Plain definition
Cohort analysis groups customers by a shared starting event, such as acquisition month, plan launch, or first paid subscription, then tracks their behaviour through comparable elapsed periods. It reveals retention, revenue, usage, and payback patterns that a blended company average can hide.
The essential discipline is matching. A January cohort at month three should be compared with each cohort at its own month three, using the same active, revenue, and cancellation definitions. Do not compare a mature cohort with a cohort only weeks old and call the difference performance.
Illustrative example
Suppose 100 customers start in January and 80 remain active three months later. A February cohort also has 100 customers, but only 70 remain after its own three months. The comparison points to a change in acquisition quality, onboarding, product, or measurement that needs investigation.
New customers added in March do not belong in either starting cohort. Keeping cohorts separate lets management test a specific change and see whether later cohorts improve.
| January cohort active after three months | |
|---|---|
| January cohort starting customers | |
| Convert to percentage | |
| January month-three retention |
Cohorts need enough volume and stable definitions; a tiny segment can produce a volatile percentage without a reliable lesson.
Why it matters
Cohorts turn aggregate growth into evidence about customer experience over time. They are the foundation for matched CAC, LTV, retention, and payback analysis.
Practical consequence
Choose one cohort event, preserve the original membership, and build a calendar that compares elapsed periods. Reconcile customer and recurring-revenue movements to records, document any definition change, and annotate product, pricing, or channel changes. Use the findings to run a focused experiment, then compare its cohort with a later matched cohort.