Demand Prediction, Predictive Shipping, and Product Allocation for Large-Scale E-Commerce

In this paper, we study the problems of multiple-product demand prediction, predictive shipping mechanisms, and products allocation across multiple warehouses for large-scale e-commerce. These research problems are triggered by an exploratory data analysis on the transactional level data from Alibaba and its logistics arm Cainiao, including detailed transaction orders, inventory and logistics for 7,013 different products and 130 warehouses. First, we develop a multiple-product demand prediction system and identify unique features presented in the data, which improves prediction accuracy significantly compared to standard machine learning models. A new clustering-based regularization method is developed with the aid of representation learning to capture the product interactions, augment data and prevent over-fitting. Second, we propose and analyze in theory a novel shipping mechanism - Predictive Shipping, which utilizes demand prediction to arrange shipping before orders are placed. We provide analytical bounds on the performance improvement via this new mechanism, which originates from the improvement on demand prediction. Third, we formulate and solve a large-scale products allocation problem across warehouses and show that a change in the current products allocation distribution of Cainiao can potentially be beneficial. Numerical experiments with both real data and synthetic data are conducted to demonstrate our findings.