Customer churn is the rate at which customers or subscribers discontinue a service during a specified period. In a subscription business, reducing churn requires more than predicting which users are at risk: it requires asking what intervention would change their outcome. section 1.0 section 8.3.1
Churn becomes a causal target when the workflow asks what produced it or what could prevent it:
- Causal discovery asks which observed factors contribute to subscribers churning. section 1.0
- Average treatment effect estimation measures how a changed recommendation algorithm affects churn relative to a baseline. section 1.0
- Conditional average treatment effect estimation examines whether losing popular content affects subscriber segments differently. section 1.0
- Probabilities of causation ask why a subscriber stayed or churned after an intervention, rather than merely assigning risk. section 8.3.1
Promotion and retention
Let X indicate whether a promotion was sent and Y indicate whether the subscriber stayed. For subscribers who received the promotion and stayed, the probability of necessity asks whether they would have churned without it. This distinguishes a promotion that was merely associated with retention from one that was necessary for it. section 8.3.1
