<p>Last-mile routing is one of the biggest challenges supply chain managers face today. Despite significant advances in solving the routing problem and its many variants, when dealing with real-world instances of last-mile delivery route optimization, there is still a significant gap between the current body of knowledge in theoretical route planning and real-world route execution. The main reason is that the quality of a route is not defined solely by its duration or cost, but by a multitude of additional factors related to geography, infrastructure, and customers that are hard or impossible to address with classical optimization methods. This article presents the decision tree-based algorithm developed by Team MEGI for the Amazon Last-Mile Routing Research Challenge, co-hosted by Amazon and the MIT Center for Transportation and Logistics, which aimed to incorporate the tacit knowledge of drivers into routing algorithms. The resulting application was ranked 4th out of the 45 finalist teams. The key components of the algorithm are the decision tree and an innovative sequencing engine that takes advantage of the Amazon districting strategy. In addition, we present a second version of the algorithm that significantly improved the results of the version submitted to the competition.</p>