GA4 Cohort Analysis: Setup, Retention and Calculation Examples

Cohort analysis follows a group that shares a starting condition and examines what happens afterward. In GA4, define inclusion, return behavior, value and calculation method before interpreting the table. A fictional four-user example makes standard, rolling and cumulative calculations easier to distinguish.

Define the cohort and the behavior to follow

For example, group users first acquired in the same week and follow their subsequent behavior. This separates new acquisition from what happens to an existing group, which a single total-user trend can obscure.

The official Cohort exploration guide explains inclusion and return criteria. Inclusion means joining the analytical cohort; it does not necessarily mean registering an account. The resulting pattern helps formulate questions, but does not automatically reveal why users left or which intervention caused a change.

Choose inclusion and return criteria before opening GA4

The standard “Any transaction” inclusion criterion uses the first transaction within the exploration date range, not a guaranteed lifetime first purchase; the table’s initial-purchase cohort therefore requires a separately verified first-purchase event or group based on the full purchase history.

These are design examples, not claims that a particular website already implements those events. Verify success conditions, duplication, consent and production hosts before relying on purchase or registration data. Button clicks are not automatically successful customers.

Configure granularity, breakdown and values in Explore

Open Cohort exploration in Explore and select dates, inclusion and return criteria. Then choose daily, weekly or monthly granularity, the calculation, any breakdown and the displayed value. Confirm available options in the actual interface.

• Use dates with enough completed follow-up time.
• Fix whether the value represents users, events or revenue.
• Save the breakdown dimension and its reporting meaning.
• Keep settings together with the resulting table.

Weekly granularity uses Sunday through Saturday, not a rolling seven-day window beginning with each user’s first contact. Someone joining near the week boundary can have a shorter week zero, which matters when comparing behavior.

Calculate standard, rolling and cumulative cohorts

The following fictional cohort includes four first-touch users in week zero and uses purchase as the return criterion. Yes means the user met that criterion during the period.

Standard concerns the current period, rolling requires every period since inclusion, and cumulative includes at least one qualifying period since inclusion. Rates here are hand calculations with an explicit denominator, not a promise that every GA4 cell automatically displays this percentage.

Read week zero and compare mature observation windows

Only A and C purchase in week zero. Dividing those two by the four first-touch users gives 50%, not 100%. Week zero depends on the inclusion criterion, return criterion and selected value.

Likewise, a recent cohort’s unfinished week two should not be compared with an older cohort’s completed week two. Keep incomplete or unavailable cells distinct from measured zeros. Compare the same mature elapsed window and record both calendar boundaries and report dates.

Separate users, event counts and retention rates

One user purchasing twice can represent one purchaser but multiple events. Changing the value does not automatically produce any desired conversion rate. Use the GA4 metric dictionary and explicitly define numerator, denominator and period.

For a fixed cohort, rolling user counts cannot increase as the required sequence lengthens, but they can remain unchanged. They do not necessarily fall every period. Revenue or another selected value should not inherit user-count monotonicity without examining its definition.

Account for device identity, retention and small cohorts

GA4 cohort construction does not consider User-ID and has device-based limitations. A person changing devices is not guaranteed to appear as one continuous user. Record consent and identification changes alongside the analysis.

Explore history is affected by data retention settings. Extending a setting cannot restore already-deleted data. In a small cohort, one user can move a rate substantially, so show counts with percentages. Check breakdown coverage, thresholds and quality indicators rather than inventing reasons for unavailable values.

Turn observations into hypotheses and verification plans

An observation such as fewer purchasers in week one can suggest several hypotheses: unclear guidance, different acquisition intent or measurement changes. Comparing acquisition periods alone does not establish the causal effect of a campaign.

For a fictional guidance improvement, fix the audience and success condition before following repeated use. Use the cohort table to choose questions and candidate timings for a test. It does not guarantee optimal timing or improved retention.

Continue with the guide that matches your next task: Time on Page vs GA4 Engagement Time: How to Measure and Interpret It; Owned Media SEO: Plan, Maintain, and Evaluate Your Content; GA4 Free vs Analytics 360: Limits, Costs and When to Upgrade; What Is CPA? Formulas, Target Setting and Measurement Checks.