Invoice Processing Agent · Use case
How can AI match supplier invoices to SAP purchase orders on our own infrastructure?
An invoice processing agent reads each supplier invoice, extracts its fields and compares them with the purchase order held in SAP. Invoices that match are listed for approval, and every mismatch goes to a person with the difference shown. The agent does not post to SAP. It runs on your own infrastructure, so invoices and supplier data stay with you.
Invoices keyed by hand.
And supplier data that cannot leave.
- ✗ 01Accounts payable teams key invoices by hand and compare each one with its purchase order line by line.
- ✗ 02Invoices arrive as PDFs and scans in many layouts, and supplier and pricing data cannot go to a public AI service.
Where it stops for a person
On any mismatch, and before anything is posted.On any mismatch, and before anything is posted. The agent does not post an invoice or a match to SAP. A person does, after approving it.
It matches.
A person approves and posts.
Step · 01
Invoices are uploaded as PDFs or scans.
Step · 02
The agent converts each invoice to text and extracts the supplier, the order number, the lines and the totals.
Step · 03
It reads the matching purchase order from SAP.
Step · 04
It compares the two and marks each invoice as matched or mismatched, with the difference shown.
Step · 05
A person approves the matched invoices and resolves the mismatches.
Reads from · connectors
SAP
File Upload
The agent connects to SAP through the SAP connector to read purchase orders. Invoices arrive through File Upload.
Acts through · MCP servers
Markitdown
Handwriting OCR
MCP servers that turn files into text: Markitdown for documents, and Handwriting OCR for scanned or handwritten pages. Handwriting OCR is a hosted service: a page sent to it leaves your environment, so leave it out where invoices must stay inside. Nothing is written back to SAP.
What you set up
- Connect SAP with read access to purchase orders.
- Set what counts as a match, field by field.
- Name who approves matched invoices and who resolves mismatches.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does the agent post invoices to SAP?
No. It reads purchase orders from SAP and prepares the match. Posting stays with a person.
What happens when an invoice does not match?
It goes to a named person with the differing fields shown. The agent does not resolve a mismatch.
Can it read scanned invoices?
The agent connects to MCP servers that convert PDFs and scans to text before the fields are extracted.
Does invoice data leave our network?
Not with Markitdown: the agent, the model and the invoices stay on your own infrastructure. A page sent to Handwriting OCR goes to that service, so use it only where that is allowed.