01 / BUSINESS DECISION
Start with the decision.
Prioritize retention experiments using attrition risk, expected contribution and intervention cost, while treating treatment benefit as a separate causal question.
02 / PRACTICAL IMPLICATION
What the evidence can support.
A high-risk customer is not necessarily persuadable. The intended decision framework combines predictive risk with experimental evidence of incremental retention; it does not equate a churn score with treatment value.
03 / EVIDENCE
A defined question. Work still ahead.
Study design only. No churn model, lifetime-value estimate, evaluated treatment policy or repository has been published for this direction.
Planned evaluation
- Temporal holdouts with a fully observed outcome window, calibration and a simple recency/frequency baseline.
- Evaluate discounted contribution forecasts at stated horizons and sensitivity to censoring and discount rates.
- Use randomized retention assignment to evaluate incremental treatment value separately from predictive churn performance.
No evaluated result, client impact or live model is claimed for this planned study.
04 / DATA
Know where the evidence comes from.
A documented synthetic customer-event history is planned. No real customer records or outcome claims are included.
05 / METHODOLOGY
Match the method to the design.
Define contractual or noncontractual attrition first. Establish a recency/frequency baseline, then compare calibrated risk or survival models. Express lifetime value as discounted expected contribution over a finite, explicit horizon. Evaluate treatment effects only under an appropriate experimental design.
06 / LIMITATIONS
Conditions that matter.
- Prediction features precede the scoring date, and outcome windows are complete or censored appropriately.
- Contribution margin, retention definition and time horizon are declared before evaluation.
- Intervention assignment supports causal inference; predicted churn alone does not identify response.
Long-run value is sensitive to unobserved future behavior and business changes. No individualized intervention recommendation is supported at this planned stage.
07 / CODE & NEXT STEPS
The next useful piece of work.
Implementation has not begun. A repository and source download will be linked when they exist and contain reviewable work.