A two-way match is an accounts payable check that compares a supplier's invoice with the purchase order before payment, confirming that price, quantity and terms agree within tolerance. Unlike a three-way match, it does not check that the goods or services were received. It is used for services, subscriptions and low-value purchases where no receipt is recorded.
Why does a two-way match matter once agents act?
A two-way match answers one question: is the business being charged what it agreed? It does not answer the other: did the business get it? When a person processes the invoice, that second question is often answered informally by someone who knows the supplier. An agent has no informal knowledge. If the only check is invoice against order, the agent will happily pay for a service that was never delivered.
So the gap needs a deliberate fill. For services, that is usually a confirmation from the budget holder that the work was done, captured in the record before payment. Making that requirement part of the agent's bounds, rather than a habit someone might skip, is the sort of control AI agents for finance depend on.
How does it work in practice?
- Match the invoice to the order: supplier, PO number, line description, unit price and quantity.
- Apply the tolerance: small differences the business accepts without review.
- Check the service period: the invoice should fall within the dates the order covers.
- Confirm delivery: where the policy requires it, a named person confirms the service was received.
- Approve for payment or hold: matched and confirmed invoices move on; anything else is held with a reason.
The agent executes the matching and checks as routine work. Confirmation of delivery comes from a person. Exceptions, and their outcomes, belong in the same log used for three-way matching, such as the three-way match exception log template, so patterns across suppliers are visible.
What does that look like in an example?
A university's IT department has a purchase order for twelve months of network support at 2,000 a month. Each month the supplier invoices 2,000. The agent matches each invoice to the order and checks it falls within the contract period. In month seven the invoice is for 2,600, with a line for "additional on-site visit". The order does not cover it, so the match fails on price. The agent holds the extra 600 and drafts a query to the IT manager, who confirms the visit happened and raises a change to the order. The 2,000 base charge, which matched and was confirmed, proceeds.
How is a two-way match different from a three-way match?
A three-way match adds the goods receipt as a third document, so it checks that the business received what it is paying for. A two-way match omits it. Three-way is the stronger control and the norm for physical goods; two-way is a practical choice where receipts are not recorded. The risk with two-way is not the match itself but the missing proof of delivery, which is why it needs a confirmation step. Both are covered in three-way matching with an AI agent.
Which related terms should you know?
- Purchase order: the agreed commitment the invoice is matched against.
- Tolerance: the allowed difference before the match fails.
- Non-PO invoice: an invoice with no order, which cannot be matched at all.
Whether to let an agent match at all when orders are vague or incomplete is discussed in can an AI agent match invoices to purchase orders?
Questions people ask
When is a two-way match enough?
For services and subscriptions where a named person confirms delivery, and for low-value goods where the cost of recording receipts outweighs the risk. The policy should set the categories and value limits, and the agent should apply them as written rather than choosing per invoice.
Can an AI agent confirm that a service was delivered?
Not on its own. It can gather evidence, such as a timesheet or a completed ticket, and ask the budget holder to confirm. The confirmation is a person's statement that the business received what it is paying for, and it should be recorded against the invoice.