AI-Driven Fleet Route & Fuel Optimization

The Challenge

A regional logistics company operating 1,200+ delivery vehicles across 6 states was spending $18M annually on fuel. Routes were planned manually by dispatchers using static zones, ignoring real-time traffic, vehicle load factors, and driver hours-of-service (HOS) constraints. Drivers frequently exceeded HOS limits, creating compliance risk and forcing unplanned overnight stops.

Fuel costs had increased 31% year-over-year, and the company had no visibility into which routes, vehicles, or driving behaviours contributed most to consumption.

Our Approach

We built a route optimisation engine deployed as a gRPC service on Google Kubernetes Engine. The engine solves a constrained vehicle routing problem (CVRP) that accounts for real-time traffic (Google Maps Platform), vehicle-specific fuel consumption curves, payload weight, driver HOS remaining, and customer delivery windows.

Fuel Modelling: Each vehicle type has a calibrated fuel consumption model that accounts for speed, gradient, payload weight, and ambient temperature. These models were trained on 18 months of telematics data from the fleet's existing GPS trackers and fuel card transactions.

Dynamic Re-routing: Routes are not just planned at the start of the day. The engine continuously monitors traffic conditions and re-optimises remaining stops every 15 minutes. Drivers receive updated turn-by-turn directions via a custom mobile app. Re-routing decisions are constrained to avoid exceeding HOS limits or missing delivery windows.

Dispatcher Dashboard: A real-time dashboard shows fleet position, route progress, predicted arrival times, and fuel consumption against the planned baseline. Dispatchers can manually override routes when needed, and the system logs the cost delta of each override for performance review.

Results