Fictional example: growth and churn together
A period starts with 100 customers. Of these, 10 leave and 90 remain. Twenty new customers join and are still present at the end, giving a final total of 110. Initial-cohort churn is 10 / 100 = 10%, despite portfolio growth.
| Fictional case measure | Value | Rule |
|---|---|---|
| Original customers | 100 | Population followed at the start. |
| Original customers lost | 10 | Leaving under the chosen definition. |
| Original customers remaining | 90 | 100 − 10. |
| New customers remaining | 20 | Counted separately. |
| Final total | 110 | 90 + 20. |
| Initial-cohort churn | 10 % | 10 / 100 × 100. |
Define loss precisely
- Choose the unit: person, account or subscription. Specify cancellation request, effective ending, payment failure and reactivation; these events differ.
- In this cohort calculation, new customers do not alter the starting denominator. Document any different convention used by a tool or report.
- Separate customer churn from revenue churn. Net revenue churn can include expansion among existing customers and become negative; it is not a customer-loss rate.
Test retention actions without guaranteeing results
Distinguish voluntary departure, payment failure and changed needs. A predictive score does not establish that an offer will prevent churn. Evaluate actions with a suitable comparison group, costs, adverse effects and the actual outcome.
Primary references
Churn needs an explicit exit event
Cancellation, access termination, inactivity and payment failure can represent different phenomena. Define the chosen churn measure, starting population and period. Handle reactivation and plan changes separately; otherwise internal transfers may count as both losses and new acquisition.
- Define the exact exit event and date.
- Separate customer churn from revenue churn.
- Check pauses, reactivation and plan migration.
Is an inactive user necessarily lost?
It depends on normal usage frequency. Define a service-appropriate threshold and test observed returns; absent events may also reflect interrupted collection.
Primary reference: Google SRE — Data processing pipelines. Related method: read the practical guide.