A cohort exploration is a report in GA4’s Explore section that groups users by when they first did something, usually their first visit or first purchase, and then shows how many in each group came back, bought again or took another chosen action in the days, weeks or months that followed. It answers questions such as whether customers who first arrived in January return as often as those who arrived in March.
How a cohort exploration works
You choose four settings:
- Cohort inclusion What puts a user into a group. First touch, meaning their first visit, is the default; you can also use any transaction, any key event or a specific event such as a first purchase or sign-up.
- Return criteria What counts as coming back, for example any event, any transaction or one specific event.
- Granularity Daily, weekly or monthly groups.
- Calculation Standard counts anyone active in each period; rolling counts only users who were active in every period up to that one; cumulative adds up the totals over time, which suits purchases and revenue.
The result is a grid. Each row is a cohort, such as everyone who started in a given week; each column is a period after that start; each cell is shaded by how high its value is. You can change the metric, for example to active users, user retention, transactions or purchase revenue, and break the cohorts down by a dimension such as first user source or device category.
Why it matters
Most GA4 reports blend new and existing visitors into one total, which hides whether people stick around. A cohort view pulls them apart. For a subscription business, a restaurant group’s ordering app or an online shop that relies on repeat custom, the share of customers who come back in months two and three often deserves more attention than this month’s traffic, because it shows what each new customer is likely to be worth.
Breaking cohorts down by acquisition channel turns it into a budget decision. If customers first won through paid social return far less often than those from organic search or email, their real cost is higher than the ad platform’s cost per acquisition suggests, and the channels deserve to be valued differently. The same figures feed straight into estimates of customer lifetime value.
Common mistakes
- Leaving data retention at two months. Explorations can only use data inside your property’s data retention period. A new GA4 property keeps it for two months unless someone changes it, which makes monthly cohorts almost useless. Raising it to 14 months on standard GA4 helps from that point on; it does not bring back data already deleted.
- Reading cohorts as people. GA4 users are mostly browsers recognised by their client ID, so a customer who returns on another device, or after Safari has expired the cookie, looks like a lost user plus a new one.
- Using first touch when you care about customers. Retention among everyone who ever visited is dominated by one-off browsers. To study repeat purchase, set inclusion to the first transaction.
- Over-reading small cohorts. A weekly cohort of forty users can swing wildly from one period to the next. Use monthly groups or a longer date range.
- Missing thresholding. When Google signals is on, GA4 can withhold rows with low counts, which blanks out cells in small cohorts.
How to act on it
- In GA4 Admin, open Data collection and modification, then Data retention, and set event data retention to 14 months if it is not already.
- In Explore, start a cohort exploration with inclusion set to the first purchase or first key event, return criteria set to a repeat of the same action, monthly granularity, and at least six months of data.
- Break it down by first user default channel group and compare the second- and third-month figures for each channel.
- Take the differences into your budget discussion: a channel whose customers come back deserves more room than its first-order cost suggests.
This is the sort of evidence that should shape where next quarter’s money goes, and it is part of how a written digital marketing strategy arrives at its budget split between channels.