TL;DR
Three-way matching is an accounts payable control that compares three documents before an invoice is paid: the purchase order (what was ordered), the goods receipt (what was received), and the supplier invoice (what is being billed). When all three agree on quantities, prices, and amounts within tolerance, the invoice is verified and can be approved for payment. It catches errors and fraud, being billed for goods never received, overbilling against the PO, quantity discrepancies, and it is the standard control for purchase-order-based spend.
Done manually, three-way matching is slow and labor-intensive: someone gathers the three documents, compares them line by line, and clears or investigates each invoice. The invoices that match cleanly are easy but still take manual effort; the ones that do not, the exceptions, are where the real time goes. A touchless three-way match is the goal: matching invoices are processed automatically end to end with no human touch, and only the exceptions are routed to a person, so the AP team spends its time on the invoices that actually need judgment.
Reaching touchless matching depends on handling two things well. First, tolerances: small, immaterial differences (rounding, minor shipping variances) should pass automatically rather than creating exceptions, so the process is not drowned in trivial mismatches. Second, exceptions: the invoices that genuinely do not match (a price variance, a partial delivery, a missing goods receipt, a non-PO invoice) require reading and reasoning to resolve, which is exactly where traditional automation stalls and routes everything to humans. Getting truly touchless means automating not just the clean matches but the resolution of the exceptions.
This is where AI that can read any document and reason about discrepancies extends three-way matching to touchless: it reads invoices in any format, matches them within tolerance, and reasons about the exceptions, resolving the ones it can and routing the rest with context, rather than dumping every mismatch into a manual queue. And because AP is an audited, controlled process, the matching must stay accurate and auditable throughout.
This post explains what three-way matching is, how it works step by step, the exceptions that stall it, and how AI reaches touchless matching. To see it in action, watch the Touchless Three-Way Match webinar, and for the automation itself, see 3-Way Match Automation.
What three-way matching is
Three-way matching is a verification control in accounts payable that compares three documents before a supplier invoice is approved for payment, to confirm the invoice is legitimate and accurate. The three documents are the purchase order (the PO, which records what the company ordered, in what quantity, at what price, and was approved before the purchase), the goods receipt (which records what was actually delivered and received), and the supplier invoice (which is the supplier's bill for what they say they provided). When the three agree, on the items, quantities, prices, and totals, within acceptable tolerances, the invoice is verified: the company is being billed for what it ordered and actually received, at the agreed price, and the invoice can be approved and paid.
The purpose of the three-way match is control. It answers three questions at once: did we order this (PO), did we receive it (goods receipt), and are we being billed correctly for it (invoice)? By requiring all three to agree, it catches a range of problems: being billed for goods that were never delivered, being billed for more than was received, being charged a higher price than the PO agreed, duplicate invoices, and outright fraudulent invoices. This is why the three-way match is a cornerstone AP control and a standard requirement for purchase-order-based spend, and why it features heavily in financial controls and audits.
It is worth distinguishing three-way matching from its lighter and heavier variants. A two-way match compares only the PO and the invoice (ordered vs billed), without confirming receipt, it is faster but does not verify that goods actually arrived, and is used for spend where there is no goods receipt to check (like services). A four-way match adds a fourth document, an inspection or quality-acceptance record, confirming the goods were not just received but received in acceptable condition, and is used for quality-critical spend. The three-way match sits in the middle and is the standard for physical goods, because it confirms receipt without the added overhead of an inspection step. For a fuller comparison, see two-way vs three-way vs four-way match: when to use each.
How three-way matching works, step by step
The three-way match follows a clear sequence, and seeing the steps clarifies both how it works and where it slows down:
1. The purchase order is created and approved. Before the purchase, a PO is raised specifying what is being ordered, quantities, prices, and terms, and it is approved. This is the authorized record of the intended purchase, and the reference the later documents are checked against.
2. Goods or services are received and recorded. When the order arrives, receipt is recorded, a goods receipt note capturing what was actually delivered (items and quantities). This confirms what physically came in, independent of what was ordered or billed.
3. The supplier invoice arrives. The supplier sends an invoice billing for what they say they provided, referencing the PO.
