The rapid expansion of the e-commerce sector has led to complex multi-warehouse networks, where optimizing the order allocation process is a critical operational challenge. While Machine Learning (ML) offers powerful predictive solutions, its widespread adoption in logistics is significantly hindered by the “black box” problem. The lack of model transparency erodes managerial trust, creating a crucial void in the literature for practical, trust-enhancing applications of Explainable Artificial Intelligence (XAI) in supply chain management. To address this gap, this paper proposes and validates a decision support system that integrates a high-performance XGBoost model with the SHAP (SHapley Additive exPlanations) interpretation method. The system was trained and evaluated on a real-world dataset spanning 24 months and over 50,000 orders. Our findings are twofold: first, the model achieved an accuracy of 72.9%, significantly outperforming a baseline heuristic (67.1%). Second, and more importantly, the SHAP analysis demystified the model’s decision-making logic, revealing that it prioritizes order complexity and financial value (features min_font_size, all_products_value) over mere geographical proximity. This work demonstrates that combining high predictive performance with transparency creates a trustworthy tool capable of overcoming key barriers to AI adoption in logistics.