Attributing Inventory Performance via Shapley-Based Counterfactual Decomposition

Inventory systems are typically evaluated using aggregate performance metrics such as out-of-stock and average inventory. In supply chain management, it is important to understand the underlying reasons for a period’s performance— specifically, how previous inventory management decisions, such as order placement, lead to the result and what their contributions are. Traditional methods are often restrictive and cannot be applied to broader cases. This paper proposes a Shapley-based decomposition framework that attributes the realized performance gap between an observed inventory policy and an optimized benchmark to individual order decisions. To numerically illustrate this method, we consider a finite-horizon periodic-review inventory system with fixed lead time and integer order quantities. Performance is evaluated using a composite metric that balances stockout frequency and normalized average inventory. Given an observed order vector and an optimized reference vector obtained via a genetic algorithm, we compute the average marginal contribution of each decision using a Monte Carlo Shapley estimator. The resulting decomposition allocates the total performance gap between the observed policy and optimal policy while accounting for nonlinear interactions and eliminating ordering bias. Compared to traditional methods, the proposed approach directly explains a realized benchmark-relative performance difference and is applicable to integer-constrained, non-differentiable, and simulation-based inventory systems. Managerially, it enables transparent inventory management performance evaluation and effective root-cause analysis.