Do it right the first time: Vehicle routing with home delivery attempt predictors. Issue 4 (4th January 2023)

• Record Type: Journal Article
• Authors: Lim, Stanley Frederick W. T.; Wang, Qingchen; Webster, Scott
• Is Part Of: Production and operations management. Volume 32:Issue 4(2023)
• Page Start: 1262
• Page End: 1284
• Publication Date: 2023-01-04
• Subjects: discretionary work effort -- field experiment -- goal‐gradient hypothesis -- home delivery attempt -- online retailing -- vehicle routing; Production management -- Periodicals; 658.505
• DOI: 10.1111/poms.13926
• Languages: English
• ISSNs: 1059-1478

Abstract

Abstract: Up to 20% of all business‐to‐consumer deliveries fail on the first attempt. Failed deliveries not only carry cost implications but also incur damage to retailers' brand reputation. Despite its economic significance, research has paid little attention to delivery attempt as an operational outcome or seldom accounted for its effects in routing models. This is partly due to the many factors that can influence delivery outcomes. We address this knowledge gap by first demonstrating that failed delivery attempts can be reasonably predicted using common routing, demand, environmental, and market attributes at both the individual package and route levels. For model‐building, we use transaction data from an e‐commerce retailer in South America. We then explore the value of accounting for failed delivery attempts in routing models. We propose a two‐stage greedy algorithm for solving large problem instances. Our analysis indicates that not accounting for the probability of failed attempts in routing models may create a significant downward bias in the total cost of delivery. The analysis also suggests that manipulating the sequence in which packages in a route are delivered can be a cost‐efficient lever that firms can employ at almost zero cost to profoundly affect delivery outcomes. We replicate the prediction model to a new sample from a delivery company in Singapore and calibrate it for a randomized field experiment to validate our algorithm's performance. Packages and drivers are randomly assigned to either our algorithm or the focal company's existing algorithm. Results suggest that our algorithm, on average, reduces the share of failed delivery attempts by 10% and the total cost of delivery by $13 per route. We further propose drivers' discretionary work effort and the goal‐gradient hypothesis as a mechanism for the efficacy of our algorithm. Controlling for time of day and other fixed effects, we empirically find that packages assigned to slots later in the route tend to have a lower failure rate because drivers display a higher degree of discretionary work effort toward the end of a route. Our approach can be applied to other firms' last‐mile delivery operations to improve their delivery execution.