An anticipatory shipping system for online retailers via mining customer behavior in large e-commerce promotion

Abstract

The anticipatory shipping practiced by online retailers plays an important role in improving customer satisfaction. However, online retailers face a new challenge in anticipatory shipping: they are required to ship a significant amount of products due to a surge of demand during the large e-commerce promotion, which dramatically aggravates the pressure on logistics distribution and reduces logistics efficiency. Therefore, making anticipatory shipping decisions to meet the suddenly increased demand has become an urgent problem for online retailers. Our research addresses this challenge by establishing a new anticipatory shipping system. We propose three cost-sensitive anticipatory shipping models, including cost-sensitive logistic regression (CSLR), cost-sensitive LightGBM (CS-LightGBM), and cost-sensitive CatBoost (CS-CatBoost). Their loss functions are constructed according to the cost of the anticipatory shipping system. Furthermore, we propose two new evaluation criteria to assess the effectiveness of the anticipatory shipping system. It intuitively demonstrates the cost differences after adopting the anticipatory shipping system. Moreover, we explore the real large promotion customer behavior data containing nearly three million samples. Our results find that the proposed cost-sensitive based forecasting models significantly outperform reference forecasting models. Our experimental evaluation concludes that forecasting AUC is more instructive to operational strategy than accuracy. Additionally, our empirical findings suggest that the anticipatory shipping system should be preferentially applied to high-value products. Conversely, low-value products should not choose anticipatory shipping to control logistics costs during surges.