Background: AI-driven demand forecasting expands the signal space for inventory and pricing decisions, enabling faster reactions to market changes. However, forecast error, non-stationarity, and distribution shifts raise a governance question: should decisions be designed to be robust to uncertainty, or adaptive to feedback?Methods: This review integrates inventory control, probabilistic demand forecasting, dynamic pricing, and robust optimization into a unified decision architecture. We organize prior findings around a closed-loop cycle: data ingestion, forecasting (point and distribution), policy selection (robust/adaptive/hybrid), execution, monitoring, and recalibration.Results: Robust policies protect against tail risk by optimizing over uncertainty sets and worst-case scenarios, but may be conservative and costly in stable environments. Adaptive policies leverage frequent feedback to improve average performance, yet can become unstable under regime changes, delayed signals, or strategic customer responses. The synthesis supports a hybrid design: adaptive learning within robust guardrails (service constraints, pricing move limits, and inventory safety floors).Conclusions: The practical frontier is not “robust versus adaptive” as a binary choice. Best practice is layered: robust feasibility and risk limits at the outer layer, with adaptive learning tuned inside auditable constraints. Future research should prioritize regime-switching demand, decision-focused learning, and explainable pricing and replenishment rules.
Mehmet A. Begen (Fri,) studied this question.