Most of the delay was not the truck. It was the exception that sat in a queue until someone noticed a missed scan, a bad appointment, or a load that did not match the order.
The operation already had the events. They were just scattered across the yard system, the carrier feed, and a shared inbox.
Exceptions now land with an owner while the freight is still en route. Resolution ran 2.4 times faster, because the work started before the dock, not after the complaint.
What we did
- Named the few exception types that actually move cost and service.
- Pulled those events into one queue with a clear owner.
- Measured time-to-resolution, not model accuracy.