The three documents are the purchase order (what was authorized), the goods receipt or service confirmation (what actually arrived or was performed), and the vendor's invoice (what's being billed). If all three agree on quantity, price, and item, the invoice is cleared to pay. If they don't, it's an exception that needs investigation before money moves.

This is one of the most common processes AI agents are deployed against in finance operations, because the matching logic itself is well-defined, but the actual documents rarely line up perfectly: partial shipments, price adjustments, and unit-of-measure mismatches all create exceptions a human previously had to resolve by hand.

Frequently Asked Questions

What's the difference between two-way and three-way match?

A two-way match only compares the purchase order to the invoice, skipping the receipt confirmation. It's faster but misses cases where an invoice bills for goods or services that were never actually received.

Why do most invoices fail a three-way match on the first attempt?

Common causes include partial deliveries billed in full, minor price changes not reflected in the original PO, tax or freight charges added at invoicing, and unit mismatches (billed by the case, ordered by the unit). Most are legitimate, not fraud, but each still needs review.

Can an AI agent fully automate three-way matching?

It can handle the clean matches end to end and investigate mismatches by pulling the relevant records, but final approval on genuinely ambiguous exceptions, especially ones above a set dollar threshold, typically still routes to a human.

Why does three-way match matter for audit and compliance?

It's a core internal control auditors specifically check for under SOX and similar frameworks, since it's the primary defense against paying for goods that were never ordered or never received.