Cohort Analysis for Marketers: See Who Sticks Around

Averages are comfortable and often wrong. A report that says customers buy about twice a year hides the fact that people who joined last spring behave nothing like those who joined during a heavy discount. Cohort analysis fixes that by comparing groups of customers who share something in common.

This guide explains what a cohort is, how to build a simple table in a spreadsheet, how to read it, and what decisions it can improve.

What a cohort is

A cohort is a group of people who share a starting point. The starting point is usually a date, such as the month they first bought or signed up, but it can also be the channel that brought them, the first product they purchased or the campaign they responded to.

Once you have groups, you follow each one over time and ask the same question of every group: how many were still active after one period, two periods, three periods, and so on?

This is different from a normal monthly report, which mixes old and new customers together. If total repeat purchases rise one month, you cannot tell whether loyalty improved or whether a large batch of new customers simply arrived. Cohorts separate those effects.

Questions cohorts answer well

• Do customers acquired this season stay as long as earlier ones?
• Which acquisition channel produces people who come back?
• Did a change to onboarding, pricing or the product improve retention?
• When do customers typically lose interest, and what could we do just before that point?
• Do buyers of a particular first product return more often?

These questions connect directly to customer lifetime value, since cohorts show the shape of that value over time.

Types of cohorts

Time-based cohorts

Group customers by the month or week of their first action. This is the classic version and the best place to begin. It reveals whether newer customers are stronger or weaker than older ones.

Channel or campaign cohorts

Group by where they came from, for example search, social, referral or a particular promotion. This links retention to your marketing spend. Clean campaign tagging is essential here, so review the UTM parameters guide first.

Behaviour-based cohorts

Group by what people did early on, such as completing a profile, using a feature, opening the first three emails or buying a specific product. These help you find the early actions that predict loyalty.

Build a simple cohort table

You can do this in a spreadsheet with no special software.

1. Export your data. You need one row per customer or order, with a customer identifier, the date and, if relevant, the amount.
2. Find each customer's first date. Assign them to a cohort, such as the month they first bought.
3. Calculate the period number for every later order. If someone first bought in January and ordered again in March, that order falls in period two.
4. Count the distinct customers active in each period for each cohort. A pivot table does this neatly.
5. Convert counts to shares of the starting group. Dividing by the cohort's original size makes cohorts of different sizes comparable.
6. Colour the table. Shading cells from light to dark makes patterns obvious at a glance.

The result is a triangle: older cohorts have more periods of data, newer ones fewer.

How to read the table

Read it in two directions.

Across a row shows how one cohort fades. A steep drop after the first period suggests that something goes wrong early, such as a disappointing first experience or a mismatch between promise and product. A long flat line means customers who stay tend to stay.

Down a column compares cohorts at the same age. If the second-period column gets lighter in newer rows, retention is worsening. If it gets darker, something you changed is working.

Watch the shape

• Early cliff, then flat: you lose people quickly but keep the committed ones. Focus on the first experience.
• Gradual slope: steady loss. Look at ongoing engagement and reminders.
• Bumps at regular points: seasonal buying or renewal cycles. Plan around them.
• Recovery after a campaign: a win-back effort may be reviving lapsed customers.

A hypothetical example

Imagine a small online store selling coffee beans on subscription. The owner builds monthly cohorts and notices that customers who joined through a steep first-box discount drop away sharply after the second box, while those who joined through a friend's referral keep ordering. The discount cohort looks great on the sign-up chart, but the referral cohort is more valuable over time. The owner decides to put more effort into a referral programme and to test a gentler offer, then watches the next cohorts to see if the pattern changes.

The story is invented, but it shows how cohorts can reverse a decision made from sign-up numbers alone.

Turning findings into action

A cohort table is a diagnostic, not a solution. Pair each pattern with a specific test.

1. Weak first period: improve onboarding emails, add a helpful first-use guide, or follow up personally. The customer onboarding emails post gives ideas.
2. Decline at a predictable point: send a timely reminder or offer just before it.
3. Channel differences: shift budget toward the sources that produce loyal customers, within reason.
4. Good behaviour signals: if people who take a certain early action stay longer, encourage that action sooner.

Test one change at a time and compare the next cohort against earlier ones. The A/B testing for marketers guide explains how to test sensibly.

Pitfalls to avoid

• Cohorts that are too small. A handful of customers produces noisy, misleading shapes. Use wider time bands, like quarters, when volumes are low.
• Comparing unequal ages. Always compare cohorts at the same period number.
• Defining activity loosely. Decide what counts as "active" before you build the table and keep it consistent.
• Changing several things at once. You will not know what moved the line.
• Ignoring seasonality. A December cohort may behave differently for reasons unrelated to your work.

Fit cohorts into regular reporting

Cohort tables suit a quarterly review better than a weekly one. Keep one standing page in your reporting that shows retention by cohort and by channel, and add a note whenever you launched something that might affect it. The marketing funnel metrics guide and funnel drop-off analysis show how to look at the steps before purchase, while cohorts show what happens after.

Choosing the right time unit

The size of each period changes what you can see. Weekly periods suit fast-moving products such as food delivery or apps, where behaviour shifts within days. Monthly periods suit most small shops and service firms. Quarterly periods fit businesses with long buying cycles, like home improvement or training courses, where a customer might only return once a year.

A good rule is to pick a period roughly as long as your normal gap between purchases. If customers typically reorder every six weeks, monthly periods will show the rhythm clearly, while weekly ones will look jagged and daily ones will look empty.

Measure more than head-count retention

Retention counts people, but you can also follow money. Instead of asking how many customers from the January cohort bought again, ask how much revenue that cohort generated in each later period. A cohort that keeps fewer people but spends more per order may still be your best group.

Two other versions are worth trying:

• Orders per customer by period. This shows whether loyal customers buy more often over time.
• Average spend by period. This reveals whether customers grow into larger purchases or shrink back to small ones.

Comparing these views side by side usually tells a richer story than any single table.

Share the results simply

A cohort table can look intimidating to people who have not seen one. When presenting it to an owner or client, give them one sentence on how to read it, one sentence on the main pattern and one recommendation. The marketing report storytelling post shows how to frame findings that way.

Takeaway

Cohort analysis for marketers replaces blended averages with honest comparisons. Group customers by start date, channel or early behaviour, follow each group over time, and read the table across and down. Then turn each pattern into one small test.

If you would like help setting up a cohort view for your own customer data, contact Kavin or download something useful from the resources page.