What Is a Cohort in Analytics?
Cohorts defined: grouping users by start date or behavior, why cohort views beat averages, and simple cohort analyses anyone can run.
A cohort is a group of users defined by a shared starting point — everyone who signed up in March, everyone whose first visit came from the launch campaign. Cohort analysis follows each group through time separately, which is the only way to see change: averages blend the past into the present until nothing is visible.
Why averages lie and cohorts do not
Suppose engagement looks flat quarter over quarter. Cohort view: users acquired before the redesign retain at 40%, users after at 25% — the product got worse, masked by loyal veterans propping up the average. Or the reverse: a rising average that is just one good old cohort aging gracefully while new ones churn. Every 'our numbers look fine' disaster story is an average hiding a cohort.
The two cohort types
- Time-based: grouped by start week/month. The workhorse — powers retention curves and before/after comparisons around launches and pricing changes.
- Behavior- or source-based: grouped by what they did or where they came from — users who activated in week one vs not; newsletter-acquired vs paid-acquired. This is where first-touch attribution earns its keep: acquisition source rides the visitor record, so source cohorts are a filter, not a data project.
Three cohort analyses anyone can run
- Retention by signup month: are newer cohorts holding better at the same age? The product-improvement scoreboard.
- Activation by source: which channels send users who reach value? Reallocates budgets faster than any CPL comparison (trial decomposition is this analysis with revenue attached).
- Revenue per cohort over time: does a March customer spend more by June than a January customer did by April? Net revenue retention, told as a story.
No special tooling required — a start-date property and patience. The discipline is refusing to quote any number about 'users' without asking 'which cohort?'