← Commercial research

CUSTOMERS · RETENTION & VALUE

Customer retention and lifetime value

Which customers may leave—and which interventions are worth testing?

Planned research

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.

Planned research

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.

Next step

Specify a customer lifecycle and attrition definition, then create a data generator that distinguishes baseline risk from treatment responsiveness.

← All commercial research