Churn is the “death” event of customer analytics: the moment a subscription is cancelled, a contract lapses, or activity stops long enough to count as gone. Contractual businesses observe it as a dated event, which makes lifetimes directly analyzable with survival tools; non-contractual businesses have to infer it from silence, which calls for latent-lifetime models instead.
Most retention work is built on top of a churn definition, so the definition itself deserves care: reversals, home movers, and system artifacts can all masquerade as churn, and a noisy target quietly degrades every model trained on it.
