Logistics runs on documents and exceptions. Everything that moves smoothly is already automated. What costs money is the shipment that does not fit the template, and that is where model work earns its keep.
Document intake
Bills of lading, packing lists, customs paperwork, invoices — scanned, faxed, photographed on a phone in a yard. Extraction models turn them into structured records with a confidence score per field. Low confidence goes to a human queue instead of into the database.
Exception handling
A late truck, a short delivery, a damaged pallet. The routine part is finding the right playbook and the right contact; the judgement part stays with your coordinator. An assistant that drafts the note, attaches the evidence and pre-fills the claim takes the twenty minutes out and leaves the decision in.
ETA and planning support
Prediction here is a data problem before it is a model problem. Historical dwell times, carrier behaviour and seasonality live in systems that rarely talk to each other. Half of this work is integrations and API work, and it has to happen first.
What is different about this sector
- Paper is authoritative and often unreadable
- Partners send data in formats they will not change for you
- A wrong automated decision moves physical goods, so reversibility matters
- Users are on handhelds in bad light and worse connectivity
Where we start
A two-week look at your highest-volume exception type, with a written estimate of what automating it is worth. If the answer is not much, we say so. That review is the AI readiness assessment.
More sector notes on industries, or the whole AI map.