Cohort Analysis for Marketing: How to Measure Retention in GA4

Short answer
What is cohort analysis in marketing?
Cohort analysis is a method that treats users who share a starting event in the same period as one group, then tracks whether that group returns in later weeks or months. In marketing, it reveals retention, repeat purchases and channel quality without the distortion of blended averages.
Cohort analysis lets you group customers who started at the same time and watch how each group behaves over the following weeks or months. An average retention rate often hides the truth. A cohort view shows which month, which channel and which campaign actually brought loyal buyers. In this guide, Talha Aslan and team walk through how we build, read and act on cohort tables in GA4.
For example, customers who placed their first order in March form one cohort. Customers who first ordered in April form another. So you compare the two groups at the same age, not in the same calendar month. In other words, you see what happened one month after the first order for each group, side by side.
This matters most when ad spend grows. New customer counts can rise while older customers quietly disappear. A total revenue chart hides that pattern. A cohort table exposes it.
Why does cohort analysis beat an average retention rate?
An average retention rate mixes customers who arrived at very different times. As a result, new and old customers cancel each other out. In a fast growing business, newcomers always dominate. So the average mostly reflects first week excitement.
A cohort view, on the other hand, judges each group at its own age. That makes these questions answerable:
- Are recent customers dropping off faster than older ones?
- Did the cohort after a price change perform worse?
- Has a new ad channel added volume but lowered quality?
- Did a loyalty campaign actually lift second orders?
Put simply, an average answers "how much," while a cohort answers "who, when and why." If you track churn as a single percentage, our guide to calculating churn rate explains where that number comes from. Cohort analysis then splits it into groups.
How do you read a cohort table?
A classic cohort table looks like a triangle. Rows show cohorts, meaning start periods. Columns show time since the start: week 0, week 1, week 2 and so on.
Read a row from left to right and you see one group fade over time. Read a column from top to bottom and you compare different groups at the same age. Reading along the diagonal also helps. It catches an event that hit every cohort in the same calendar week, such as a site outage.
The lower right corner stays empty. That happens because newer cohorts have not reached that age yet. Don't treat the gap as a problem. It only means time has not passed.
On a first pass, we suggest checking three things. First, how sharp the drop is from period zero to period one. Second, where the curve flattens. Third, whether the newest cohorts break away from the older ones.
Where is the GA4 cohort exploration and how do you set it up?
In GA4, the cohort view lives among the templates in the Explore section, not in the standard reports. According to Google Analytics Help, cohort exploration lets you study the behavior of user groups that share a common attribute over time. Menu labels can change, so look for the matching template in your property.
You can follow this order:
- Open a new cohort exploration in Explore.
- Pick the inclusion criteria, meaning the event that puts a user into a cohort.
- Pick the return criteria, meaning the event that counts as "came back."
- Choose cohort granularity: daily, weekly or monthly.
- Choose the calculation type and add a breakdown if needed.
- Set the metric you want in the Values field.
Each of these six settings changes the result. That's why our team writes the reason for every setting into a report note. Otherwise, three months later nobody remembers what the table measured. For a refresher on how the platform works, see our overview of GA4.
How should you choose inclusion and return criteria?
Google's help article lists these options for inclusion: first touch (acquisition date), any event, any transaction, any conversion, and a specific event. Return criteria offer similar options.
The right choice depends on your business question. For instance, a content site can include on first touch and return on any event. For an online store, that setup misleads. Browsing a site does not mean buying.
Here's a practical rule. Make the return criterion the behavior that earns you money. In a subscription product, that might be a renewal payment event. For a service firm, it could be a form submission or a quote request. Then the table answers "how many created value again," not just "how many came back."
Also check that your events fire correctly. If the event you choose fires inconsistently, the cohort table inherits the gap. Our article on missing GA4 conversions offers a checklist.
Should you use daily, weekly or monthly cohorts?
Granularity should match the speed of your purchase cycle. For fast moving products or mobile apps, daily and weekly views make sense. For products people buy a few times a year, a monthly view creates less noise.
Very fine granularity produces small groups. In a small group, one user can swing the rate by many points. That leads to false alarms. Very coarse granularity, however, can hide an important shift for a whole month.
We usually work in two layers. First, a weekly table catches early signals. After that, a monthly table confirms whether the pattern holds. In short, the weekly view asks "is something happening?" while the monthly view asks "is this a real trend?"
Also keep data retention in mind. Explorations depend on the data retention setting of your property. Check current options in your GA4 admin settings and the official help page.
What do standard, rolling and cumulative calculations mean?
GA4 cohort exploration offers three calculation types, and each answers a different question. Based on the official documentation, the difference looks like this:
| Calculation type | Who counts in a cell? | Best for which question? |
|---|---|---|
| Standard | Users who met the return criteria in that period only | How many people were active this week? |
| Rolling | Users who met the criteria in that period and every earlier period | How large is the core audience that never lapsed? |
| Cumulative | Users who met the criteria in any period up to that point, as a running total | How much value has the cohort built up so far? |
For example, if you pick a revenue metric with cumulative calculation in a purchase cohort, each cell shows revenue accumulated up to that age. That gives you an early estimate of customer lifetime value. Rolling calculation, meanwhile, paints a far more honest picture for subscription businesses where continuity matters.
