When Not to Build a Churn Model Yet

If you cannot name who acts when a risk score rises, a model only adds theatre. If your event stream still renames properties every release, scores will drift without anyone noticing. If monthly active users are tiny, statistical noise will drown signal.

Run a short readiness audit first: label clarity, feature coverage, leakage risk, and an owner map for interventions. Often the honest next step is instrumentation or a cohort review — not a modelling sprint.

Prediction earns its keep when scores change what someone does this week. Until that path exists, invest in the foundations.