Cohort analysis groups users who share a common starting point — like the month they signed up — and tracks how each group behaves over time. It reveals retention, churn, and revenue patterns that blended, all-user averages completely hide.
What Is a Cohort?
A cohort is a group of users who share a characteristic or experience within a defined period — for example:
- All customers who signed up in January 2025
- Users who made their first purchase during a Black Friday sale
- Subscribers who joined after a specific marketing campaign
The word "cohort" comes from ancient Roman military units — in business, you're grouping users to battle for retention rather than territory.
Cohort Analysis Example
The classic view is a retention table: each row is a signup-month cohort, each column is how many of them are still active N months later.
| Cohort | Month 0 | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| Jan signups | 100% | 68% | 55% | 48% |
| Feb signups | 100% | 72% | 60% | 54% |
| Mar signups | 100% | 80% | 71% | — |
Reading down a column shows whether retention is improving cohort-over-cohort (here it is — March's onboarding clearly works better). A single blended "60% retention" number would have masked that trend entirely.
How to Read a Cohort Retention Curve
Plot any row and its shape tells you most of what you need to know:
- The cliff: a steep drop in the first period or two, then a gentler slope. Almost every product has one — it's the users who never found value. A steep, early cliff points at onboarding or activation problems.
- The flattening tail: after the early drop, a healthy curve levels off — the survivors are sticky. A curve that keeps sloping down never reaches stable retention, a warning sign for long-term lifetime value.
- The smile: retention that dips then rises later as dormant users return or expand. It's the strongest pattern, common in products with network effects or recurring need.
The goal isn't a flat 100% line — that never happens. It's a curve that flattens sooner, higher, and improves cohort-over-cohort.
Types of Cohort Analysis
| Type | Groups users by |
|---|---|
| Time-based | When they became customers (signup week/month) |
| Behavior-based | Actions they took (used a feature, hit a milestone) |
| Size-based | Purchase amount or company size |
How to Do a Cohort Analysis
- Define your cohorts: pick the shared characteristic (e.g. signup month).
- Choose a metric: retention rate, revenue, or engagement.
- Track it over time: measure each cohort across successive periods.
- Compare cohorts: read down the columns to spot trends and act on them.
Common Cohort Analysis Mistakes
- The partial-period trap: the most recent cohort hasn't finished its latest period yet, so its final number looks artificially low. Grey out or exclude incomplete periods before you compare.
- Tiny cohorts, noisy results: a 20-user cohort swings wildly on a few churns. Group into larger buckets (quarter instead of week) until each cohort is big enough to be stable.
- Ignoring seasonality: a cohort acquired during a holiday promo or a specific campaign can behave very differently. Compare like with like.
- Billing cycle masking churn: annual-contract customers look "retained" until renewal simply because they're still paying — engagement, not billing status, is the truer signal of retention for those cohorts.
Metrics to Track
| Metric | What it shows |
|---|---|
| Retention rate | % still active over time |
| Churn rate | % who stopped using the product |
| Customer Lifetime Value | Revenue per customer over their lifetime |
| ARPU | Average revenue per user |
| Engagement | Login frequency, feature usage |
Why Cohort Analysis Matters
- Accurate LTV: see how customer value actually develops over time.
- Product insight: identify which features drive long-term retention.
- Marketing optimization: find which channels bring the best-retaining customers.
- Churn prevention: spot exactly when and why customers leave.
- Forecasting: project revenue from real cohort behavior, not averages.
Tools for Cohort Analysis
Google Analytics (basic web cohorts), Mixpanel and Amplitude (event/product cohorts), Tableau (custom visualization), or direct SQL queries for full control.
Cohort Analysis FAQ
What is cohort analysis in simple terms?
It's grouping users by a shared starting point — usually when they signed up — and watching how each group behaves over time. Instead of one blended average, you see how January's customers compare to March's at the same age.
What is an example of cohort analysis?
A retention table where each row is a signup-month cohort and each column shows the % still active 1, 2, 3 months later. Comparing rows reveals whether newer cohorts retain better than older ones.
What are the types of cohort analysis?
Three main types: time-based (grouped by signup date), behavior-based (grouped by an action taken), and size-based (grouped by spend or company size).
Why is cohort analysis better than overall averages?
Blended averages mix new and old users, hiding trends. Cohort analysis isolates each group so you can see whether retention, revenue, or engagement is genuinely improving over time.
What is cohort analysis used for?
Measuring retention and churn, judging onboarding and product changes, comparing acquisition channels by how well their users stick, calculating accurate lifetime value, and forecasting revenue from real behavior instead of blended averages.
What is the difference between cohort analysis and time-series analysis?
Time-series tracks one metric for your whole user base over calendar time. Cohort analysis splits users into groups by a shared start point and lines them up by their own age, so a January signup at "month 3" is compared to a March signup at "month 3" — isolating behavior from calendar noise.
