Approximate Dynamic Programming Captures Fleet Operations for Schneider National
Schneider National needed a model that would replicate the behavior of their team of dispatchers. We used the modeling and algorithmic framework of approximate dynamic programming to optimize the movements of 6,000 drivers, each described by a 15 dimensional attribute vector, over a month. The model closely replicates historical performance, and also produces accurate estimates of the marginal value of each of 500 types of drivers. Numerous projects have produced millions in savings.