Cohort analysis is a method to track how groups of users behave over time. A cohort is a group sharing a common starting trait—most often the week or month of first visit or signup. By following each cohort’s activity in fixed time buckets (weeks/months), you can compare retention rate, churn rate, activation, and eventual customer lifetime value (LTV) without mixing in users who started at different times. It’s different from broad segmentation because time alignment is the core.
Common cohort types:
- Acquisition cohorts: grouped by first seen date or source/medium.
- Behavioral cohorts: grouped by an action (e.g., completed onboarding).
- Attribute cohorts: grouped by plan, geo, or device.
Why it matters (web analytics angle)
Cohorts expose product-market fit and channel quality. Stable or improving retention suggests the product is sticky; falling retention hints at onboarding or feature gaps. Comparing cohorts by acquisition channel helps you spot the high-quality traffic versus vanity clicks. Cohorts also feed downstream metrics like LTV, payback, and repeat purchase rate.
How to calculate (short and clean)
Let a cohort have N0N_0N0 users at start.
- Retention at month m: Retentionm=Active users from cohort at month mN0\text{Retention}_m = \frac{\text{Active users from cohort at month m}}{N_0}Retentionm=N0Active users from cohort at month m
- Churn at month m: Churnm=1−Retentionm\text{Churn}_m = 1 – \text{Retention}_mChurnm=1−Retentionm
Mini-example: 100 users acquired in January.
| Month since signup | Active users | Retention |
|---|---|---|
| 0 (signup) | 100 | 100% |
| 1 | 45 | 45% |
| 2 | 30 | 30% |
| 3 | 24 | 24% |
Reading this: retention decays and then (ideally) flattens. Compare multiple cohorts (e.g., different channels or onboarding variants) to see which curve stays higher longer.
Analyst tips
- Define “active” once and hold the line (e.g., at least 1 session or 1 key event per period).
- Use consistent buckets (ISO weeks or calendar months).
- De-dupe users; don’t mix visitors and users unless that’s the point.
- Seasonality can masquerade as retention changes—compare like with like (e.g., Jan vs Jan).