Home / Wiki / C / Cohort Analysis

Cohort Analysis

Industry Strategy and Deal-MakingUpdated 16 Nov 2026

Cohort analysis is a method of studying customer behaviour by grouping customers who share a starting point, such as the month they signed up, made a first purchase or took out a loan, and then tracking each group separately over time. Each group is called a cohort. Comparing cohorts at the same age shows whether retention, spending or credit losses are improving or getting worse, which a single overall figure cannot show.

The method is widely used by subscription and software companies, online retailers, consumer apps and lenders. It is closely linked to the churn rate and customer retention rate, which it breaks down by age, and to customer lifetime value (CLV), which it helps estimate. In lending the same approach is usually called vintage analysis. Cohort tables are a standard feature of business intelligence (BI) tools and dashboards.

01

How a cohort table works

A cohort table puts each cohort in a row and each period since the start in a column. The value in a cell is usually the retention rate:

Rc,t=Ac,tNc×100R_{c,t} = \frac{A_{c,t}}{N_c} \times 100
  • Rc,tR_{c,t} is the retention rate of cohort cc in period tt, in per cent.
  • Ac,tA_{c,t} is the number of customers from that cohort still active in period tt.
  • NcN_c is the number of customers in the cohort at the start.

In the example below, 1,000 customers signed up in January, 1,250 in February and 1,500 in March. Of the February cohort, 1,000 were still active one month later, so its month-1 retention is 1000÷1250×100=80%1000 \div 1250 \times 100 = 80\%; 875 remained after two months, giving 70%.

CohortCustomers at startMonth 1Month 2Month 3
January1,00082%74%70%
February1,25080%70%
March1,50078%

Reading down a column compares cohorts at the same age. Here month-1 retention has slipped from 82% to 78%, a warning sign that a blended figure might hide if the business was growing fast. Reading along a row shows how one cohort decays. The table forms a triangle because younger cohorts have fewer periods of history.

02

Retention curves

Plotting each row as a line gives a retention curve. Most curves fall steeply in the first periods, as customers who were a poor fit leave, and then flatten. Where the curve levels off indicates the share of customers who become long-term users. A curve that keeps falling towards zero suggests weak product fit, while later cohorts whose curves sit above earlier ones point to improvements in the product, pricing or onboarding.

03

Revenue cohorts

Cohorts can also track revenue rather than headcount, by dividing each cohort's monthly recurring revenue (MRR) in each period by its starting MRR. Because customers who stay often upgrade, a revenue cohort can rise above 100% even while some customers leave. This is the pattern measured by net revenue retention (NRR) and described as negative churn. Comparing a cohort's cumulative gross profit with its customer acquisition cost (CAC) shows the CAC payback period directly from data rather than from a formula.

04

Uses in lending and e-commerce

Lenders group loans by the month or quarter they were made and track cumulative delinquencies, defaults and charge-offs by months on book. Comparing vintages at the same age shows whether recent underwriting is riskier than before, often well before losses show up in the overall non-performing loan (NPL) ratio, and supports the calibration of a credit scoring model.

Online retailers group customers by the month of their first order and track repeat-purchase rates and cumulative spending. This shows whether customers acquired through a promotion, a marketplace or a new marketing channel come back as often as others, and how long it takes for them to become profitable.

05

Limitations

Small cohorts produce noisy percentages, so a single bad month can look like a trend. Cohorts acquired through different channels, prices or countries may behave differently, and a change in that mix can be mistaken for a change in the product. Cohort tables show association rather than cause: to learn whether a change actually improved retention, businesses use controlled experiments such as A/B tests, the commercial counterpart of the randomized controlled trial (RCT). Definitions also matter, since an "active" customer may mean a paying subscriber, a user who logged in or a buyer in the period, and results are only comparable when the definition is fixed.

Last updated 16 Nov 2026