Order queue

What this demo shows

Orders arrive by e-mail and as PDFs. Instead of retyping them into the ERP, the operator gets the extracted data, sees where in the document each value comes from, resolves the flagged issues and approves. Only then is the import file created.

1. Extraction. An AI model pulls the order number, delivery date, address and lines from the message and the attachment. Every value comes with a quote from the document. If the quote is not in the text, the line needs confirmation.
2. Checks. Plain code, without AI, checks item codes in the catalogue, customer codes, units and packs, withdrawn products, addresses on file and duplicate orders.
3. Human decision. The operator resolves ambiguous lines, unit conversions, replacements and missing data. Missing information goes into a ready-made request to the customer.
4. Import file. After approval a CSV file is created. In a real deployment the format matches the client's ERP import (CSV, XML, EDI or API).

Three examples

• Metalplast Demo: a correct order in the e-mail body. Everything checks out.
• Agro-Serwis Demo: a PDF order with customer codes, “2 rolls” and “3 packs” instead of catalogue units, and a fitting that matches two items.
• Termo-Linia Demo: a resent order (duplicate), a withdrawn grease with a replacement in the catalogue, an incomplete address and no delivery date.

Saved and live extraction

For the three examples the extraction was prepared in advance (Claude did it while the demo was being built), so the demo works offline. The “Extract with AI” button runs a live extraction when the page runs in Claude. Then you can also edit the message text or paste your own. Use only fictional or anonymised data.

What this demo does not show

Integration with a specific ERP, multi-user login or measured savings. In a pilot, time and errors are measured on the client's real cases, against criteria agreed before the start.