Analytics for Cross-Border e-Commerce: Inventory Risk Management of an Online Fashion Retailer

This study presents a data-driven analytics framework applied to a Chinese fashion retailer engaged in cross-border e-commerce. While the retailer successfully fulfills overseas orders through online platforms, it faces significant inventory management challenges in its overseas warehouses due to operational complexities such as extensive product assortments and high demand uncertainty. Traditional model-driven approaches often prove inadequate in addressing these dynamic challenges. Therefore, this research proposes a novel data-driven methodology for inventory management in overseas warehousing operations. We implement a two-stage predictive analytics approach. First, we classify all products into two categories: \(\mathcal A\) -items, which are profitable to store in overseas warehouses, and \(\mathcal B\) -items, which are not. Second, we predict demand for SKUs within each \(\mathcal A\) -item. Building on these predictions, we develop prescriptive models for optimizing inventory decisions. These include a deterministic model that treats predicted demand as certain, and a stochastic model that explicitly incorporates demand uncertainty through maximum entropy distributions. (1) Comparative analysis of machine learning techniques demonstrates that random forest outperforms other methods in both classification and demand prediction tasks. (2) The deterministic model can be efficiently solved as a linear program, while the stochastic model with maximum entropy distributions is solved using Karush-Kuhn-Tucker conditions. (3) Empirical application shows that predictive classification yields substantial cost reductions (up to 20% on average) by preventing \(\mathcal B\) -items from being shipped to overseas warehouses. Furthermore, the stochastic model provides near-optimal inventory solutions for \(\mathcal A\) -items, with performance losses as low as 0.00% compared to perfect foresight benchmarks.