Data-driven perspective for improving first-time attempts in last-mile deliveries: evidence from a chinese home appliance retailer
研究提出一个两阶段成本敏感树集成链模型，利用中国物流运营商数据识别高风险失败订单并优化干预点，帮助提升首次配送成功率。
Our study addresses the critical challenge of optimising first-time delivery success in home appliance retail. Existing approaches typically consider last-mile delivery outcomes as a single-stage prediction task. However, limited attention has been given to understanding interdependencies within the multi-stage sequential process. Additionally, there is a lack of focus on implementing proactive dynamic interventions to address emerging issues that may affect delivery success rates. To fill this gap, we propose a data-driven framework featuring a novel two-stage cost-sensitive tree ensemble chain (TCTEC) model. This model effectively addresses the severe imbalance inherent in real-world delivery data and captures the intricate interdependencies between order processing stages and their associated success predictors. Using operational data from a major Chinese logistics provider, our model accurately identifies high-risk failures and highlights actionable intervention points, particularly through adjusting retention times at last-mile hubs. By prioritising orders based on failure types and critical operational features, our framework enables targeted resource allocation, thereby enhancing first-time delivery success rates. Our research provides logistics operators with practical tools to refine distribution strategies, which yields measurable improvements in last-mile efficiency and cost reduction. Consequently, our work advances both theoretical understanding and practical applications in delivery optimisation.