Built into operations, not bolted on.

Logistics is full of AI proposals that assume the operator will change their behavior to accommodate the model. They will not. The systems are entrenched, the margins are thin, and the cost of a wrong recommendation is direct.

The use cases that actually land are the ones that integrate with the TMS, the freight forwarder's document workflow, or the dispatcher's screen - without asking the operator to leave their tools. Exception handling, document classification, and demurrage prediction are early winners.

Trade documentation and customs compliance are unusually high-leverage targets. These workflows are document-heavy, rule-heavy, and slow - exactly where LLMs reduce cost without changing what the operator does day-to-day.

Readiness → Build → Operate, on operations time and operations margins. The sponsor is usually an operations leader with IT as the integration gatekeeper, and the decision is unsentimental: the return has to be direct - cost out, time saved, or errors avoided - and it has to work without pulling crews off the tools they already use.

Integration is the whole game. Systems like the TMS and EDI flows are entrenched, and the cost of a wrong recommendation is immediate, so we design to integrate with existing systems rather than replace them, and we keep humans in control of consequential calls. We would rather automate a document step reliably than automate a dispatch decision unreliably.

A typical first build targets a high-volume, document- or exception-heavy workflow - exception handling, trade-document classification, or customs-filing review - integrated into the dispatcher's screen or the forwarder's document flow, with a measurable operational metric agreed in Readiness.