4. The three documents are matched. The core step: the invoice is compared against the PO and the goods receipt to check that they agree, the quantities billed match those ordered and received, the prices match the PO, and the totals reconcile, within set tolerances. If everything agrees, the match passes.
5. Matched invoices are approved; mismatches become exceptions. An invoice that passes the match is verified and can proceed to approval and payment. An invoice that does not match, beyond tolerance, becomes an exception that cannot be paid until the discrepancy is investigated and resolved.
Done manually, each of these steps, especially step 4, requires a person to gather the documents, compare them, and clear or flag the invoice, which is why manual three-way matching is slow and consumes so much AP time. The aim of automation is to perform the matching automatically, so that matched invoices flow straight through and only the exceptions from step 5 require human attention.
Why three-way matching stalls: tolerances and exceptions
If three-way matching is conceptually simple, why is it so often a bottleneck? Two things determine whether matching flows smoothly or stalls: tolerances and exceptions.
Tolerances keep trivial differences from becoming exceptions. In the real world, the three documents rarely agree to the penny. Rounding differences, minor shipping or freight variances, small quantity differences from partial shipments, and unit-of-measure conversions create tiny discrepancies on many invoices. If every one of these triggered an exception requiring investigation, the matching process would drown in trivial mismatches. Tolerances solve this: a tolerance is a threshold (a percentage, a dollar amount, or both) within which a difference is considered immaterial and allowed to pass automatically. Well-set tolerances let the clearly-fine invoices flow through while flagging only the material discrepancies, and getting tolerances right is essential to an efficient match, too tight and everything becomes an exception, too loose and real problems slip through.
Exceptions are where the time actually goes. When an invoice does not match beyond tolerance, it becomes an exception, and exceptions are the real bottleneck in AP. Common exceptions include a price variance (the invoice price does not match the PO), a quantity discrepancy (billed more than received, or a partial delivery), a missing goods receipt (the invoice arrived but receipt was not recorded), and a non-PO invoice (there is no PO to match against at all). Each of these requires someone to investigate, why is the price different, was this partially delivered, where is the receipt, and determine how to resolve it. This is judgment work, not mechanical matching, and it is where AP teams spend the majority of their matching time. Traditional matching automation handles the clean matches but cannot resolve these exceptions, so it routes them all to people, which is why matching automation so often plateaus with the clean invoices automated and the exceptions still manual.
The implication is important: reaching a truly touchless three-way match is not just about automating the clean matches (the easy part), it is about handling the tolerances well and resolving the exceptions (the hard part). An automation that only clears the perfect matches leaves the bulk of the actual work, the exceptions, on the AP team's desk.
What touchless three-way matching means, and how AI gets there
A touchless three-way match is the goal most AP teams are working toward: a process where matching invoices are processed from receipt to payment automatically, with no human touch, and only the genuine exceptions are routed to a person. Touchless does not mean no humans anywhere; it means humans are involved only where judgment is actually needed, on the exceptions, while the matching invoices flow through on their own. The measure of success is the touchless rate: the percentage of invoices that complete matching (and move toward payment) with no manual intervention.
The reason most AP operations fall short of touchless is the exceptions. Reaching a high touchless rate requires automating not just the clean matches but as much of the exception handling as possible, and that requires more than rule-based matching, it requires the ability to read varied documents and reason about discrepancies. This is where AI changes what is achievable:
AI reads invoices in any format. Rather than relying on rigid templates, AI extracts the data from invoices however they arrive (varied layouts, PDFs, scans), so the matching is not blocked by document format, a common cause of invoices falling out to manual handling.
AI matches within tolerance and reasons about discrepancies. It performs the match and, when there is a discrepancy, reasons about what it is, a price variance, a partial delivery, a timing difference, a missing receipt, rather than simply flagging no match. This is the difference between detecting an exception and understanding it.
AI resolves the exceptions it can, and routes the rest with context. For exceptions where the resolution is clear (a within-policy price variance, a partial delivery that will be completed, a receipt that can be located), AI can resolve them automatically; for those that genuinely need human judgment, it routes them to the right person with the discrepancy explained and the context assembled, so the human resolves them quickly. This is what lifts the touchless rate: not just clearing perfect matches, but handling the exceptions that would otherwise all go to people.
