INFORMS Analytics Magazine — churn modelled as binary classification on RFM / engagement signals (logistic regression / decision trees / ensembles)

Claim. Customer churn is conventionally framed as a binary classification problem: at a given horizon, a customer either stays or leaves. The standard methods (logistic regression, decision trees, gradient-boosted ensembles) take RFM and engagement features as inputs and predict the probability of churn.

Quote.

"A customer either stays or leaves… This statistical method predicts the probability of a customer churning."

Source. INFORMS Analytics Magazine (pubsonline.informs.org, accessed 2026-06-21).

Confidence. Industry-consensus for the capability. The framing as binary classification is textbook; the methods are conventional.

Caveats. Capability is not outcome. The model has to perform above base rates, on enough data to clear noise, and the firm has to act on the at-risk flags. None of those is automatic — and as Express Analytics / INFORMS — "data quality and availability can fundamentally undermine a model's reliability" notes, data quality and availability "can fundamentally undermine a model's reliability."

Implication / use. Use to ground the retention-decision domain as a capability claim. Critically: do NOT pair with vendor-recycled magnitudes (5x-25x, 25-95pct) — see Quarantine: "5x-25x cheaper to retain," "5pct retention → 25-95pct profit," "80pct profits from 20pct customers," "AI churn → 20-30pct retention improvement" — vendor-recycled, untraced to primary.