How do you build a purchase cohort?
A purchase cohort groups customers by the date of their first order. For ecommerce and any business that relies on repeat sales, it's the most valuable cohort type.
In GA4, you can use "any transaction" as the inclusion criterion. The return criterion also becomes "any transaction." Then the table shows how many customers ordered again in each period after their first purchase.
There's a limit, though. GA4 recognizes users through cookies and user IDs. If a customer returns on a different device or after clearing cookies, GA4 may count a new user. So, where possible, build the purchase cohort from your order system's customer ID as well. Our team uses the GA4 table for direction and the order database table for decisions.
The revenue curve of a purchase cohort also underpins lifetime value math. We covered the formula side in our customer lifetime value guide.
Which ad decisions does channel cohort analysis change?
Channel cohort analysis splits customers by the channel that first brought them and compares how valuable each channel's customers become over time. In GA4, you do this by adding first user source or default channel group as a breakdown dimension.
This often flips the ranking in your ad dashboard. For example, one channel delivers cheap first orders, yet those buyers never return. Another channel costs more per first order, but its customers keep coming back for months. Optimizing only for the first order shifts budget to the first channel and loses money over time.
So we suggest asking these questions before moving budget:
- How much cumulative revenue did each channel's cohort produce by month three?
- Does the gap between channels open in month one or later?
- Are the breakdown groups large enough to support a decision?
To weigh that revenue against what you paid to acquire it, use our guide on reducing customer acquisition cost.
Who should interpret cohort analysis results?
Building the table is easy. The hard part is turning what you see into sound decisions. When a month's cohort looks weak, the cause could be seasonality, stock, price, ad creative or a tracking error. Separating those causes takes both data skills and field knowledge.
If nobody on your team owns this work, a digital marketing specialist can help you set up the first table and agree on a reading rhythm. At Talha Aslan and team, we build cohort tables as a decision tool tied to the ad and content calendar, not as a standalone report.
Still, you can start on your own. Pick one question, build one cohort table and check it on the same day each month. A steady habit teaches more than a complex dashboard.
How do you interpret a sample cohort table?
The table below is purely an example scenario. It does not belong to a real client, and the numbers exist only as a reading exercise. Picture an online store's monthly purchase cohorts and the number of customers who ordered again after their first purchase.
| Cohort (first order month) | New customers | Returned month 1 | Returned month 2 | Returned month 3 |
|---|---|---|---|---|
| Month A | 400 | 80 | 60 | 52 |
| Month B | 420 | 84 | 63 | (not yet) |
| Month C (discount campaign) | 700 | 70 | (not yet) | (not yet) |
A total new customer chart would call month C a big win. The cohort row, however, tells another story. Month C's month one return share is about half that of months A and B. In other words, the discount brought many one time buyers.
The takeaway is not "discounts are bad." The right move is to build a flow that encourages a second order for the discount cohort, then measure the result in the next cohort.
What are the most common cohort analysis mistakes?
Most mistakes we see in the field come from interpretation, not setup. Knowing them in advance lowers the risk of a wrong call.
- Comparing tiny groups. A breakdown with a few dozen users makes noise look like a trend.
- Comparing an immature cohort with older ones. Empty cells in a new cohort are not zeros.
- Leaving the return criterion on "any event." A single page view is not loyalty.
- Mistaking a tracking change for a behavior change. A tag update, a consent setup change or a site migration can break the table.
- Ignoring seasonality. Holiday cohorts behave differently by nature.
Also, if unassigned traffic runs high, the channel breakdown becomes unreliable. In that case, we recommend fixing GA4 unassigned traffic first.
How does cohort analysis relate to churn and CLV?
These three ideas show the same picture from different angles. Churn rate gives the share of customers you lose in a period as one number. Customer lifetime value estimates the total value a customer leaves over the relationship. Cohort analysis shows how both numbers form within each group.
For example, your overall churn rate may look steady while recent cohorts lose customers faster. Older, loyal customers keep the average up. So the single number stays quiet, while the cohort table warns you early.
The cohort curve also helps with lifetime value estimates. The point where the cumulative revenue curve flattens shows when most customer value forms. That way you can plan more realistically how long to wait before an acquisition cost pays back. On the marketing side, the flows we describe in our retention marketing strategy article act on that curve.
Which marketing actions should cohort analysis lead to?
A cohort table stays a pretty heat map until it drives action. We suggest pulling at least one test idea from every review. These pairings work well in practice:
- A sharp drop after period zero calls for a welcome email, a setup guide or a short reminder flow after the first order.
- A weak cohort from one channel calls for comparing that channel's targeting and ad promise with the site experience.
- Newer cohorts trailing older ones call for questioning recent product, pricing or delivery changes.
- A curve that flattens after a certain month calls for a loyalty offer for customers who pass that point.
