Churn prediction
Estimating the probability that a specific customer will churn, using behavioral and commercial signals — distinct from a health score, which describes current condition rather than forecasting an outcome.
Churn prediction is the practice of estimating, per customer, the probability of cancellation within some horizon — usually the next renewal. Models range from weighted scorecards to machine-learned classifiers, but they consume the same raw material: usage trends, engagement decay, support friction, sentiment, and commercial context like renewal proximity and discount history. The output is a probability or a risk tier, not a certainty.
Churn prediction differs from health scoring in intent: a health score describes present condition, while a churn model forecasts a future outcome, and the two can disagree — an account can look healthy today while trending toward the exit. In practice the deciding factor is explainability. A bare probability gets ignored, because a CSM cannot argue with it or act on it; a prediction that shows its contributing factors tells the team both that an account is at risk and what the save play should address.