Traditional ML for freight matching at Trucker Tools

Trucker Tools builds freight-management software for brokers and carriers — load tracking, load boards, carrier sourcing — over a network of more than 350,000 motor carriers. Consulting through Mission Cloud, I worked on the matching problem underneath all of it: given a load, which carriers should see it? And given a carrier, which loads are worth surfacing?

Most of the difficulty was in the data rather than the models. Accessing OLTP databases, transforming data into features ready for data analysis and science. Understanding the business domain and building robust feature engineering pipelines was therefore a good part of the project.

Load-carrier matching is a recommendation problem with a great deal of structure already in it. The clustering side grouped carriers by revealed behavior rather than declared preferences. The lanes actually run, the equipment actually used, the timing patterns actually kept. Behavioral segments gave the recommender something meaningful to generalize across, and made the cold-start case tractable: a carrier with little history can be matched through the cluster it most resembles. After that, traditional Bayesian statistics provide an interpretable and evidence-based starting point for recommendations.