At its core, invoice matching answers one question: does what's being billed correspond to what was actually authorized and, where applicable, actually received. The specific documents compared, and how many of them, vary by the matching type a company uses.
Beyond PO-based matching, invoices for services under a contract or statement of work get matched against those terms instead, since there's often no physical receiving record for a service the way there is for goods. The matching logic has to adapt to what kind of purchase it is.
Is invoice matching just another name for three-way match?
Three-way match is one specific method of invoice matching. The broader term covers any comparison of an invoice against supporting documentation, whether that's a PO, a receipt, a contract, or a statement of work, depending on what kind of purchase it is.
How is matching a services invoice different from matching a goods invoice?
Goods have a physical receiving record, was it delivered, in what quantity. Services often don't, so matching relies on other evidence: timesheets, milestone confirmations, or a manager's sign-off that the work was actually completed as billed.
What percentage of invoices typically match cleanly on the first pass?
It varies widely by company and vendor mix, but a meaningful share of invoices, sometimes a third or more, need some manual touch on the first attempt, which is exactly the volume an AI agent is positioned to absorb.
How does an AI agent decide what counts as a close-enough match?
By applying configurable tolerance thresholds (a small percentage or dollar variance) set by the finance team, auto-clearing what falls inside that band and escalating anything outside it, rather than requiring an exact-to-the-cent match on every invoice.