Behavioral Deviations in Delivery Delays: Physical and Communication States Modelling in Last-Mile Logistics

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Last-mile delivery delays are typically attributed to operational constraints, such as scheduling inefficiencies and capacity limitations. However, delivery delays can also arise from unexpected behavioral deviations. Even when deliveries are optimally planned in advance, behavioral disruptions may still occur, such as customers changing delivery addresses while couriers are already en route or receivers being absent upon arrival. Most existing studies incorporate such uncertainties into delivery optimization models to obtain robust operational solutions. Rather than treating human behavior as an exogenous disturbance to delivery optimization, this study conceptualizes behavioral deviations as endogenous components of delay resolution process. Resolving delays requires coordination among multiple human actors, including couriers, customer service operators, suppliers, and receivers. In this process, communication becomes critical for identifying the root cause of delays, attributing responsibility, and determining solutions. Delay outcomes such as whether deliveries can be received or rescheduled, depend on the activities of communication, decision-making, and physical actions. Specifically, information should be transmitted to the appropriate actor and decisions should be made as quickly as possible to guide the subsequent operational execution. As such, communication states, which have received limited attention in last-mile delivery research, are deeply embedded in delay resolution and interact dynamically with physical delivery states. While tracking physical delivery states has been recognized as effective for improving resource planning and reducing delay risks, jointly monitoring both physical and communication states is essential for understanding the underlying drivers of delivery delays. To this end, an information system is designed and implemented to capture both types of states. By examining these integrated states, behavioral factors can be distinguished from operational constraints in last-mile delivery delays. This study aims to optimize the delay resolution process to reduce the delivery delays and the time required to resolve the delays. Specifically, this study addresses three research questions: (1) Description: What casual pathways and behavioral patterns lead to delivery delays in last-mile logistics? (2) Prediction: How can multi-actor behavior be modeled in delay resolution process to predict delay duration and outcomes? (3) Prescription: How can information system support last-mile delivery by jointly tracking physical and communication states? To answer these questions, we first decompose archival delay-related text conversations into standardized physical and communication states using a large language model. Our dataset, collected from a third-party logistics company in China includes 2,833 time-stamped communication records. Based on these data, we develop a Markov-chain-based multi-agent process model to estimate the influence of physical and communication state transitions on the delay likelihood, duration, and outcomes. Building these analytical insights, we further design and implement an information system in the same logistics company. Leveraging the information system, the delay resolution process is tested in field.

Links and sources

Comments