How to Read Cohort Analysis and Retention

Premise — What cohort analysis can and can't tell you

What cohort analysis can answer is the question, "How long did a group of users who signed up in a particular period keep using the product afterward?" If you look only at total membership or monthly active users (MAU), growth in new acquisition can mask the churn of existing users, so you fail to notice that the foundation is actually wearing thin. Only by dividing into cohorts can you isolate the reality of "retention" independently of the momentum of new inflow.

Conversely, cohort analysis alone cannot explain the reason "why they quit." It is a method for detecting changes in the numbers; identifying causes requires combining it with other means such as user interviews. Here we take weekly cohorts by sign-up month as an example and cover the procedure for reading them.

Procedure — How to build cohort tables and retention curves

1. Decide the cohort's reference date. The basic approach is to align it with the timing when the product's value delivery begins—sign-up date, first-payment date, first-activation date, and so on.

2. Decide the observation interval. Weekly is natural for a product used frequently; monthly is natural for a B2B business tool premised on monthly use. If the interval is too short, the noise from a small denominator stands out.

3. For each cohort, treating the reference date as 100%, calculate the share of users who performed the "reference behavior" at each elapsed period. The closer the "reference behavior" is to a value action specific to that product (e.g., saving an article, generating a report) rather than logging in, the more precise it becomes.

4. Build a table that stacks cohorts vertically, and compare the columns for the same elapsed period horizontally. If newer cohorts show improving retention, product improvements may be taking effect.

Worked example — A cohort table for a hypothetical SaaS product

The following are hypothetical figures for illustration. Please view them not as real benchmarks but as an example for demonstrating how to read a table.

Cohort (sign-up month)  week0  week1  week2  week4  week8
2026-01                 100%   62%    51%    44%    41%
2026-02                 100%   58%    49%    43%    42%
2026-03                 100%   60%    52%    45%    44%

In this example, retention rises slightly from week4 to week8. A shape that levels off and then reverses upward is the shape growth practitioners call the "smile curve," an idea repeatedly taken up in the body of essays by Reforge founder Brian Balfour. Comparing the three cohorts, newer cohorts have slightly higher retention at week1, suggesting that the effect of onboarding improvements may be starting to show.

Where people stumble

Treating retention and churn as a simple complement (retention + churn = 100%) holds in a business where subscription cancellation is clearly defined, but it does not hold in a product where a "stopped using but didn't cancel" state exists, as in freemium. Unless you also account for "reactivation"—returning from dormancy—you will compute a churn rate more pessimistic than reality.

Another pitfall is over-interpreting cohorts with small denominators. In a product where only a few dozen people sign up per week, a behavioral difference among just a few people can move retention by more than 10 points, and a curve that looks like it's smiling may be nothing more than coincidental noise. That said, a small denominator is no reason to give up on cohort analysis itself; you can cope with adjustments such as widening the observation interval to monthly or treating several months together as one cohort. On the other hand, with a product like a B2B product whose usage frequency is closer to quarterly than weekly, mechanically applying a weekly cohort table risks misreading a normal usage cycle as churn—so caution is needed there too.

Read next

Retention captured through cohorts becomes easier to operate once you distill it one level down into a metric you track day to day. The thinking behind metric design is covered in "Designing and Operating a North Star Metric." Also, since the shape of a retention curve is one of the signals of PMF, reading it together with "How to Measure Product-Market Fit" reveals the connections between the metrics.