How to Run Cohort Analysis for Retention

Your total customer count can grow every month while the business quietly leaks users out the back door. Cohort analysis is how you see that leak. This guide shows you how to group customers by when they started, build a simple retention table in a spreadsheet and turn the patterns into decisions your team can act on this quarter.

Most founders track headline numbers: total users, monthly revenue, new signups. Those numbers matter, but they blend old and new customers together. If you signed up 300 people last month and lost 250 existing ones, your total still went up. It just does not feel like a business that is about to break.

Cohort analysis separates those groups so you can see how each batch of customers behaves over time. It sounds technical, but the core idea fits on a napkin, and you can build your first version in an afternoon with data you already have. Late May is a handy moment to do it, too: you have roughly five months of 2027 cohorts to compare before mid-year planning begins.

What cohort analysis actually tells you

A cohort is simply a group of people who share a starting point. In retention work, that starting point is usually the month or week someone signed up, made a first purchase or started a paid plan. You then follow each group forward and ask a simple question each period: how many of them are still here?

When you lay those answers out in a grid, patterns jump out that a single growth chart would hide.

• Early drop-off. If most of a cohort disappears in the first period, your onboarding or first experience is likely the problem, not the product's long-term value.
• Where the curve flattens. A retention curve that levels off means you have found a core group who keep getting value. A curve that keeps sliding toward zero means you have not yet.
• Cohort-to-cohort improvement. If customers who joined in April stick around better than those who joined in January, the changes you shipped in between are probably working.
• Seasonal or channel effects. A cohort that came from one promotion or one channel may behave very differently from the rest, and you will only see it when the groups are separated.

In short, cohort analysis answers "is the product getting better at keeping people?" rather than "are we getting bigger?" Both questions matter, but the first one tells you whether growth will last.

Decide what "retained" means before you touch the data

This is the step most teams rush, and it shapes everything that follows. A customer can be "retained" in several ways, and each one tells a different story.

Logo retention

Is the customer account still active or still paying? This is the cleanest measure for subscription businesses and for anyone selling to other businesses. You count accounts, not usage.

Activity retention

Did the person do something meaningful in the period? For an app, that might be completing a core action such as sending an invoice, booking a class or publishing a post. Avoid weak signals like "opened the app" or "logged in," which can include people who arrived, felt confused and left.

Revenue retention

How much of the cohort's starting revenue is still coming in? This view captures upgrades, downgrades and cancellations together. It can rise above 100% if remaining customers expand, which is why it is often tracked alongside logo retention rather than instead of it.

Repeat purchase retention

For shops, cafes and service businesses without subscriptions, a "retained" customer might be one who bought again within the period. Pick a period that matches how often people naturally buy. Monthly may suit a coffee subscription, while quarterly may fit a hair salon or a print shop.

How to build a cohort retention table step by step

You do not need a data team for this. If you can export a list of customers with a start date and a list of activity dates, you can build the table in any spreadsheet tool.

1. Export two lists. First, every customer with a unique ID and their start date. Second, every qualifying event (payment, booking, order or core action) with the customer ID and event date.
2. Assign each customer a cohort. Convert the start date into a cohort label, such as "2027-01" for January. Weekly cohorts suit fast-moving consumer apps, while monthly cohorts suit most small businesses.
3. Work out each event's period number. For every event, calculate how many months (or weeks) after the customer's start it happened. The start month is period 0, the next month is period 1 and so on.
4. Count unique active customers. Build a pivot table with cohort labels as rows, period numbers as columns and a count of unique customer IDs as the values.
5. Convert counts into percentages. Divide each cell by the size of that cohort in period 0. Period 0 becomes 100%, and later columns show the share still active.
6. Add conditional formatting. Shade higher percentages darker. The resulting "triangle" makes patterns easy to read at a glance.
7. Chart a few rows as lines. Plot three or four cohorts on one line chart so you can compare their curves directly.

Your table will look like a staircase, because newer cohorts have had less time to age. That is normal. Do not compare a two-month-old cohort's month six figure with anything, because it does not exist yet.

A cohort retention template you can copy

Paste this layout into a spreadsheet and replace the numbers with your own. The figures below are made-up examples for illustration only.

The "Notes" column is the secret weapon. Each time you change pricing, onboarding, support or marketing, write it next to the cohort that first experienced it. Months later, you will be able to connect the dots between what you did and how retention moved.

Reading the curves: what good and bad look like

Once your table exists, resist the urge to compare your numbers with figures you have seen online. Retention varies hugely by product type, price, customer size and how "retained" is defined. The most useful comparison is your own cohorts against each other.

Signs things are working

• The curve drops early and then flattens into a steady line.
• Newer cohorts sit above older ones at the same period number.
• Revenue retention for older cohorts holds up or rises as customers upgrade.

Signs something needs attention

• The curve never flattens and keeps sliding period after period.
• One cohort, often a big promotional one, behaves much worse than its neighbors.
• Retention looks fine but the cohorts are getting smaller, which points to an acquisition problem rather than a product one.

Slicing cohorts to find what drives retention

A basic table tells you that retention changed. Slicing tells you why. Once the core table is working, try splitting cohorts by one attribute at a time.

• Acquisition channel. Do customers from referrals stay longer than those from paid ads or a marketplace listing?
• Plan or price point. Do annual plans, higher tiers or certain bundles keep people longer?
• First action. Do customers who complete a specific setup step in week one retain better? This is often where the biggest insight hides.
• Customer type. Freelancers, small teams and larger accounts may behave very differently.
• Region. If you sell in both the US and Europe, holidays and buying habits can shift patterns.

Keep the slices few and the questions specific. Ten slices at once usually produce noise, not insight. And remember that a pattern is a clue, not proof. If customers who use a feature retain better, it might be because the feature creates value, or because your most committed customers were always going to find it. A small experiment, such as nudging new users toward that feature, helps you tell the difference.

Turning the analysis into action

A cohort table that nobody acts on is just a pretty spreadsheet. Build a light routine around it so it shapes real decisions.

• Update monthly. Add the newest cohort and the latest period for every existing cohort. Many teams do this in the first week of the month.
• Share one chart, one sentence. In your team meeting, show the line chart and say one thing that changed. For example: "March customers are holding better at month 2 than any cohort so far."
• Pick one retention lever per quarter. Onboarding, early support, pricing, product education or a win-back email. Focus beats scattering effort.
• Log every change in the Notes column. This is how you learn which bets paid off.
• Talk to customers in both groups. Interview a few who stayed and a few who left. Numbers show where to look, and conversations show what to fix.

If you later move to an analytics tool, the logic stays the same. You will simply automate the export and pivot steps. Many product analytics and billing tools include cohort views, so check what your current stack already offers before adding anything new.

Frequently asked questions

The bottom line

Cohort analysis is one of the most honest views you can have of your business. By grouping customers by start date, defining "retained" clearly and tracking each group over time, you can see whether your product is getting better at keeping people, and which changes actually helped.

Block out an afternoon this week, export your data and build the first table in a spreadsheet. Add a Notes column, share one chart with your team and pick a single retention lever for next quarter. Small, steady improvements in those curves compound into a much stronger business.