Unpacking churn with survival models

Survival modeling is a statistical technique used to analyze and model time-to-event data, such as the time until death, failure of a mechanical device, or occurrence of a disease.

The article discusses using survival modeling for predicting customer churn and understanding causal mechanisms. The proportional hazards assumption, which requires the effect of covariates to remain stable over time, is often violated in real-life implementations.

The author proposes a solution by introducing an interaction term between time and the problematic variable, which allows for consistent coefficients and more accurate interpretation of the effects of variables over time.

This method is demonstrated using a churn example and the Lifelines package in Python. Overall, survival models are powerful tools for unpacking churn, but the issue of time-dependent covariates is common and can be addressed using established solutions.