The order intake that reads itself

The starting point

Orders arrived the way customers like to send them: as a PDF, an email, an attachment, sometimes a scan. Around a hundred a day, in every format, and each one had to be read and keyed into the ERP by hand. Article numbers, quantities, prices, delivery dates, field by field.

The work was pure transfer: information that already existed on paper had to exist in the system, and people spent their days carrying it across. Every manual entry was also a chance for a typo, and typos in an order become deliveries that come back.

What we built

A reading assistant in front of the ERP. Every incoming order is picked up the moment it arrives, whatever the format. An AI model reads the document, extracts the positions and matches customer and article numbers against the master data in the ERP. What matches cleanly is entered as a draft order; anything unclear, a missing article number, an unusual price, an ambiguous quantity, is flagged for a person.

Nothing is confirmed on its own. The team reviews the exceptions, releases the drafts, and the order takes its normal path from there. Built with standard components (workflow automation plus a language model), running in the house, on the ERP that was already there.

The result

80 percent of manual entry gone, measured against the ERP log: the share of orders arriving as pre-filled drafts versus orders typed from scratch, over four weeks. Transfer errors dropped with the typing. The team’s day shifted from keying in the routine to checking what actually needs judgment.

The AI reads every incoming order, and my team only checks the exceptions. And control stays with us.

CEO, VGG Handels AG

What comes next

The same assistant is set to take on supplier invoices and order confirmations, the neighbouring paperwork of the same process. Each extension follows the same rule. AI proposes, the team decides.