Learning from Route Plan Deviation in Last-Mile Delivery

Last-Mile Delivery

Last-mile delivery is a critical problem due to its substantial economic impact on operational cost, and deviations from the planned routes can potentially prevent companies from minimizing these costs.

We study whether delivery crews systematically, consistently and substantially deviate from the planned stop sequence of their routes.

Additionally, we analyze what drives these deviations and whether they add economic value or not.

With this objective, we perform regression analysis and build classification models, using one-year data across two countries, Mexico and the USA.

Our models predict whether a driver will deviate from the planned route, and the impact of the deviations on the route’s distance.

The findings show that the customers’ ZIP codes are highly useful to predict deviations.

Additionally, drivers are more likely to deviate and increase the route’s distance when more customers are visited, suggesting that this is where the company should focus its efforts.

As routes have a larger number of customers or visit a larger number of ZIP codes, drivers are more likely to make worse decisions: the proportion of deviated routes increase, as well as the proportional distance,  added to these routes.

A variable created to describe the location of all the customers visited by the route, Centroid ZIP, showed good predictable capabilities.

However, models that use this variable need training data of the particular regions where it will be used.

Download MIT's Learning from Route Plan Deviation in Last-Mile Delivery Now!