We are testing AI order import from documents

We are testing an AI order importer in LiteTMS that reads a transport document, prepares a structured order draft, matches it against data already in the workspace, and shows the forwarder exactly what still needs attention before the order is created.

The goal is simple: stop making a forwarder type information that is already sitting in a PDF, scan, office document, or email.

Two ways to start without a blank form

Matchcodes already handle repeat work. A forwarder can save a proven order pattern, find it by code or name, and use it as the starting point for the next similar job. The route, crew, documents, geofencing, and finance sections can be included in the saved pattern.

The AI importer tackles a different case. The order arrives from outside the system and its layout belongs to the customer, not to LiteTMS. Instead of building a template for every sender, the forwarder uploads the document and asks LiteTMS to read it.

That gives the order creation screen two practical shortcuts. Use a Matchcode when the work repeats. Use AI when the source is a document that somebody would otherwise have to copy field by field.

What LiteTMS reads and prepares

The current test flow accepts PDFs, common image formats, DOC and DOCX files, and EML email files. It reads the complete source rather than asking the user to paste selected fragments.

From that material, it prepares one structured transport order. Depending on what the document contains, the draft can include:

1. The ordering party and a separate payer.
2. Loading and unloading stops, addresses, dates, time windows, references, and contact details.
3. Cargo description, weight, quantity, packaging, pallets, loading metres, volume, ADR class, and temperature.
4. Vehicle and trailer registrations, driver details, price, currency, VAT, and payment terms.

Missing information stays missing. The extraction rules explicitly tell the model not to guess a tax number, registration plate, price, or any other value that is absent from the source.

The review screen is where the useful work happens

Reading the file is only the first half of the job. LiteTMS then compares the extracted data with records already held by the company.

Contractors can be matched by tax number or company name. Stops are resolved against real places. Vehicle registrations are checked against the fleet register. The review screen separates a firm match from a suggestion and from an item that remains unmatched.

This distinction matters. A system that fills every box confidently can hide expensive mistakes. LiteTMS instead produces a clear working summary: which documents were read, what was extracted, what was matched, what is only a suggestion, and what still needs a manual decision.

The forwarder can select another suggested contractor, search for the correct place, correct extracted cargo or finance values, and decide whether to assign a recognised vehicle. Only after that review does the user confirm and create the order through the normal LiteTMS order writer.

What we are testing now

The pipeline is working in LiteTMS and is going through its final tests before it becomes a standard part of order creation. Documents from transport companies vary wildly. A clean generated PDF, a photographed page, an email thread, and a scan with handwritten notes all present different problems.

We are checking more than extraction accuracy. We are testing whether the matching summary is honest enough, whether weak suggestions are clearly marked, and whether a forwarder can correct an exception faster than typing the order from the beginning.

We are introducing the feature gradually and checking it against more document layouts. That gives us time to improve the matching summary and manual correction flow before the full launch.

Why this changes a forwarder's day

Manual entry steals attention twice. First the forwarder reads the document. Then they read it again while copying names, addresses, dates, references, cargo, and rates into the TMS. The difficult part of the job starts only after the typing is finished.

AI order import moves that attention to the exceptions. The document becomes a draft, the company data is matched where possible, and the forwarder reviews a short list of decisions. Matchcodes do the same for recurring work from the other direction: start with a trusted internal pattern instead of an empty form.

That is the efficiency gain we care about. Less copying. More time for the parts of forwarding that require judgement.

The AI importer is in its final testing stage and will soon join the standard order creation flow. You can create a LiteTMS workspace and follow its progress on the blog.