validationvv1.4SR 11-7 · Independent Validation Pass
Cardholder Attrition Propensity
90-day voluntary attrition risk
owner: Customer Data Science · last validated May 2, 2026
Outcome it's accountable for
$41M retained annual revenue+19% vs untargeted retention
metric: retained revenue vs no-model
0.831
AUC
0.522
KS
0.689
Precision
0.701
Recall
Top feature drivers (SHAP)
spend_trend_90d0.31
rewards_redemption_rate0.22
tenure_months0.17
dispute_count0.14
support_contacts0.16
Disaggregated performance
| segment | AUC | approval |
|---|---|---|
| Mass | 0.828 | - |
| Affluent | 0.836 | - |
| Private Client | 0.841 | - |
| Small Business | 0.819 | - |
Tested for disparate impact across segments - no material divergence.
Intended use
Rank cardholders by 90-day voluntary-attrition risk to target retention offers. Marketing decision-support only.
Training data
24 months of customer_360 snapshots + attrition labels. Time-based validation split; calibrated probabilities.
Limitations
Cold-start weakness for < 6-month tenure; offer-fatigue not modeled. Recalibrated quarterly.
Fair lending & adverse action
Retention targeting reviewed for fair-treatment; no pricing or credit impact. Offers audited for consistency across segments.