all.

A worked example

Common mistakes

• Reading aggregate metrics as cohort behavior.Aggregate retention can move because cohort behavior changed OR because cohort mix changed. Without the cohort cut, the two are indistinguishable.
• Picking cohort boundaries to fit the story.Drawing a cohort line at "the week we shipped the change" and another at "the week we wanted the chart to look good" is data-dredging. Pre-commit cohort boundaries before reading the data.
• Treating small cohorts as definitive.A 50-user cohort with 8 retainers (16%) is one user away from 18% or 14%. Below ~200 per cell, treat cohort numbers as directional, not as verdicts.

Related terms