The Challange
Northbay’s dispatch team found out about late deliveries when customers called to complain. They had years of shipment data sitting in spreadsheets and a TMS, but nobody had time to look for patterns, and service credits were eating into margins every quarter.
The Solution
We combined TMS history, weather feeds and carrier performance data into one training set, then built a gradient-boosted model that scores every active shipment for delay risk each hour. High-risk loads appear on a simple dispatcher board with the likely cause, so the team can reroute or warn the customer early.
Result
82% accuracy in flagging late shipments 24+ hours ahead 31% drop in service credits within two quarters Dispatcher time on status calls cut by roughly half
Technology Stack
Python, XGBoost, AWS SageMaker, PostgreSQL, React
