Reading a Retention Curve Starts With Cohort Analysis

What is cohort analysis?

In product and revenue reporting, cohort analysis is the practice of grouping users by a shared starting event, such as the week they signed up, and then tracking that same group's behavior over time instead of measuring the whole user base as one pool. A cohort is one such group, and a cohort table lines several cohorts up side by side so a team can compare how each one behaved in its first week, its fourth week and its twelfth week. The output that makes cohort analysis useful is the retention curve, which plots the share of a cohort still active in each period after the starting event.

What is a cohort?

A cohort is a group of users who share one defining event that happened in the same window, most often a signup date, a first purchase date or a first use of a specific feature. Every user in a cohort gets measured on the same clock, counting periods since that shared event rather than by calendar date, so a user who signed up in week one and a user who signed up in week ten are both read at "week 4 since signup" on the same axis. That shared clock is what makes cohorts comparable to each other, since two cohorts can be lined up period by period even though they started on different calendar dates.

What is the difference between an acquisition cohort and a behavioral cohort?

An acquisition cohort groups users by when they arrived, such as everyone who signed up during the same week, while a behavioral cohort groups users by what they did, such as everyone who completed onboarding or everyone who upgraded to a paid plan. Acquisition cohorts answer whether a product is getting better or worse at keeping the people it brings in over time, since each new signup week becomes its own line on the chart. Behavioral cohorts answer whether a specific action changes what happens afterward, since a team can compare the retention curve of users who took that action against the retention curve of users who did not. Both cohort types run on the same retention math; only the grouping rule that sorts users into rows changes.

How is retention calculated inside a cohort?

Retention for a cohort in a given period is the number of users from that cohort still active in that period divided by the number of users the cohort started with.

cohort retention rate = active users in period / users in original cohort x 100

A cohort of 500 users who signed up in January, with 175 of them still active in month three, has a month-three retention rate of 35%.

How do you read a retention curve?

A retention curve plots a cohort's retention rate against periods since the starting event, and the shape of that line, not any single point on it, is what tells a team whether the product holds on to users. Most curves drop sharply in the first few periods as users who tried the product once and did not come back leave the cohort, then bend and flatten into a slope that is much shallower than the opening drop. Read the flattening point as the closest thing to a stable base: a curve that keeps falling in period ten has an ongoing churn problem, while a curve that has leveled off by period four has settled into a group of users who keep coming back.

Cohort A and cohort B both settle into a flat tail, but B loses fewer users on the way there and settles at a higher floor, which is the difference a single average retention rate would never surface.

Why does an average retention rate hide what a cohort table shows?

A single average retention rate hides the pattern because it blends cohorts that are at completely different points on their own curve into one number, so a business that is improving and a business that is declining can report the same average by coincidence. A user acquired eight weeks ago who has already flattened into the stable tail counts the same as a user acquired last week who is still in the steep opening drop, even though their retention behavior means different things for the product's future. Two products can show an identical 45% average retention this month while one has cohorts stabilizing at 53% and the other has cohorts still falling past 53% toward an unknown floor, and only the cohort table tells those two situations apart.

What should you do after reading a cohort table?

Compare recent cohorts against older ones in the same column to see whether retention is trending up or down at each period, since a newer cohort scoring higher than an older cohort at week 4 is early evidence that a recent product change helped. Segment behavioral cohorts against each other, such as users who completed a specific onboarding step against users who skipped it, to test whether that step is the reason for a stronger curve rather than a coincidence. Pair the cohort table with funnel data from the same period, since a cohort that drops early often stalled at a specific step rather than losing interest all at once.

How does cohort analysis connect to churn and engagement?

A cohort's retention curve and its churn rate describe the same users from opposite directions, since the users a cohort loses between two periods are exactly the users counted as churned in that window. Reading retention curves next to engagement rate and the underlying customer journey shows not just how many users a cohort kept but what those users actually did before the ones who left stopped coming back. Flowsery's real-time analytics tracks user journeys and goal events as they happen, so the activity feeding a cohort table comes from the same data as the rest of a team's reporting instead of a separate export.

Frequently Asked Questions

What is the simplest definition of cohort analysis?

Cohort analysis groups users by a shared starting event, such as their signup week, then tracks that same group's behavior forward in time instead of averaging the whole user base together. The output is usually a retention curve or a cohort table showing how each group's activity changes period by period.

What is the difference between a cohort and a segment?

A cohort is defined by a shared starting event and is tracked forward from that fixed point in time, while a segment is any group defined by a shared trait, such as country or plan type, that can be measured at any single point in time. A cohort always has a time axis built in; a segment does not need one.

How many users do you need before cohort analysis is useful?

There is no fixed minimum, but a cohort with too few users produces a retention curve that swings sharply from one user leaving or staying, making the pattern hard to trust. Wait until weekly or monthly cohorts each contain enough users that one person's behavior does not move the whole line before drawing conclusions from the shape.

Why does a retention curve drop sharply at the start and then flatten?

The sharp early drop reflects users who tried the product once without finding a reason to return, while the flatter tail that follows represents the smaller group who found ongoing value and kept coming back. Reading where a curve flattens, not just how far it fell, tells a team whether it has found a stable core of users.

Can two cohorts have the same average retention but different curves?

Yes, one cohort can drop fast early and flatten at a low floor while another drops slowly and settles at a higher floor, yet still average out to a similar overall retention figure across the whole period measured. Only the period-by-period cohort table, not a single average, shows which of those two situations is happening.

How is an acquisition cohort different from a behavioral cohort?

An acquisition cohort groups users by when they arrived, such as their signup week, and is used to see whether retention is improving release over release. A behavioral cohort groups users by an action they took, such as completing onboarding, and is used to test whether that specific action changes the retention curve compared to users who did not take it.

How long should each period in a cohort table be?

Most cohort tables use weekly or monthly periods based on when the shared starting event happened, then track each cohort forward on that same clock. Weekly periods catch the steep early drop in more detail, which matters for a product where churn shows up fast. Whatever length a team picks, every row in the table has to use it, or the columns stop lining up.

Can the same average retention number mean two different things for a business?

Two products can both report a 45% average retention rate this month while one has cohorts flattening at 53% and the other has cohorts still falling past 53% toward an unknown floor. The average treats both businesses as identical, but only the cohort table shows which one has found a stable base and which one is still losing ground.

Why pair a cohort table with funnel data?

A cohort table shows that retention dropped, but not where in the product that drop happened. Pairing it with funnel data from the same period shows whether a cohort that drops early stalled at a specific step, rather than simply losing interest all at once. That distinction changes what a team fixes: a stuck step versus a product nobody wants to keep using.

How are a cohort's retention curve and its churn rate related?

A cohort's retention curve and its churn rate describe the same users from opposite directions. The users a cohort loses between two periods are exactly the users counted as churned in that window, so a curve that drops from 48% to 40% between two periods is reporting the same group the churn rate counts as lost.