Labelling Soft Churn Honestly

If your training labels only mark formal cancellations, the model will miss the accounts that hollow out months earlier. Soft churn is real attrition with delayed revenue impact.

Define soft churn with observable rules: no meaningful feature use for a fixed window, or a collapse below a cohort-specific activity floor, while the subscription remains active. Document exceptions for dormant-by-design seats.

Keep soft churn separate from hard cancellation in evaluation. A model that catches soft churn early may still miss abrupt exits driven by budget cuts or competitor switches.

Honest labelling is slower than dumping all inactive accounts into one bucket. It is also the difference between a score people trust and one they ignore.