The Value of Last-mile Delivery in Online Retail
Preprint
- 1 January 2023
- preprint
- Published by Elsevier in SSRN Electronic Journal
Abstract
Last-mile delivery refers to the final leg of shipment in which packages are moved from a local transportation hub to end-customers. It is known to be the most expensive part of delivery, due to operations complexity and lack of economy of scale. Given the growing importance of last-mile delivery in online retail, companies need to decide whether to outsource this service to customers through pickup stations where they can collect packages or to offer home delivery. In this paper, we study the economic value of last-mile delivery. We conducted a quasi-experiment in collaboration with Cainiao, Alibaba's logistics subsidiary, in which home delivery service was rolled out to some pickup stations sequentially in 2021. This allowed us to comprehensively evaluate the causal impact of last-mile delivery. Using a staggered difference-in-differences identification, we found that last-mile delivery significantly increases sales and customer spending on Alibaba's retail platform. Last-mile home delivery relies heavily on physical labor and thus has capacity constraints. To optimally prioritize that capacity, we used an uplift model with causal machine learning to target the most responsive customers. Specifically, we proposed a novel capacity- and fairness-aware uplift model that builds in capacity and fairness constraints to optimize the targeting policy, thereby maximizing profits without compromising social equity. Our findings suggest that online retailers should carefully weigh the costs and benefits of last-mile delivery and tailor their logistic strategies.Keywords
This publication has 49 references indexed in Scilit:
- The Operational Value of Social Media InformationProduction and Operations Management, 2018
- Coordinated Logistics with a Truck and a DroneManagement Science, 2018
- Estimation and Inference of Heterogeneous Treatment Effects using Random ForestsJournal of the American Statistical Association, 2018
- Double/debiased machine learning for treatment and structural parametersThe Econometrics Journal, 2018
- Double/Debiased/Neyman Machine Learning of Treatment EffectsAmerican Economic Review, 2017
- XGBoost2016
- Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal InferencePolitical Analysis, 2007
- The use of the area under the ROC curve in the evaluation of machine learning algorithmsPattern Recognition, 1997
- Statistics and Causal InferenceJournal of the American Statistical Association, 1986
- The Truck Dispatching ProblemManagement Science, 1959