The situation
Axiom brokers freight. The paperwork that comes with that — bills of lading, rate confirmations, carrier invoices — arrives as PDFs, dozens of them a day, each one formatted differently depending on the carrier. Someone had to open every single one and type the relevant fields into their TMS by hand.
The team was spending over six hours a day doing this. It was slow, it was boring, and mistakes happened often enough to cause real problems. The people doing it were capable of much more.
What we built
We built a system that reads the documents instead. It uses a language model to extract the structured data out of whatever the carrier sends — and carriers send everything in different layouts with different field names. The extracted data gets checked against their business rules, then written straight into the TMS through their existing API. No one has to touch it.
For the documents the system isn't confident about — about 6% of the volume — we built a simple review screen. Someone can look at the flagged document, confirm or fix the extraction, and move on in under 30 seconds. Nothing gets dropped.
We also put a dashboard together so operations managers could see what was flowing through, what was getting flagged, and how accurate the system was running. More for peace of mind than anything else — but it mattered.
What happened
The system went into full production ten weeks after we started. Thirty hours of weekly data entry effectively disappeared. Error rates dropped 80%. The people who'd been doing that work got moved onto customer relationships and growing the book of business — the stuff that's hard to automate and actually moves the needle.
The ROI wasn't complicated to calculate. It was immediate.
What the client said
"We didn't think AI could handle the variation in our documents. It handles formats we didn't even know we had. The ROI was immediate."
— VP of Operations, Axiom Logistics