Cohort analysis
Cohort analysis groups customers by the month of their first purchase, then tracks each group's repeat orders, revenue and margin over the months that follow. It is measured as retention and cumulative value per customer at each month of age, and ecommerce operators use it to see whether newer customers repeat better or worse than older ones.
| Cohort size | Customers whose first order fell in the cohort's start month. The denominator is fixed and never changes |
| Month n | Months since first purchase (cohort age), not calendar months |
| Cumulative value per customer | Cohort revenue - or better, contribution margin - through month n, divided by cohort size |
Worked example
Take a store's January cohort: 1,000 first-time customers spending $85,000 on their first orders, an $85 average order value (example numbers, flagged as such).
Ninety days in, the cohort is worth $120.10 per customer against $85.00 on day one - 41% more than its first order. Whether that clears the cost of acquiring the cohort is a margin question, not a revenue one.
What is a good cohort retention curve?
There is no universal benchmark, and any single number would mislead. The shape of a healthy curve depends on the category's repurchase cycle - consumables repurchase in weeks, durables in years - on your contribution margin structure, and on the channel mix the cohort was acquired from.
Two things matter more than any threshold: whether the curve flattens above zero, meaning a persistent repeat base survives rather than decaying to nothing, and whether younger cohorts sit above or below older cohorts at the same age. A brand whose month-3 value per customer falls cohort over cohort is getting worse at retention even while total revenue grows. The floor worth computing is your own: a cohort has to return its acquisition cost in contribution, not revenue, and our free margin LTV calculator computes that payback floor from your numbers.
Cohort analysis vs related metrics
| Metric | What it measures | How it differs |
|---|---|---|
| Retention rate | Share of customers who stay active in a period | One blended average across the whole file. Cohorts show the same idea separately by customer age, so improvement and decay stop cancelling out. |
| Repeat purchase rate | Share of customers who ever order twice | A single summary with no time axis. A cohort curve shows when the second order arrives and whether it keeps arriving. |
| Customer lifetime value (LTV) | What a customer is worth over their life | A forward projection. Cohort curves are the observed data any credible LTV model is fitted on. |
| RFM segmentation | Which current customers are most valuable | Groups by current behaviour, so membership shifts. A cohort is fixed at first purchase and never moves. |
Common mistakes
- Judging immature cohorts. A 60-day-old cohort cannot show month-3 behaviour yet. A blank cell is not a zero.
- Tracking revenue while acquisition is paid from margin. Discount-led repeat orders flatter the revenue curve while contribution shrinks underneath it.
- Comparing calendar months instead of cohort age, so seasonality reads as a retention change.
- Mixing acquisition channels in one cohort, then acting on an average no single channel actually produced.
- Ignoring returns. Early cohort revenue overstates the keep until the returns window closes.
Frequently asked questions
In Blufire, S4 Customer Value & Segmentation builds cohort value on contribution margin - CLV, the margin-true RFM cube, lifecycle states and migration, and Financial Buckets - and S8 Persona Analytics turns those groups into audiences.
Updated July 2026