Customer Retention Analytics

Churn risk, scored live in your browser

This runs the actual trained XGBoost model client-side — no server, just the same 400 boosted trees the Python pipeline produced, walked in JavaScript. Change the inputs below and the prediction updates instantly.

View the full project on GitHub Verifying model…
0.849ROC-AUC (XGBoost)
7,043customers scored
27engineered features
$5.47Mannual revenue at risk identified

Try it — describe a customer

This assumes a flat billing history since signup (no simulated price drift), and uses dataset-wide averages for the price-comparison features. See the full feature list for exactly how each one is engineered.

Account
Services
Billing
— —

Key factors

    What the full analysis found

    Generated by the pipeline in this repo, from the same model driving the predictor above.

    Churn rate and revenue lost by contract type
    Month-to-month customers churn 15x more than two-year customers, and carry most of the revenue lost.
    Heatmap of churn rate by tenure and bill tier
    A high bill on a new account is the single worst combination — 72% churn.
    Top churn drivers by gain, XGBoost
    Contract status and billing friction dominate the model's feature importances.
    Cumulative gains chart
    Contacting just the top 10% by risk score catches a third of everyone who'll actually leave.