Cohort Analysis Explained for Beginners
Cohort analysis groups users by something they share — most commonly, the week or month they first signed up — and then tracks how each group's behavior changes over time. Instead of looking at one blended, overall number (like "our retention rate is 60%"), cohort analysis lets you compare, for example, how well January's new users stuck around versus how well June's new users stuck around, which reveals whether the product is actually getting better or worse at keeping people, something a single overall average can hide.
Quick facts
- A cohort is simply a group of users who share a starting point in time — usually the month or week they signed up.
- Cohort analysis is most commonly used to study retention — what percentage of a cohort is still active after 1 week, 1 month, 3 months, and so on.
- The output is usually shown as a table or chart, with one row per cohort and one column per time period since signup — this specific layout is often called a cohort retention table, or sometimes described visually as a "cohort waterfall."
- Cohort analysis reveals trends that a single blended average number hides, since averaging across all users mixes old, established users with brand-new ones.
- It's a foundational tool for understanding both retention/churn and how well onboarding or product changes are actually working over time.
Why a single overall number isn't enough
If a product simply reports "60% of our users are active this month," that number blends together long-time loyal users with brand-new signups from last week — two very different groups with very different behavior. A product could have terrible new-user retention while still showing a healthy-looking overall active user number, simply because its established, loyal base is large and stable. Cohort analysis fixes this by isolating each starting group and tracking it on its own, so a change in new-user quality shows up clearly instead of getting hidden in a blended average.
A worked example: a simple cohort retention table
Imagine a habit-tracking app looking at retention by monthly signup cohort:
| Signup month | Week 1 | Week 2 | Week 4 | Week 8 |
|---|---|---|---|---|
| January | 100% | 62% | 41% | 30% |
| February | 100% | 65% | 44% | 33% |
| March | 100% | 58% | 35% | 22% |
| April | 100% | 70% | 50% | 38% |
Reading this table: every cohort starts at 100% in their first week (everyone was active when they signed up, by definition). Looking across the rows shows how retention decays over time for each specific group. Looking down the columns reveals a trend: March's cohort retained noticeably worse than February's at every time point, while April's cohort retained the best of all four months. This immediately raises a useful question — what happened in March that hurt new-user retention, and what happened in April that helped it? A single blended monthly retention number would never have surfaced this specific, actionable pattern.
How to actually use cohort analysis to improve a product
The real value of cohort analysis comes from connecting a change in cohort behavior to something the team actually did. If March's cohort retained poorly, the next step is investigating what changed that month — a pricing change, a confusing new onboarding flow, a bug, or a shift in what kind of user was being acquired through marketing. If April's cohort retained noticeably better, and the team shipped a new onboarding flow that month, that's a strong, testable signal the new flow genuinely helped — far more convincing than an anecdote or a gut feeling, because it's backed by a clear before-and-after comparison across similar groups of users.
Cohorts beyond just signup date
While signup-date cohorts are the most common, cohorts can be grouped by almost anything meaningful: which acquisition channel brought them in (cohort by marketing source), which pricing plan they started on, or which version of an onboarding flow they experienced. Grouping by acquisition channel, for example, can reveal that users from one channel retain far better than users from another — useful information for deciding where to invest marketing spend, not just how to improve the product itself.
Common mistakes when doing cohort analysis
- Comparing cohorts of very different sizes without noting it. A cohort of 20 users behaves much more erratically than a cohort of 2,000 — a single unusual user can swing a small cohort's numbers dramatically, which can be misread as a meaningful trend when it's really just noise.
- Only looking at retention, and never other cohort-based metrics. Cohort analysis works for revenue, feature adoption, and referral behavior too, not just whether someone stays active — limiting it to retention alone misses other useful patterns.
- Not connecting a cohort trend to an actual explanation. Spotting that March's cohort retained worse is only half the job — the real value comes from investigating and confirming why, then acting on it.
- Drawing conclusions from cohorts that are still too new to have long-term data. A cohort that signed up two weeks ago can't yet tell you anything about 3-month retention — comparing incomplete cohort data to fully mature cohorts produces misleading conclusions.
FAQ
What's the difference between cohort analysis and a retention rate? Retention rate is usually a single, blended number for a given period. Cohort analysis breaks that same underlying idea down by starting group, over time, which reveals trends and patterns a single blended retention number can't show on its own.
How many cohorts do you need before the analysis becomes meaningful? There's no fixed number, but comparing at least 3-4 cohorts of reasonably similar size makes trends easier to distinguish from random noise — a single cohort, or comparing wildly different-sized cohorts, makes patterns much harder to trust.
Can cohort analysis be used for revenue, not just user retention? Yes — revenue cohort analysis tracks how much revenue each signup group generates over time, which is especially useful for understanding LTV patterns and whether newer customers are becoming more or less valuable over time compared to older ones.
What tools are commonly used to build cohort analysis? Most product analytics platforms (like Amplitude, Mixpanel, or a company's own internal data warehouse and BI tool) have built-in cohort analysis features, since this is one of the most common and valuable ways product teams study user behavior over time.