Numbers alone won't tell you why customers stay or leave. The words in reviews and support tickets answer that. We cover this in our guide to turning voice of customer data into copy.
What data can you use for cohort analysis outside GA4?
GA4 offers a quick start, yet it doesn't need to be your only source. Your order system, CRM or email platform also store enough data to build cohorts.
The simplest method is a spreadsheet. First, pull the first order date and later order dates for each customer. Next, group customers by first order month and calculate how many months after the first order each purchase came. Finally, build the triangle with a pivot table.
This method gives you a precise customer identity. On the other hand, carrying channel data into order data takes extra work. So we suggest combining both sources: GA4 for channel questions, order data for loyalty questions.
When you share files that contain personal data, remove names and contact details. A customer ID and dates are enough for a cohort. Also flag returned and canceled orders in a separate column. Otherwise, a cohort with heavy returns looks more loyal than it really is.
How does cohort analysis work for B2B and service businesses?
In B2B and service businesses, repeat purchases are rare, so the cohort definition changes. Instead of "first order," you can use "first inquiry," "first meeting" or "contract start."
For example, group companies that filled out a quote form into monthly cohorts. Then track how many in each cohort moved to a meeting and how many signed. That is really your funnel laid out on a time axis.
As a result, the true length of your sales cycle becomes visible. A campaign may look weak in month one, then mature by month three. Without a cohort view, you risk switching it off too early.
With small volumes, quarterly cohorts work better than monthly ones. Otherwise each cell holds just a handful of companies, and the table can't support a decision.
How do consent settings affect cohort analysis?
When some users decline analytics cookies, GA4 handles them differently. As a result, your cohort table shows the measurable share of customers, not all of them. That doesn't make the table useless. It does remind you to read ratios and trends rather than absolute counts.
Cohort behavior can shift suddenly after you change your consent banner. That shift reflects measurement, not customers. So log every consent change with a dated note.
Also, keep personal data out of cohort reports. A cohort needs group behavior, not identity. We summarized the legal basics in our guide to building a GDPR compliant website. That content is not legal advice.
When do device and campaign breakdowns make sense?
Channel is the most common breakdown, but not the only one. Device category, landing page or campaign name also work as breakdowns. Each one tests a different hypothesis.
For instance, if the mobile cohort's second order rate clearly trails desktop, the problem may sit in checkout. If a cohort from one landing page looks weak, the page's promise may not match the product experience.
The rule is simple: use one breakdown at a time. Opening two dimensions at once shrinks cells fast. Start with your strongest hypothesis, confirm it, then move to the next dimension. That keeps the table readable and helps the team agree on one conclusion.
How should you present cohort analysis to leadership?
Showing the raw triangle in a leadership meeting often confuses people. Instead, pull a two or three sentence finding from the table and attach the table as evidence.
We use this structure in presentations. First, a one sentence finding. Then, the single row or column that proves it. Finally, the proposed test. A line like "The last two cohorts return less in month one; we're testing the delivery time change" drives a faster decision than a colorful heat map.
Also note how you built the table in every presentation. Without the inclusion criterion, return criterion and calculation type, any reading stays open to misunderstanding.
How does cohort analysis differ for subscriptions and repeat purchases?
In a subscription model, a customer either stays or leaves, so rolling calculation fits naturally. Each cell shows customers who kept paying without a break. That way you see the true core audience.
In a repeat purchase model, a customer can skip a month and come back the next. Therefore, standard calculation fits better. Rolling calculation would show a needlessly harsh drop for this model.
Hybrid businesses that sell both subscriptions and one time products should build two separate tables. Mixing both behaviors in one table blurs each of them. In short, choosing a calculation type without knowing your business model is the fastest way to reach a wrong conclusion from correct data.
Which metric should you track alongside cohort analysis?
A cohort table points a direction. It gains meaning once you tie it to the main goal of the business. That's why we suggest pairing it with one core metric the whole team follows.
For example, if your core metric is "customers who order at least once a month," the cohort table shows which groups feed that number. We cover this idea in our article on the north star metric.
Also watch leading signals. Small events such as add to cart, account creation or product page visits hint at how a cohort will behave later. Our guide on micro conversions in GA4 shows how to define them.
What is a quick checklist to start cohort analysis?
If you want to build your first cohort table this week, follow this list in order. Each item stands for a decision. Skip one, and the table answers the wrong question.
- Write down the single question you want answered.
- Define the inclusion event and the return event.
- Choose a granularity that fits your purchase cycle.
- Pick the calculation type that fits your question.
- Add one breakdown, such as channel or device.
- Note the date of your last tracking change.
- Put the review date on the calendar and write your first test idea.
If your team can't sustain the setup, review and action loop, it may be time to hire a digital marketing expert who can speed it up. At Talha Aslan and team, we plan every step from tracking setup to ad decisions with you. We don't promise results, but we show the data behind every decision openly.
Core guide: What Is Digital Marketing? Its types, examples and how it is done, on one page with an interactive film that travels through the brain.



