Exception handling that keeps a freight network moving

The challenge

Roughly nine thousand shipment exceptions a week arrived as email. Coordinators triaged them by hand, cross-referencing three systems to decide whether a delay needed a re-route, a customer notification or nothing at all. Response time depended entirely on who was on shift.

The approach

1. 01

   Ingested carrier events and exception emails into a single normalised stream.

2. 02

   Used an extraction step to pull shipment identifiers, cause codes and affected stops.

3. 03

   Encoded the re-route decision as deterministic rules with an explicit escalation branch.

4. 04

   Routed low-confidence and high-value exceptions to a review queue with full shipment context.

The outcome

Most exceptions now resolve automatically within minutes of the carrier event. Coordinators spend their time on the genuinely ambiguous cases, and every decision has a traceable reason code attached.