Predictive: churn, CLV & propensity

GET /v1/ml/summary and POST /v1/ml/score expose churn, customer-lifetime-value (CLV) and purchase-propensity scoring.

Two ways it runs

• With the ML sidecar (a separate Python service, configured via ML_SIDECAR_URL): scores come from trained models on your actual data.
• Without it: the platform falls back to in-stack heuristics automatically — predictive segmentation still works, just with a simpler model, no separate service to run.

Using the scores

Scores are written back onto the contact as ml_churn_score, ml_clv and ml_propensity — from there they're just traits, segmentable exactly like any computed trait (trait:ml_churn_score), so a high-churn-risk segment can feed straight into a win-back journey without any custom integration code.