Cohort analysis
Cohort analysis tracks how groups with a shared characteristic, such as signup month or acquisition channel, behave over time. It helps businesses compare retention, revenue or other outcomes at the same point in each group's lifecycle.
Key takeaways
- Compare groups at the same age: month 2 after signup against month 2, not against another group's first month.
- Define who enters the cohort and what counts as a successful outcome before calculating results.
- For customer retention, keep the original cohort size as the denominator. Removing customers who churned inflates the rate.
- Company-wide growth can hide weak retention if enough new customers replace those who leave.
How cohort analysis works
Start with a business question: "Are customers acquired this quarter staying longer than customers acquired last quarter?"
Then define:
- Entry rule: First purchase, signup or another qualifying event.
- Outcome: A repeat purchase, active subscription or product action.
- Time interval: Days, weeks or months since entry.
- Measurement window: How long each group has had to reach the outcome.
Google Analytics' "[GA4] Cohort exploration" documentation separates cohort inclusion from return criteria and cohort granularity. Those settings determine who belongs, what counts as returning and how time is grouped.
That distinction matters. Logging in once is not the same as renewing a subscription. A chart labeled only "retention" leaves the reader guessing.
Better retention produces $1,500 more monthly revenue
An illustrative subscription business reviews results through April 30. Each monthly cohort began with 100 new paying customers at $100 per month. There are no reactivations, price changes or account expansions.
Month 0 is the acquisition month. Later cells show the percentage of the original customers still subscribed at that month's end.
| Acquisition cohort | Starting customers | Month 0 | Month 1 | Month 2 |
|---|---|---|---|---|
| January | 100 | 100% | 60% | 40% |
| February | 100 | 100% | 70% | 55% |
| March | 100 | 100% | 80% | Not yet observed |
For February's month 2:
- Customer retention = 55 remaining customers ÷ 100 starting customers = 55%.
- Monthly revenue = 55 × $100 = $5,500.
- January's comparable revenue = 40 × $100 = $4,000.
February therefore produces $1,500 more monthly revenue at the same customer age, from the same initial customer count. Retention improved by 55% − 40% = 15 percentage points.
March's missing month-2 result is not zero. May falls outside the reporting window.
For a founder, this separates improved customer economics from growth purchased through more acquisition spending. It does not prove a product change caused the improvement: February's customers might have come from a different channel.
What to question before presenting the chart
Question a retention chart if it:
- Omits starting cohort sizes, so a small group looks as convincing as a large one.
- Treats unobserved future periods as zero.
- Changes the qualifying activity between cohorts.
- Shows only combined results despite a major change in customer mix.
There is no useful universal retention target across subscriptions, retail and daily-use apps. Compare like-for-like outcomes.
On an investor slide, include cohort dates, starting counts and the exact retention definition. Those labels make the claimed improvement checkable.
Cohort analysis vs. segmentation
| Method | Main purpose | Example |
|---|---|---|
| Cohort analysis | Compare group behavior over elapsed time | Repeat purchases in the months after each customer's first order |
| Segmentation | Group customers by shared traits, without necessarily aligning time | Revenue from enterprise customers versus small businesses |
Segmentation can be part of cohort analysis. For example, compare paid-search and referral cohorts at the same number of weeks after signup.
Questions people ask
What do "cohort 1" and "cohort 2" mean?
They are labels, not standard categories. Cohort 1 might mean January signups and cohort 2 February signups, or two different acquisition channels. The report should state each group's entry rule rather than make readers infer it.
What does "cohort-based" mean?
It means people are organized into groups with a shared starting point or characteristic. In reporting, their outcomes are tracked together. In a cohort-based course, participants usually start and progress together; that describes the delivery format, not necessarily an analysis.
What are the main types of cohort analysis?
Two common approaches are acquisition cohorts, grouped by when customers joined, and behavioral cohorts, grouped by an action or pattern of use. Retention, revenue and repeat-purchase analysis describe the outcomes measured, rather than a separate rule for forming the group.
