Fleet Routing

Continental fleet routing rebuilt around a streaming optimizer. Real-time adaptation to weather, hours-of-service, and dock windows — the dispatcher's job, made calmer.

The state of play.

The dispatchers were running 2,840 trucks against a routing engine that recomputed twice a day. By the time a route reached the cab, weather had moved, a dock had slipped, and the driver was solving the puzzle by phone. The cost of the gap was measured in fuel, in driver hours, and in the dispatcher's afternoon.

What we built.

We rebuilt the routing core in Go as a streaming optimizer over Kafka, with CockroachDB for state and Mapbox for the geometry. Routes recompute on the order of seconds, with explicit constraints for hours-of-service, dock windows, and weather. The dispatcher's console was redesigned around exception handling — the calm cases become quiet, the loud cases become legible.

What shipped.

Eighteen percent reduction in fuel per route in the first full quarter. ROI hit at six months. The dispatcher floor is, by their own measurement, fifteen decibels quieter at 4 p.m.

We measured the wins in fuel. The drivers measured them in fewer phone calls. Both numbers went the right way.