This document discusses how to model customer churn through machine learning. It defines churn as customers leaving or stopping usage. There are two types of churn - for subscription models where leaving can be clearly defined, and non-subscription models where leaving must be approximated. The document recommends predicting churn through classification models to identify potential churners, using customer behavioral and profile features over time. It also discusses evaluating models on validation data and using models to predict future churn and inform retention offers.