Before a business pays an invoice, one question matters: did we actually get what we are being billed for, at the price we agreed? Three-way matching is the standard control that answers it. Here is how it works, where it breaks down when done by hand, and how AI has changed the math.
The three documents
Purchase order (PO). What you ordered and the price you agreed to pay. Created when the order is placed.
Receiving document. Proof of what actually arrived — a goods receipt, delivery note, or packing slip, checked in by your receiving team.
Invoice. What the supplier is billing you.
A three-way match compares all three. If the quantities, items, and prices agree, the invoice is approved for payment. If they do not, it is held for review.
A quick example
You order 100 units at $50 each — the PO says $5,000. The warehouse receives 90 units and logs it. The supplier invoices $5,000 for 100 units.
A three-way match catches the gap instantly: you received 90, you are billed for 100. Without the match, that invoice likely gets paid in full and the $500 becomes a dispute you may never recover — or never even notice.
What three-way matching catches
- Quantity mismatches — billed for more than was delivered
- Price changes — invoice price higher than the PO price
- Duplicate invoices — the same invoice submitted twice, sometimes months apart
- Invoices with no PO — spending that bypassed approval
- Fraudulent invoices — bills for goods that never existed
The freight world runs the same control under a different name: matching the carrier invoice against the rate confirmation and the BOL or POD is three-way matching for transportation, and it catches the same categories of errors and fraud.
Why manual matching fails
The logic of three-way matching is trivial. The labour is not:
- The PO lives in the ERP, the receipt in the warehouse system, and the invoice in someone's inbox — just gathering the three documents takes longer than comparing them.
- Line-level checking is tedious. A 40-line invoice against a 40-line PO invites skimming, and skimming misses things.
- Volume forces shortcuts. Many AP teams only fully match invoices above a dollar threshold, which means small errors — and small frauds — flow through unchecked, repeatedly.
The result: the control exists on paper but is applied to a fraction of spend.
How AI automates the match
Modern AP automation removes the labour rather than the control:
- 1Extraction. AI reads the invoice — any vendor, any layout, no templates — and pulls every line item, quantity, price, and reference. The same happens for POs and receiving documents. This is what AI document extraction does in seconds per document.
- 2Matching. The system compares the three documents line by line: quantities, unit prices, totals, part numbers, references.
- 3Exception routing. Clean matches are approved automatically. Only true discrepancies reach a human, with the mismatched fields highlighted side by side.
The economics flip: instead of matching a sample of invoices thoroughly or all invoices superficially, you match everything, at line level, every time — and your team only touches the invoices that deserve attention.
Two-way, three-way, and four-way matching
- Two-way: invoice against PO only. Faster, but cannot catch delivery shortfalls.
- Three-way: invoice against PO and receipt. The standard for goods.
- Four-way: adds an inspection or acceptance record. Used for regulated or quality-critical purchases.
Most businesses should default to three-way matching for physical goods and two-way for services where there is no receiving step.