The pattern everyone recognises
A customer sends an order. Sometimes via an EDI connection, but more often as a PDF attachment or in the body of an email. A staff member opens the message, reads the addresses, weights, and references, and keys them into the TMS or planning system. Then a quick check: do the addresses look right, is the weight plausible, is the delivery date achievable? With twenty orders a day, that is manageable. With two hundred, it becomes a full-time job. And the moment a staff member is out sick or the market picks up, planning grinds to a halt because of data entry, not because of truck capacity.
Why standard EDI does not fully solve the problem
EDI has existed for decades and solves part of the problem, for the customers that support it. In practice, however, a transport company has dozens of clients, ranging from large shippers with full EDI connections to SMEs that send a PDF built from a homemade template. That mix persists. You cannot require all your customers to implement EDI, especially when the relationship depends on low barriers to entry. So there will always be a group of orders processed manually. That group is precisely the bottleneck.
What automatic order entry with AI does differently
An AI-driven order entry system reads incoming messages regardless of format. A PDF with a consignment note, an email with loose lines, a screenshot from a portal: the AI extracts the relevant fields, loading and unloading address, reference number, weight, number of packages, requested date, and places a draft order in the TMS. The planner sees the order already filled in and only needs to approve or adjust it. The keying work disappears; the control stays with the person. This is the principle we call human in the loop: the AI does the groundwork, the staff member decides.
When does it work and when does it not?
Automatic order entry works well when incoming documents have a recognisable structure, even if the template varies by customer. Documents with consistent fields, a fixed layout, or repeat customers are ideal. It becomes more difficult with orders that are largely free text, where the customer describes in prose what needs to happen without any structure. That requires additional training effort or a different approach. To be straightforward: not every order flow is suitable for full automation. Sometimes partial automation, where the AI fills in the most common fields and routes exceptions to a staff member, is already a significant step forward.
What you need to get started
The barrier is lower than most companies expect. You do not need a data platform, a large IT department, or years of preparation. What you do need: a clear picture of the five to ten most common order formats, access to a representative set of historical documents to test against, and a planner willing to review critically during the first few weeks. In most cases, integration with the existing TMS can be achieved via an API or file exchange. We build this as an AI Worker running on top of the existing system, without requiring you to replace the core system.