This study examines privacy-aware and interpretable customer response forecasting for cross-channel digital marketing decision support. It proposes a gradient-boosted decision-tree framework that combines consent filtering, k-anonymity, differential privacy, temporal feature engineering, probability calibration, and SHAP-based explanations. The model uses a 12-month event window containing 142,380 customer campaign observations from email, paid search, display, social media, and mobile push channels. Performance is assessed against logistic regression, random forest, and shallow neural-network baselines using AUROC, AUPRC, and Brier score. The proposed model records an AUROC of 0.834, an AUPRC of 0.427, and a Brier score of 0.152, compared with 0.801, 0.389, and 0.165 for the strongest baseline, respectively. Differential privacy reduces AUROC from 0.847 without privacy noise to 0.834 at ε = 1.0, representing a 1.5% relative reduction. Calibration lowers the Brier score from 0.168 to 0.152. Engagement frequency, interaction recency, and channel diversity provide the largest explanatory contributions. The findings indicate that privacy controls and interpretable forecasting can support practical campaign prioritization while retaining competitive predictive performance.