Effective inventory management requires deciding when to order, how much to order, and how much protection to keep in the form of safety stock and safety lead time. Many recent data-driven approaches still treat lead times, purchasing costs, and defect rates as fixed inputs, and usually optimise replenishment separately from safety buffers. This study proposes a data-driven framework that jointly determines replenishment decisions, safety stock, and safety lead time while accounting for time-varying lead times, purchasing costs, and defect quantities. The approach also uses a custom loss function to reflect the asymmetric consequences of forecast errors in the inventory setting. The method is tested on a real automotive dataset. Results show that it produces inventory decisions that are close to those of an ideal decision-maker and improves cost performance compared with the benchmark approaches considered. Overall, the study shows that integrating forecasting, safety-parameter tuning, and replenishment optimisation can support more effective inventory decisions in dynamic operating environments.
Gabellini et al. (Fri,) studied this question.