Invoice intake: extract, validate, post or review
Supplier invoices arrive by email or upload, an AI step extracts the fields, a deterministic layer validates them, and each one is posted, queued for a human, or rejected with the reason.
The problem it solves
AI extraction is easy. The expensive mistakes come after it: the duplicate that gets paid twice, the TRN that does not match the master, the invoice over the limit that nobody approved. Those decisions should not be made by a model.
What it does
- Extracts supplier, TRN, invoice number, date, subtotal, VAT, total and lines from the document.
- Validates against the supplier master and open purchase orders: TRN format and match, duplicate invoice number, lines add to subtotal, total equals subtotal plus 5% VAT, extraction confidence, auto approve limit, supplier limit, PO match.
- Routes by result: AUTO APPROVE posts the bill to the accounting system; REVIEW opens a ticket in the human queue with the reasons; REJECT alerts accounts payable.
- Every decision carries its rule code so a reviewer can see why in one line.
How it was tested
- 12 sample invoices: 7 posted automatically (including one matched to an open PO), 3 sent for review (over limit, VAT total wrong, new supplier), 2 rejected (duplicate number, TRN with 14 digits).
- 12 of 12 outcomes as expected.
Screens
Who it is for
Any business receiving more than 50 supplier invoices a month.
See it running on a call → hamza.jamal@finamatik.com
Built and tested on sample data before any live credential is connected. Every system Finamatik ships is adapted to the client's own tools and owned by the client outright.