The invoice that checks itself

The starting point

Invoices arrived the way invoices do: as PDFs in a shared mailbox, in every layout a supplier can think of. Each one had to be opened, assigned to the right case, checked line by line against the underlying order, and typed into the accounting system. Deviations surfaced late, often only when the numbers no longer added up at the end of the month.

None of this was anyone’s favourite work. It was simply necessary, and it was eating the mornings of the very people whose judgment the business needed elsewhere.

What we built

A matching assistant on top of the systems already in place. When an invoice lands in the mailbox, a workflow picks it up, reads it with an AI model, and extracts the positions: amounts, quantities, references. It then matches them against the corresponding order and marks every deviation it finds, a price that moved, a quantity that doesn’t fit, a position that shouldn’t be there.

The result is a pre-filled booking proposal with the deviations highlighted. Nothing is booked on its own. The team reviews, corrects where needed, and releases. Built with standard components (workflow automation plus a language model), so the setup runs in-house and the team can operate it without us.

The result

Processing time per invoice halved, measured over four weeks of before-and-after timing on the same invoice types. Roughly nine out of ten invoices now arrive as a pre-checked proposal, and deviations are visible the day the invoice arrives instead of at month-end. Same team, twice the throughput.

Invoices that used to swallow whole mornings are matched before the first coffee. It simply works.

Regional Director, LBS Süd

What comes next

The same pattern extends to neighbouring paperwork: order confirmations and reminders are next. The assistant stays in the house, and every extension follows the same rule. AI proposes, the team decides.