Churn Early Warning System

Risk scoring + weekly save list to make retention proactive.

At a glance

Primary KPI + guardrails, documented assumptions, and a weekly operating rhythm.

Primary: retention lift vs holdoutSave rateCost per saveIncremental profitGuardrails: unsubscribes, margin %

Tools

SQLPythonPower BI

Artifacts

DashboardNotebookSQL snippetsExec summary (PDF)

Links

Dashboard: (add link) Repo: (add link) Exec summary PDF: (add link)

Overview

I defined churn, engineered customer features, built an explainable risk score, and operationalized it as a weekly action queue. The output is a dashboard and playbook that prioritizes outreach, caps discount spend, and measures incremental retention lift with holdouts.

Bottom line

If we do nothing, the business keeps arguing about the numbers. If we do this, the decision becomes measurable and repeatable.

The problem

Retention was reactive: churn showed up after customers disappeared, and incentives were spent without clear prioritization or measurement.

What I built

A feature table (recency, frequency, monetary, discount behavior), a churn label (e.g., no purchase in 90 days), an explainable model score, and a weekly top‑risk list with tiered actions.

Key insights

Risk is concentrated. Targeting the top risk decile is materially more efficient than broad discounting—especially when actions are tiered and capped.

Recommendation

Run a 4‑week pilot on the top 10% risk segment with 10% holdouts in each tier. Track saves, cost per save, and incremental profit weekly.

Next iteration

Add richer signals (web/app events), and test uplift modeling once measurement infrastructure is established.