Yunqiang Yin

Dujuan Wang

Huaxin Qiu

Sutong Wang

Qihang Xu

Yugang Yu

Joshua Ignatius

Problem definition: Promoting successful first-time delivery is the key to optimizing last-mile operational revenue, especially in the home appliance retail business for costly return shipping. This paper investigates how to identify and explain the potential delivery failure of each order at the last-mile hub under the promised delivery time, and accordingly regulate further distribution to realize a better outcome for its first-time delivery. Academic/practical relevance: Improving first-time attempts in last-mile deliveries is crucial to increasing customer satisfaction and retention as well as attracting new customers. To the best of our knowledge, this paper leads a data-driven research paradigm to predict the first-time delivery success of online orders. By targeting the orders which may likely fail to meet first-time deliveries, we can readjust order processing schedules at last-mile hubs to avoid delivery failures due to operational shortfalls. Methodology: We extend the multi-output interpretable classification model to forecast the delivery results by exploiting the complicated relationship between order status and relevant operational predictors. Considering the imbalanced data distribution in the success-failure prediction problem, we develop a two-stage cost-sensitive random forest chain (TCRFC) model to predict success rate and identify multiple outputs of potential delivery failures. This data-driven methodology reveals the importance of modeling the order flow process and integrating decision-making with predictive models in order delivery outcome forecasting. Results: After testing on the logistics operational-level dataset shared by RiRiShun Logistics (RRS), the proposed approach uncovers the key factors affecting the first-time delivery success rate, while demonstrating superior performance in predicting order delivery results. In addition, our work allows for better quality delivery decisions of order distribution processing based on real-time order prediction results. The simulation experiments estimate that the adjusted last-mile delivery plan can increase the success rate of the possible failed orders' first-time delivery by about 82.60% and save nearly $16,000 for return shipping compared to that of the original operations. Managerial implications: With effective predictions of order delivery outcomes at the last-mile hubs, logistics providers can make more targeted delivery commitments for the possible failed orders to increase successful first-time deliveries and promote customer loyalty. According to the identified type of potential failure, the operators may properly decide early or deferring delivery action of the corresponding order. And to further utilize resources for operational improvement to the best advantage, the operators can give priority to the orders which have key features with more significant impact on increasing the first-time delivery success rate.

Joshua Ignatius

Books

Your next favorite book is out there