In short: Reading unstructured email, extracting packing lists and invoices, drafting quotes from your own rates, and what still needs a human.
Most writing about AI in freight is either a vendor's promise or a consultant's forecast. Neither helps the owner of a ten-person forwarder decide what to try on Monday. This article sticks to work that current tools do reliably, work they do with supervision, and work they shouldn't be trusted with yet. No market sizes, no predictions.
The biggest change in the last few years is that software can now read a badly written email the way an experienced operator would. "Hi, need price for 2 x 40 from Ningbo, FOB, to our warehouse in Dandenong, cleared please" contains a container type, a quantity, a port, an Incoterm, a delivery suburb and a clearance instruction. Older rule-based tools needed a form to get that. Current language models get it from the email, including the forwarded thread, the signature block and the reply that says "sorry, make that 3".
That matters because reading is where much of the quoting time goes. It's also where the errors start: a missed "HC", a delivery address in the second email of a thread, an Incoterm that changed halfway through.
Packing lists, commercial invoices and supplier spreadsheets arrive as PDFs and Excel files in a dozen layouts. Extracting carton counts, dimensions, gross weights and invoice values from them used to mean retyping. Current tools handle mixed layouts well, including scanned PDFs, and can total cartons and cubic metres across a packing list.
Two cautions. First, a document can be wrong at the source: a supplier's packing list that says 0.5 cbm for a pallet that's clearly 1.2 is still wrong after extraction. Second, totals should be shown next to the line items they came from, so a person can spot a carton row that was read twice.
Given a structured request and your rate sheet, drafting a quote is lookup and arithmetic. The model reads the request; ordinary code finds the rows and does the sums. That split is important. A language model should not be making up a freight rate or doing currency conversion in its head. It should be choosing which row of your sheet applies, and the numbers should come from the sheet.
Done that way, a draft for a standard request is usually right, and when it's wrong it's wrong in a way you can see: a different port pair, a missed zone, a container size misread. A person still checks every draft before it goes out. The point is that checking a draft is much faster than writing one. Our article on freight quote automation covers which requests to start with.
When a request is missing the delivery postcode or the dimensions, a model can list exactly what's missing and write the question email. This is low-risk and high-value: it gets something back to the customer the same hour, which matters more than people think (see why quote response time decides who wins the freight).
The failure modes are predictable, which makes them manageable.
Start where the work is most repetitive and the risk is lowest: incoming quote requests on lanes you already price. It needs no integration, because it works on email you already receive, and it doesn't touch your freight system's records. If it saves time on quoting, move on to document extraction for bookings. Leave customs and live exceptions until last.
Run it alongside your current process for two weeks. Quote the way you always do, compare the drafts, and count how many you'd have sent with no edits or one. That number, on your own requests and your own rates, is the only evaluation that matters.
Comparing general assistants? See ChatGPT for freight forwarders and Microsoft Copilot for freight forwarders.
Looking beyond quoting? See the best AI tools for freight forwarders in 2026, by job.
Laneworks does the email and quoting part and nothing else. You forward a customer's quote request, with any PDF or Excel attachments, and about two minutes later a draft comes back: the reply, what it read from the email, and the rate-sheet row behind every price. Missing details produce the question email instead. It works by email alongside CargoWise or whatever system you use, and you send everything yourself. There's a sample draft on the site.
Forward one real quote request to try@getlaneworks.com and see the draft for yourself. It's priced from a sample tariff until you send your rate sheet; after that, every line comes from your rates.
Forward a quote request