Cohort analysis groups users or customers who experienced the same event in the same period, such as their first purchase in the same month. It then follows each group separately: how many are still active after a week, a month or a year?
Cohort analysis follows groups that started at the same time: acquisition and behavioural cohorts, the retention table, churn and GA4.
What cohort analysis means
The term comes from demography and sociology. In 1965, Norman Ryder defined a cohort in the American Sociological Review as the aggregate of individuals, within some population definition, who experienced the same event within the same time interval. Until then, he wrote, the defining event had almost always been birth, but that was only a special case.
In marketing, the event is usually the start of the customer relationship. Google Analytics describes a cohort as a group of users who share a common characteristic, and gives as an example all users with the same acquisition date.
Two kinds are distinguished. Acquisition cohorts group by the time of first contact, such as the first visit, sign-up or first purchase. Behavioural cohorts group by an action. Amplitude describes a behavioural cohort as users grouped by the actions they take within a specific time window, such as watching three episodes on the day they signed up. The first kind shows whether newer customers stay longer than older ones. The second shows whether a particular behaviour goes along with longer retention.
How it works
The result is usually a retention table. Each row is a cohort, identified by its start period and its size. Mixpanel describes it the same way: the first column shows the day, week or month in which users performed the starting event, and another column shows how many there were. The following columns count the periods since the start. In Google Analytics, for example, the "day 1" column of the 1 January cohort shows how many of those users returned on 2 January.
An example with made-up numbers: in January, a shop gains 400 new customers. The following month, 120 of them buy again, or 30%. Of the 500 new customers in February, 175 buy again the following month, or 35%.
The table is read in two directions. Along a row, you see how fast a cohort shrinks. Down a column, you compare cohorts at the same age, in the example the January and February cohorts after one month. If the value rises from row to row, newer customers are being retained better. Cells for the youngest cohorts are often not yet complete because their period is still running. Mixpanel marks such cells separately.
GA4 offers cohort exploration for this. There you set the inclusion criterion (for example first visit, first transaction or a specific event), the return criterion and the granularity: daily, weekly from Sunday to Saturday, or monthly. There are three ways of calculating. Standard counts in each cell the users who meet the return criterion in that particular period, regardless of the others. Rolling counts only those who meet it in that period and all previous ones. Cumulative counts those who meet it in any period of the exploration. Google names further limits: a maximum of 60 cohorts, cohorts based on device data only without User-ID, and thresholding on demographic dimensions when a cohort is too small. How far back an exploration reaches also depends on data retention. For user-level data in standard properties, it can be set to 2 or 14 months.
Example with made-up numbers from the text, other cells left without a value
↓Column: cohorts at the same age: January against February after one month, 30% against 35%. If the value rises, newer customers are being retained better.
Hatched: that month is not over yet
Reading down compares customers of the same age.
Why it matters
A total figure mixes old and new customers. If the churn rate rises in a given month, the cause may be lower-quality new customers, a price change for everyone, or simply that the base held an unusually large number of young customers that month, if young customers leave more often. Cohort analysis separates these cases because it shows each group at its own age.
Customer lifetime value also depends on how long customers stay. The retention table supplies the share still active after each period, broken down by start month or by channel. That makes it possible to check whether customers from a particular campaign are still buying after three months or came only once.
The difference between returning and staying matters. In GA4's standard calculation, a user can be absent in week 2 and reappear in week 3. Someone who does not return in a period has not cancelled. For statements about churn, the rolling calculation fits better, or an analysis from your own CRM, where cancellations are recorded as events.
Best practices
Before analysing, decide which event starts a cohort and what counts as a return.
Compare shares, not absolute numbers, when cohorts differ in size.
Choose the granularity to match the purchase rhythm: daily for apps, monthly for infrequent purchases.
Leave the incomplete cells of young cohorts out of the comparison.
Split cohorts by channel or campaign when you want to assess acquisition.
Check GA4 data retention before analysing cohorts over more than two months.
Common mistakes
Equating a return in one period with retention.
Comparing the values of the most recent cohort with older ones while its period is still running.
Reading cohorts of a few dozen customers as a trend.
Overlooking that GA4 cohorts are based on devices and do not take User-ID into account.
Reading a drop in overall retention as a deterioration without looking at the cohorts separately.
Manuel Riveiro RodriguezCEO & Digital Strategist
A technical audit covers this and everything else in one pass.
Cohort analysis groups users or customers who experienced the same event in the same period, such as their first purchase in March. It then shows, separately for each group, how many are still active after each week or month. This makes it possible to compare older and newer customers fairly, each at the same stage of their relationship.
What is the difference between acquisition and behavioural cohorts?
Acquisition cohorts group by the time of first contact, for example the month of sign-up or first purchase. Behavioural cohorts group by an action, for example everyone who used a particular feature in their first week. The first kind compares generations of customers, while the second shows which behaviour goes along with longer retention.
How do you read a retention table?
Each row is a cohort and each column a period since the start. Along a row, you see how fast a group shrinks. Down a column, you compare cohorts at the same age, for example all of them in their third month. The most recent rows often contain incomplete cells because their period has not finished yet.
How are cohort analysis and churn related?
Cohort analysis shows when customers leave, broken down by start month or channel. A single churn rate for the whole customer base hides this because it mixes old and new customers. But someone who does not return in a period has not necessarily cancelled. In GA4, a user can reappear later.
How do you create a cohort analysis in GA4?
According to Google's help, you open the template gallery in Explore and choose cohort exploration. There you set the inclusion criterion, such as first visit, the return criterion, the granularity and the calculation type: standard, rolling or cumulative. Google limits it to 60 cohorts and builds them from device data, without User-ID.
