Where the problem lies in current transport planning
In most transport companies, orders arrive by email, PDF attachments, WhatsApp or customer portals. That information then has to be entered manually into the TMS. Loading address, delivery address, weight, reference number, requested delivery date: every field a click. At twenty orders a day that is manageable. At a hundred it becomes a full-time task that is also prone to errors. As a result, planners spend a large part of their day rekeying data instead of planning. That is precisely the gap where AI adds value.
How AI email processing in transport works
AI email processing in transport runs on document recognition and language models trained on the structure of transport orders. An incoming email with a PDF attachment is opened automatically. The system identifies the relevant fields: loading and delivery location, date, weight, reference number, special instructions. These are converted into a structured order object and submitted to the TMS. The planner sees a proposal with a confidence score. When the confidence is high, approval takes a single click. When there is doubt, the planner makes adjustments. The system does not learn from those corrections on its own; instead, the recognition rules are periodically maintained and refined by the team. This is the approach we follow with Dottle, our document AI.
Automated order entry: what actually changes for the planner
Automated order entry changes the planner's work, but does not replace it. What disappears: manual rekeying, searching for attachments, counting lines on a packing slip. What remains: assessing exceptions, liaising with customers about discrepancies, working through capacity constraints and making decisions when requirements conflict. Those are exactly the tasks where an experienced planner adds value. AI handles the groundwork; the human decides. In our projects, planners almost always describe this as the biggest shift: from typing to thinking.
When does automating transport planning software with AI not fit?
Not every operation is ready for AI automation of order entry. If orders vary widely in format, language and structure with no recognizable pattern, recognition accuracy will be too low to deliver value. If the TMS itself cannot be reached via an API or structured input, integration falls back on fragile screen automation, which is not a solid foundation. And if volume is small enough that the planner handles it comfortably, the business case is thin. Honest advice: start with a measurement. How many orders per day, how many unique customer formats, how stable is the structure of incoming messages? Those three questions determine whether automation makes sense.
Bonsai AI Workers or a new core: which approach fits?
If the existing TMS works well enough and only the input side is the problem, you build an AI Worker around it. It reads incoming messages, processes the documents and feeds the TMS without touching the core system. This approach has a shorter lead time and a lower initial investment. If the TMS itself is outdated, too rigid for the current customer mix or a constraint on growth, rebuilding the core system is sometimes the smarter move. A modern, purpose-built core with AI built in offers more room to operate than a layer on top of a system that has already reached its limits. Bonsai can do both, and the conversation always starts with the same question: where is the real bottleneck?