This is where a deterministic, agentic platform like Kognitos fits touchless three-way matching, honestly scoped. Kognitos is not an ERP or the accounting system of record; it works alongside the existing AP and ERP systems that store the POs, receipts, and invoices. Where it fits is the reasoning-and-exception work that makes matching touchless: reading the invoices (in any format), performing the three-way match within tolerance, and, crucially, reasoning about and resolving the exceptions (price variances, quantity discrepancies, missing receipts, non-PO invoices) that traditional automation cannot, resolving what it can and routing the rest with context, all deterministically and with a full audit trail. Two things make the approach fit three-way matching specifically. First, because matching is a financial control that auditors scrutinize, it must be accurate and auditable, and because Kognitos executes deterministically and logs every step in plain language, every match and every exception resolution is traceable and reconstructable, which is what AP controls and audits require. Second, the value is aimed exactly where the manual effort is, on the exceptions, not just the clean matches, which is what actually raises the touchless rate rather than automating only the easy invoices. This is the capability demonstrated in the Touchless Three-Way Match webinar, delivered by 3-Way Match Automation, and built on the deterministic approach in deterministic AI vs generative AI for finance controls.
Book a working session with a Kognitos solutions engineer • Try Kognitos free
How to approach touchless three-way matching
For an AP or finance team working toward touchless three-way matching, a practical sequence:
Get the tolerances right first. Set tolerance thresholds (percentage and/or dollar) that let immaterial differences pass automatically while flagging material ones, so the process is not clogged with trivial exceptions. This alone raises the share of invoices that flow through cleanly.
Automate the clean matches. Automate the matching of invoices that agree within tolerance so they flow through to payment without manual effort, this is the baseline of a touchless process.
Focus on the exceptions, that is where touchless is won or lost. The touchless rate is limited by the exceptions, so prioritize handling them: use AI that can read varied invoices and reason about discrepancies to resolve the exceptions it can and route the rest with context, rather than sending every mismatch to a person.
Handle the common exception types deliberately. Price variances, quantity and partial-delivery discrepancies, missing goods receipts, and non-PO invoices each have different resolutions; handling them intelligently, rather than as one undifferentiated exception queue, is what drives the touchless rate up.
Keep it accurate and auditable. Because matching is a financial control, ensure the automation is accurate and produces a complete, traceable record of every match and exception resolution, for both control and audit.
Measure the touchless rate. Track the percentage of invoices matched with no human touch, and the exception types that most often break touchless, so you can target improvements where they matter. This connects to the broader effort in how to automate accounts payable.
The throughline: reaching touchless three-way matching means getting tolerances right, automating the clean matches, and, above all, resolving the exceptions with AI that can read and reason, while keeping everything accurate and auditable. The clean matches are the easy part; handling the exceptions is what actually makes three-way matching touchless.
Putting it together
Three-way matching compares the purchase order (what was ordered), the goods receipt (what was received), and the supplier invoice (what is billed) before payment, verifying the company is paying for what it actually ordered and received, which is why it is the cornerstone control for purchase-order-based AP spend. Done manually it is slow, and the real bottleneck is the exceptions, the invoices that do not match because of a price variance, a partial delivery, a missing receipt, or a missing PO. A touchless three-way match, where matching invoices flow through automatically and only exceptions reach a person, depends on two things: well-set tolerances that let immaterial differences pass, and, most importantly, resolving the exceptions, which requires reading varied documents and reasoning about discrepancies rather than routing every mismatch to a human. This is where AI extends matching to touchless: reading invoices in any format, matching within tolerance, and resolving or intelligently routing the exceptions, while keeping the process accurate and auditable as a financial control demands. The clean matches are easy; making three-way matching genuinely touchless is about handling the exceptions, which is exactly where reasoning-capable AI delivers. For the reading capability behind it, see document automation; for the full solution, Finance & Accounting Automation Solutions.
Last updated: July 2026. This article is informational and does not constitute financial advice. Matching controls and tolerances vary by organization; configure them in line with your control requirements.