An Empirical Study of Time Allotment and Delays in E-Commerce Delivery

Problem definition: We study how having more time allotted to deliver an order affects the speed of the delivery process. Furthermore, we seek to predict orders that are likely to be delayed early in the delivery process so that actions can be taken to avoid delays. Methodology/results: We use the JD.com transaction dataset provided by Shen et al. (2020). We first employ a Regression Discontinuity design to examine the effect of exogenous variations in time allotment between same-day and next-day orders on delivery duration. Subsequently, we fit random forest classification models to predict delays and identify the key predictor variables. We draw methods from causal inference and machine learning to help identify early on orders that will likely be delayed, in order to increase the likelihood of preventing at least part of the potential delays during the delivery process. We see that when there is more allotted time to deliver an order, workers spend disproportionately longer in earlier stages of the delivery. Such behavior causes workers in the later stages to "speed up", as spending more time in earlier stages leaves less time for later stages of delivery. Based on the feature importance analysis, we find that such speedup effect is mainly driven by orders that are at risk of being delayed due to a prolonged first leg, whereas other factors (such as product and demographic characteristics) lend relatively little support in predicting delays. Managerial Implications: Our delay prediction model can use information about earlier legs for early detection of potential delays. As the speed and duration of each leg varies with the allotted time, managers should carefully evaluate how much time and resources they allot for each stage of the delivery and try to identify early any orders at risk of being delayed