Abstract
Digital marketing can support customer protection, yet crime-risk analytics may reproduce surveillance and exclusion. This chapter develops a Responsible AI Marketing Risk Observatory for aggregate zone-week decisions without individual profiling. A panel of 5,616 synthetic observations supports temporal model comparison, isotonic calibration, companion-model SHAP diagnostics, fairness auditing, and conformal prediction sets. Calibrated logistic regression achieved ROC AUC 0.912, Brier score 0.094, and ECE 0.028. The TPR disparity was 0.110; empirical coverage was 0.940, with 16.6% ambiguous sets requiring review. These results support methodological feasibility within the simulation, not field effectiveness or an unconditional coverage guarantee. The observatory links estimates to reversible safeguards and accountable human decisions. Customer trust and actual harm reduction remain unmeasured and require external validation.

This publication has 23 references indexed in Scilit: