What AI can and cannot do for a transport company

Transport is receiving the same wave of AI promises as every other industry, and the useful version is narrower and duller than the pitch. Worth separating the parts that genuinely work now from the parts that are being sold ahead of themselves.

What it does well

Reading documents. A CMR, a delivery note, a weighbridge ticket, a supplier invoice, photographed badly in a loading bay. Extracting the reference, the date, the quantity and the signature status is a solved problem in a way it was not five years ago, and it is worth more in this industry than almost anywhere else.

Understanding an order in prose. An email that says two pallets from Poznań to Lyon, loading Thursday morning, contains structured data. Turning it into a draft order removes an hour a day of typing.

Remembering. What this lane cost the last four times, how long this consignee usually keeps a driver waiting, whether this customer pays in thirty days or sixty. Perfect recall of things a dispatcher half-remembers is genuinely useful at the moment of quoting.

Drafting the repetitive. The invoice, the payment reminder, the confirmation to the customer, the note to the driver. Composed and waiting rather than written from scratch.

What it does not do

Take responsibility. A rejected invoice, a missed delivery window or a customs error has a consequence and somebody has to answer for it. That somebody is a person in your company, and no tool changes it.

Fix the physical world. A border queue, a breakdown, a consignee who closes at two, a strike at a port. Software can tell you sooner and help you re-plan, and it cannot make the vehicle move.

Replace regulated systems. Tachograph and working-time analysis, national freight reporting, electronic tolling, customs declarations. These have legal weight and specific requirements, and any product implying otherwise should be asked exactly which obligation it discharges.

Guarantee an extraction. A document read badly produces a wrong reference, and a wrong reference produces a rejected invoice weeks later. The extraction is a draft, checked by a person before it reaches a customer, and any vendor promising otherwise is describing an ideal case.

Price the market. Rates move for reasons that are not in your history: a competitor short of work, a seasonal surge, a customer with a problem. The system can tell you what this cost and what you charged; the price is still a judgement.

The honest summary

AI in transport removes typing, chasing and remembering. That is not a small claim — those three things are most of what an office does that is not talking to people.

It does not remove judgement, responsibility or the physical world. A product that pretends otherwise will be discovered by the first delivery that goes wrong, and in this industry that happens in the first month.

Which is why every outbound action here stays a draft. Not caution about the technology — an accurate description of where the technology helps and where a person is still the one who answers the phone.