← All analytics projects
01
Project 01 · Retention analytics

Customer Churn & Retention

Identify which customer profiles to prioritize and see how the score threshold changes campaign workload, precision, and recall.

ClassificationROC–AUC 0.886Thresholds
Customer records5,000
Observed churn42.4%
Holdout ROC–AUC0.886Logistic regression
Holdout sample1,000
Portfolio patterns

Where the synthetic model sees risk

Associations in generated data demonstrate analysis and delivery—not real customer causation.

Churn rate by contract

Month-to-month62.6%
One year30.5%
Two year10.6%

Churn rate by tenure band

1–12 months53.6%
13–24 months39.9%
25–48 months40.7%
49–72 months39.6%
Customer scenario

Score one account with the exported model

The browser applies the original scaler, one-hot coefficients, and logistic intercept. No Python service is called.

Account profile

Estimated churn score91.4%

High outreach priority. Score bands: High ≥65%, Medium ≥35%. This is a synthetic model score, not a calibrated real-world probability.

Decision guidance

A real retention program should validate the model on production data, estimate contact economics, and measure incremental retention with a randomized holdout.

Default reference profile reproduces the Python model at 91.4%.

Model diagnostics

Choose an operating threshold

A lower threshold captures more churners but creates more unnecessary contacts.

Flag customers at or above 0.50

Precision74.1%
Recall82.8%
F1 score0.782
Customers flagged474

Holdout confusion matrix

Actual / predicted
Stay
Churn
Actual stay
453
123
Actual churn
73
351

Top coefficient magnitudes

Contract_Two year1.890
Contract_Month-to-month1.657
MonthlyCharges0.960
InternetService_Fiber optic0.756
TechSupport_Yes0.602
PaymentMethod_Electronic check0.569