How does a purchase order actually arrive?
Most purchase orders arrive as a PDF by email. Sometimes a supplier sends an Excel file, a Word document, or a printout from their own ERP. Large trading partners use EDI or a supplier portal, but that is the exception rather than the rule. The result: your purchasing department processes dozens or hundreds of documents per week in formats that differ by supplier. Article descriptions, units, prices, reference numbers: they are all there, but rarely in the same place and rarely in the wording your system expects. That is the real problem.
Why manual data entry is especially error-prone for purchase orders
Processing a purchase order is not a simple copy job. A supplier uses their own article codes; you have internal codes in your ERP. A supplier writes '12 units box 250 ml'; your system knows that as '1 carton'. Prices sometimes exclude and sometimes include discounts. The employee entering the order has to make all those translations mentally. Every error that slips through leads to a mismatch between the goods receipt and the invoice. You resolve those mismatches manually, if you catch them in time at all. It is precisely this accumulation of small translation steps that keeps the error rate high and processing time long.
What document AI actually does when processing purchase orders
dottle reads the incoming document, extracts the relevant fields, supplier, order number, line items, quantities, prices, delivery date, and translates these into the structure of your ERP or purchasing module. The translation of supplier article numbers into internal codes is defined once as a mapping; dottle applies it consistently. Deviations, such as a price that does not match the agreed purchase price, are flagged for human review. The employee approves, corrects where needed, and confirms. The manual typing disappears; control stays with the buyer. That is the human-in-the-loop model: dottle does the reading, a person makes the decision.
Which tools automatically match invoices to purchase orders in your ERP?
Matching an incoming invoice to the corresponding purchase order is a common practical question. Document AI approaches this differently from an OCR module or a standalone scan-and-recognise solution. dottle reads the invoice, identifies the order number or article references, and automatically matches these to the open order in your ERP. Discrepancies in price or quantity are flagged; matches are processed directly. This works with SAP, Dynamics 365 Business Central, Exact, and AFAS Profit, without having to configure a template per supplier. A standalone accounting module can often do part of this, but lacks the context to read varying formats without templates.
When does automating purchase orders not pay off?
To be honest about the trade-off: with a small and stable supplier base, ten suppliers each sending a monthly order in a fixed format, automation is rarely urgent. The setup time does not outweigh the time saved. The same applies when source data is structurally poor: incomplete article descriptions, missing references, orders arriving as a scan of a scan. Automation does not fix a broken purchasing process. It makes a sound but labour-intensive purchasing process faster and less error-prone.
When does automating purchase orders pay off?
The strongest business case arises when three things come together: high volume (dozens to hundreds of orders per week), high variation in formats or suppliers, and a system in which you register orders regardless. Wholesalers, manufacturers, and distributors buying from many different suppliers are the typical candidates. As soon as an employee spends more than an hour a day retyping and checking incoming orders, it is worth examining what automation costs and delivers. Two weeks is enough to run a pilot and measure the actual time savings.