Gulf Freight ran a 240-vehicle fleet across the GCC with routing done in spreadsheets by a night shift of nine dispatchers. Every deviation - traffic, a refused pickup, a vehicle down - had to be re-planned manually, and the plan was stale within an hour of leaving the depot.

The commercial pressure was direct: late deliveries triggered penalties on their three largest contracts, and fuel was the single biggest line in cost after drivers.

We built a routing engine that treats every shipment as a live constraint problem rather than a fixed manifest. It re-solves continuously as traffic, vehicle capacity and pickup windows move, and it explains each change so dispatchers stay in the loop instead of being replaced by it.

Everything runs on AWS with a Next.js control room for the operations team and a lightweight driver app for the last mile.

The rollout went live across four depots without pausing operations. On-time delivery moved from 61% to 93%, empty running dropped by a fifth, and the night re-planning shift was redeployed into customer service rather than cut.

We spent two weeks riding along with dispatchers before writing a line of code, then shipped a planning-only prototype that recommended routes without executing them. That shadow phase ran a month and became the trust argument for the full rollout.