Cohort retention analysis: how to build the table and how to read it
Cohort analysis compares groups of customers acquired in the same period and shows whether newer cohorts stay longer or shorter. Average monthly retention hides these changes. The article describes data preparation, building the table, interpretation using a worked example, revenue retention and common mistakes.
Amounts and values in the examples are illustrative and show how the calculation works. In projects, we use the client's data.
Dominik Dębski · Lemur Media · · reading time: 4 min
Reviewed by Dominik Dębski, 1 November 2026
What is cohort analysis?
Cohort analysis is a behavioural analytics method that divides customers into groups sharing a characteristic from a given period and analyses each group separately. In retention analysis, a cohort usually consists of customers who started working with the company in the same month or quarter. Cohorts can also be defined by behaviour, e.g. customers who started using a specific feature after its launch. In "SaaS Metrics 2.0" (2013), David Skok describes cohorts calculated both by number of customers and by recurring revenue.
What data is needed?
All you need is a table with a customer ID, the start date of the relationship and information on activity or revenue in subsequent months. First, define what an active customer is, e.g. a paid subscription or at least one order in the month. The definition must be the same for all cohorts. Attributes for later segmentation are also useful: acquisition channel, segment, pricing plan, responsible sales rep.
| Customer ID | Start date | Month | Active | Revenue in the month |
|---|---|---|---|---|
| K-0001 | 15.01 | M0 | yes | PLN 2,000 |
| K-0001 | 15.01 | M1 | yes | PLN 2,000 |
| K-0001 | 15.01 | M2 | no | PLN 0 |
Each row represents one customer in one month since start. This layout lets you calculate both customer retention and revenue retention from the same table. The data comes from the billing system; the CRM is a weaker source if it does not record cancellations and plan changes.
How do you build the table?
The table is built in four steps, from assigning each customer to a cohort to dividing by the cohort size in the starting month.
- Each customer is assigned to a cohort based on the month or quarter in which the relationship started.
- For each cohort, count the active customers in each month since start (M0, M1, M2…).
- Divide the number of active customers by the cohort size in M0.
- Prepare a second table for revenue: divide the cohort's revenue in each subsequent month by its revenue in M0.
How do you read the table?
Read the table vertically: compare successive cohorts in the same month since start.
| Cohort | Customers M0 | M3 | M6 | M12 |
|---|---|---|---|---|
| January | 200 | 160 (80%) | 140 (70%) | 120 (60%) |
| April | 250 | 190 (76%) | 160 (64%) | no data |
| July | 300 | 210 (70%) | no data | no data |
The number of customers in each cohort is growing, but retention after three months falls from 80% to 70%. Combined M3 retention for the three cohorts is 74.7% ((160 + 190 + 210) ÷ (200 + 250 + 300) = 560 ÷ 750) and looks fine, even though the newest cohort is 10 percentage points weaker than the January one. Typical causes of this pattern are a change in acquisition channel, a change in customer profile or sales promises that do not match the product.
The second way to read the table is the retention curve of a single cohort, read horizontally. The January cohort loses 20 percentage points in the first three months, 10 points in the next three and 10 points in the following six. A flattening curve means that departures are concentrated at the start of the relationship, so efforts should focus on onboarding and activation. The cohort table also lets you check the effect of such efforts: David Skok describes a case in which first-month churn fell from 15% to 4%.
Customer retention or revenue retention?
Customer retention shows how many customers stay, while revenue retention shows how much money stays, taking expansions and contractions into account. Example: the January cohort starts with revenue of PLN 400,000. In M12, 120 of 200 customers remain (60%), but their revenue is PLN 360,000, or 90% of the starting value, because the remaining customers expanded their contracts. In B2B models with large differences in customer size, revenue retention matters more: losing many small customers may have less impact than one large contract shrinking.
What are the common mistakes and limitations?
- Changing the definition of an active customer mid-series without recalculating older cohorts.
- Drawing conclusions from small cohorts: with 30 customers, one customer equals 3.3 percentage points of retention.
- Treating missing data for young cohorts as zero, which drags averages down.
- Combining cohorts from different acquisition channels without separate tables.
- Ignoring seasonality when comparing cohorts from different months of the year.
- Calculating only customer retention in a base where a few customers account for most of the revenue.