Request Account
gender Female SeniorCitizen 0 Partner No Dependents No tenure 52 Contract One year PaperlessBilling Yes PaymentMethod Mailed check
Services
PhoneService Yes MultipleLines Yes InternetService Fiber optic OnlineSecurity No OnlineBackup Yes DeviceProtection No TechSupport No StreamingTV No StreamingMovies No
Charges
MonthlyCharges 80.200000 TotalCharges 4297.600000
Committed request JSON {
"gender": "Female",
"SeniorCitizen": 0,
"Partner": "No",
"Dependents": "No",
"tenure": 52,
"PhoneService": "Yes",
"MultipleLines": "Yes",
"InternetService": "Fiber optic",
"OnlineSecurity": "No",
"OnlineBackup": "Yes",
"DeviceProtection": "No",
"TechSupport": "No",
"StreamingTV": "No",
"StreamingMovies": "No",
"Contract": "One year",
"PaperlessBilling": "Yes",
"PaymentMethod": "Mailed check",
"MonthlyCharges": 80.200000,
"TotalCharges": 4297.600000
}
examples/score_request.json
Committed response No action
Churn probability
0.122990
Scoring model gradient_boosting . Rule no_action . Offer null .
Churn risk, CLV, and offer eligibility do not match a save call or an offer.
README, on this response: churn probability 0.122990 is below churn_medium 0.165703, so the rule is no_action, even though the OnlineSecurity lookalike probability is 0.618604, above the offer cutoff 0.400000.
Top reasons Path contributions along the gradient-boosting trees. They sum with the model intercept to the churn log-odds for this customer. They are not SHAP values.
Contract
-0.357808 Decreases churn risk StreamingMovies
-0.235742 Decreases churn risk MonthlyCharges
-0.169382 Decreases churn risk Add-on propensities Lookalike probability of current holding. Eligible means the customer has internet service and does not already hold that add-on.
OnlineSecurity 0.618604 · eligible TechSupport 0.478232 · eligible
CLV proxy Value 1511.445671
Monthly charges 80.200000 × expected remaining months 18.845956. Horizon 72.000000 months. The README states that the value is the six-decimal product of those two displayed figures.
Frozen cutoffs From reports/metrics.json. Cutoffs are not chosen on the test set. churn_high 0.442962 churn_medium 0.165703 clv_high 3311.917019 min_offer_propensity 0.400000
Committed chart for customer 5343-SGUBI. Contract, StreamingMovies, and MonthlyCharges lower this customer's churn log-odds. The contributions are not SHAP values. Committed response JSON {
"churn_probability": 0.122990,
"scoring_model": "gradient_boosting",
"top_reasons": [
{
"feature": "Contract",
"contribution": -0.357808,
"direction": "decreases_churn_risk"
},
{
"feature": "StreamingMovies",
"contribution": -0.235742,
"direction": "decreases_churn_risk"
},
{
"feature": "MonthlyCharges",
"contribution": -0.169382,
"direction": "decreases_churn_risk"
}
],
"next_best_action": {
"action": "no_action",
"offer": null,
"rule_id": "no_action",
"rationale": "Churn risk, CLV, and offer eligibility do not match a save call or an offer."
},
"addon_propensities": {
"OnlineSecurity": {
"eligible": true,
"probability": 0.618604,
"reason": null
},
"TechSupport": {
"eligible": true,
"probability": 0.478232,
"reason": null
}
},
"clv_proxy": {
"monthly_charges": 80.200000,
"expected_remaining_months": 18.845956,
"value": 1511.445671,
"horizon_months": 72.000000
},
"explanation_method": "Path contributions along the gradient-boosting trees. They sum with the model intercept to the churn log-odds for this customer. They are not SHAP values."
}
examples/score_response.json