Invoice matching is conceptually a comparison between documents and practically a question about tolerance. Real invoices differ from real orders in small ways all the time, through rounding, freight added at billing, part deliveries and price movements, and the entire experience of a matching process is determined by which of those differences a person has to look at.
Two-way and three-way
A two-way match compares the invoice with the purchase order, which is what you use for services and anything without a physical delivery. A three-way match adds the goods receipt and is what protects you from paying for goods that never arrived. Which applies is a property of what was bought, so a system should support both rather than forcing one.
What is compared
Supplier identity, unit price, quantity and total, ideally at line level rather than document level. Line level is what lets several invoices settle against one ordered line and lets a part delivery match cleanly. Document-level matching looks simpler and produces exceptions on perfectly normal situations, which teaches people to override rather than investigate.
Setting tolerance from evidence
Start somewhere reasonable, then look at a month of exceptions and see what actually caused them. If most are rounding, the tolerance is too tight. If genuine price increases are passing unnoticed, it is too loose. A percentage with an absolute cap is the usual shape, and reviewing it once against real exceptions is worth more than any amount of upfront theorising.
Questions people ask about invoice matching
What if the goods receipt is missing?
The match cannot complete, and the right response is to chase the receipt rather than to override. Overriding regularly is how a control becomes a formality.
Can matching be done by the accounting system?
Many can match at document level. Line-level matching with running quantities is less common and is what makes part deliveries workable.
What proportion of invoices should match cleanly?
It is capped by how many have an order and a receipt behind them, so measure that first. A published benchmark tells you nothing about your own ceiling.