Supplier invoices arrived as email attachments in formats that varied by supplier and sometimes by month. Each was opened, read, keyed into the finance system and filed. The work was repetitive, easy to get slightly wrong, and grew with the business.
Previous attempts had failed on the same point: the tooling handled the clean cases and had nothing to say about the rest, so the team kept a manual process running alongside it and gained nothing.
We built an automated path for the invoices that follow a predictable shape, and treated the exceptions as a first class part of the design rather than an afterthought.
Invoices are read on arrival and the fields that matter are extracted and matched against the purchase order.
Anything confident and matching is posted automatically.
Anything uncertain goes to a review queue with the uncertainty shown, so a person checks the specific field in doubt rather than re-reading the document.
Every decision is logged, so a posting can be traced back to the document and the rule that handled it.
We measured the exception rate during the trial before agreeing the scope, so the saving was based on the real mix of invoices rather than the tidy ones.
The routine majority is handled without a person touching it, and the remainder arrives in one queue with the reason for review attached.
Straightforward invoices post without manual keying.
Exceptions are visible and explained rather than discovered at month end.
Each posting can be traced to its source document.
The finance team’s effort scales with the exceptions rather than with volume.
The system was built to be inspected: when it declines to post something, it says why.