Cohort Analysis
Cohort analysis compares groups that share a starting event or characteristic to reveal how behavior and outcomes change over time.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/cohort-analysis/
01What it means
Cohort analysis groups people or accounts by a shared, time-bounded characteristic and follows their subsequent behavior. Acquisition cohorts might share a signup week; behavioral cohorts might share completion of a workflow. Unlike a static segment, a cohort often preserves membership so analysts can compare trajectories such as retention, conversion, feature adoption, or revenue across comparable groups.[1][2][3]
| Cohort | Entry | +1 period | +2 periods | +3 periods |
|---|---|---|---|---|
| Cohort A | Period 1 | Period 2 | Period 3 | Period 4 |
| Cohort B | Period 2 | Period 3 | Period 4 | Period 5 |
| Cohort C | Period 3 | Period 4 | Period 5 | Period 6 |
Read down an age column to compare cohorts at the same point in their lifecycle; the grid shows alignment, not performance data.
02Why it matters
Aggregate averages mix users with different tenure, contexts, and product experiences. Cohorts separate those histories, making it easier to see whether newer customers retain better, whether a release affected exposed groups, or whether a behavior is associated with later success. The method turns a single blended metric into a pattern that teams can investigate.
03When to use it
Use cohort analysis when timing, exposure, or prior behavior may explain different outcomes. It is valuable for retention studies, onboarding changes, channel comparisons, and longitudinal feature analysis. Define cohorts before inspecting too many cuts when the work is intended to support a hypothesis or decision.
04How it works
- State the question and choose a cohort-entry event or characteristic that is relevant, observable, and time bounded.
- Set membership rules, grain, identity unit, outcome event, follow-up intervals, and eligibility criteria before calculating results.
- Build a cohort table or curves showing the proportion or value observed at each comparable age.
- Compare cohorts with adequate sample sizes and annotate releases, campaigns, pricing changes, and data-quality incidents.
- Investigate notable differences with segmentation, research, and experiments while accounting for seasonality and changing population mix.
05A practical example
A budgeting app compares weekly signup cohorts after introducing bank-connection guidance. Newer cohorts show higher week-four retained budgeting, particularly among first-time budgeters. Because a seasonal campaign also changed acquisition mix, the team segments by source and runs a controlled onboarding test before attributing the improvement to guidance.
06Where it breaks
Cohort patterns are descriptive and can reflect selection effects, seasonality, external events, or instrumentation changes. Recent cohorts have incomplete follow-up, creating right-censoring. Fine segmentation produces unstable small groups and tempting stories. Behavioral cohort membership can also use information that occurs after the supposed cause, introducing leakage.
Common mistakes
- Comparing calendar-period totals instead of cohorts at the same age since entry.
- Creating many post hoc cohorts until an appealing difference appears.
- Ignoring incomplete follow-up, identity resolution, sample size, or acquisition-mix changes.
07Sources
- Cohort explorationGoogle Analytics Help · Authoritative analytics documentationCohort exploration
- Identify users with similar behaviorsAmplitude · Product analytics documentationIdentify users with similar behaviors
- Cohort analysisMixpanel · Professional analytics guidanceCohort analysis