35,000 invoice lines a month, no random sampling

Six suppliers at a Swedish importer, each line matched to the price that was agreed

89 to 94%of invoice lines matched to an agreed price
35,000lines a month, checked in a run no person could do
5 minutesinstead of two hours a week
Client
Importer, six factories and manufacturers
Delivered
2026
Read time
2 min
Reference
On a call
StackPlain deterministic codeno model calls

Who

A Swedish importer buying from six factories and manufacturers. Tens of thousands of invoice lines a month, and the CEO checking them himself.

What was wrong

He opened lines at random, compared them to what was agreed, and hoped the rest were fine. Two hours a week, every week, and it still only covered a slice.

What we found

The checking was not hard. It was enormous. Six suppliers, about 35,000 rows a month, each supplier wording its paperwork its own way. Nobody could match all of that by hand, so nobody ever had. And nothing told him what share of his spending had ever been checked at all.

What we built

  • Every line matched to the price that was agreed, however differently the two are worded
  • Its own quality pass to catch wrong flags before they reach him. A few got through early on
  • One page: billed correctly, overcharged, or not checkable at all

What changed

Five minutes of reading instead of two hours of work. Overpayments in the four figures surfaced that nobody knew were there. And for the first time he can say what share of his spending is verified.

What they own now

Plain code, running in their own systems. No AI subscription, no tokens, no per-use fee. Nothing to renew.

The first step

Name the task. I’ll tell you if AI should touch it.

Eugene Zhang, founder of Bon LabsFind your first AI build