Procure-to-Pay

The Purchase Order Process, and Why Three-Way Matching Matters

Purchase orders are the least loved part of procurement and the most load-bearing. They are what lets you check an invoice against something other than trust. This guide covers the process step by step, when a purchase order is worth raising, how three-way matching works in practice, and the benchmarks that tell you whether yours is working.

9 min read
Warehouse worker holding a goods receipt form on a clipboard above a sealed carton

A purchase order is a simple thing. It is a written instruction to a supplier: what you want, how much of it, at what price, delivered when and where.

Its real purpose only becomes clear later. When the invoice arrives, the purchase order is the only independent record of what was agreed before the supplier had any reason to shade the numbers.

Without one, checking an invoice means asking someone whether it looks about right.

The process, step by step

StepWhat happensWhat can go wrong
1. RequestSomeone identifies a need and raises a requisitionVague descriptions that nobody can price or receive against
2. CheckBudget, category and preferred supplier are confirmedSkipped entirely, so the first check is at invoice stage
3. ApproveThe right person authorises it by value and categoryToo many approvers, or approvers who rubber-stamp
4. IssueA purchase order goes to the supplier with a PO numberOrder placed by phone or email, with the PO raised afterwards
5. ReceiveGoods or services are recorded as receivedNobody records receipt, which breaks the match later
6. MatchInvoice is checked against the order and the receiptOnly two of the three are compared
7. PayPayment is released on agreed termsPaid early, late, or twice

Step 5 is the one most often missing in mid-sized companies, and it is the one that makes step 6 possible. If nobody confirms that something arrived, the invoice is being checked against an order alone.

When is a purchase order worth raising?

Not for everything. A purchase order has a cost in time, and applying it to every £30 purchase produces resentment and workarounds rather than control.

A sensible split for most mid-sized companies:

Purchase order required
Anything above a threshold you set deliberately, anything from a new supplier, anything recurring, and anything where delivery needs to be confirmed later.
Catalogue or card, no individual PO
Low-value, high-frequency, routine items from an approved supplier at an agreed price. The control lives in the catalogue rather than in each order.
Contract-backed, no PO per invoice
Rent, utilities, insurance and similar. The contract is the control; a monthly PO adds nothing.

Write the threshold down and review it annually. Thresholds that were sensible five years ago quietly become bureaucracy.

Three-way matching, explained plainly

Three-way matching means three documents must agree before an invoice is paid:

  1. The purchase order — what you agreed to buy, and at what price.
  2. The goods receipt — what actually arrived, and how much of it.
  3. The invoice — what the supplier is charging you for.

If all three agree within tolerance, the invoice can be paid without anyone thinking about it. If they do not, it becomes an exception for a human to resolve.

Two-way matching compares only the order and the invoice. It is faster, and it suits services where there is nothing physical to receive. But it cannot detect goods that were charged for and never delivered. That is precisely the weakness billing fraud exploits.

Why the match is a fraud control, not just admin

The Association of Certified Fraud Examiners studied 1,921 fraud cases across 138 countries. Billing schemes mean paying for things never received, or paying inflated amounts. They appeared in 31% of cases at organisations with fewer than 100 employees, against 22% at larger ones.

Payment tampering showed an even wider gap: 23% of cases at smaller organisations against 9% at larger ones.

The difference is structural. Smaller teams find it harder to separate who orders, who receives, who approves and who pays. A working three-way match still functions when the same small group does several of those jobs. It requires three separate records to line up.

How fast should this be?

Two benchmarks are worth holding yourself against.

On the front end, Procurify's data across more than 250 mid-market organisations shows requisition-to-purchase-order times of about 58 hours in companies of 100 to 500 people. In those of 500 to 2,000, it is 65 hours. Their guidance is that beyond 72 hours the waiting itself becomes the bottleneck. The Hackett Group's leading procurement organisations run this step 58% faster than their peers.

On the back end, Ardent Partners puts the average cost of processing an invoice at $9.40 and the average cycle at 9.2 days, with 32.6% processed with no human touch. The best-performing 20% process for $2.78 in 3.1 days, with 49.2% touchless.

MeasureTypicalStrong
Requisition to purchase order58–65 hours (mid-market)Under 24 hours
Invoice processing cost$9.40$2.78
Invoice cycle time9.2 days3.1 days
Invoices needing no human touch32.6%49.2%
Requisition rejection rate3.6%–5.5% (mid-market)Low and stable
Mid-market cycle times and rejection rates from Procurify (2023–2025 data); invoice figures from Ardent Partners' AP Metrics That Matter in 2025.

One caution on the rejection rate. A very low figure is not automatically good — it can mean nobody is checking. Watch the trend rather than the absolute number.

The five failures that cause most of the pain

1. Retrospective purchase orders

Someone orders by phone, the invoice arrives, and a purchase order is raised afterwards to make the paperwork balance. The PO now proves nothing, because it was written to match the invoice.

If this is common, treat it as a symptom. People do it when the proper route is too slow to use in real situations.

2. No goods receipt

Very common where deliveries go straight to a site or a person rather than a goods-in function. The fix is usually to make receipting someone's explicit job for each location, and to accept a simple confirmation rather than a formal process.

3. Vague descriptions

"Consultancy services — £12,000" cannot be received against, matched against, or challenged. Descriptions should say what, how many, and over what period.

4. Approval chains nobody can justify

Each extra approver adds delay and reduces attention, because approvers who see everything stop reading anything. Count your steps. If a step has never rejected anything, it is not a control.

5. Exceptions with no owner

Mismatches pile up in a queue nobody owns, then get released in a batch at month end because the supplier is chasing. That is the control failing quietly. Give exceptions a named owner and an age limit.

What to automate, and in what order

Purchase order work is high volume and rules-based, which makes it a good automation candidate — but only after the process is right.

The Hackett Group's leading procurement organisations process requisitions, purchase orders and purchase order change orders entirely electronically, and handle 27% more requisitions electronically than their peers. Deloitte's 2025 survey found purchasing among the top three areas where technology has actually been deployed, at 75%.

A sensible order: fix approval thresholds first, because that costs nothing and usually delivers the biggest time saving. Then automate purchase order creation from approved requisitions. Then automate matching, with tolerances. Then move routine items into a catalogue so most orders never need individual handling at all.

A short health check

  1. What share of your invoices arrive with a valid purchase order number? Below about 70% and the process is not really in place.
  2. How many purchase orders were raised after the invoice date last month?
  3. How long does a requisition take to become a purchase order, measured as a median?
  4. How many invoices are sitting unmatched right now, and how old is the oldest?
  5. Who owns exceptions, and what is the target for clearing them?

Those five numbers take an afternoon to gather and will tell you more than a process map.

Common questions

What is the purchase order process?

Seven steps: someone raises a requisition; budget, category and supplier are checked; the right person approves it by value; a purchase order is issued to the supplier; goods or services are recorded as received; the invoice is matched against the order and the receipt; and payment is released on agreed terms. The receipt step is the one most often missing, and without it the match cannot work properly.

What is three-way matching?

Checking that three documents agree before paying: the purchase order (what you agreed to buy and at what price), the goods receipt (what actually arrived), and the invoice (what you are being charged). If all three agree within tolerance, the invoice can be paid automatically. If not, it becomes an exception for someone to resolve.

What is the difference between two-way and three-way matching?

Two-way matching compares only the purchase order and the invoice. It is faster and reasonable for services where nothing physical arrives. But it cannot detect goods charged for and never delivered, which is exactly the weakness that billing fraud exploits. Three-way matching adds the goods receipt as independent confirmation.

Should every purchase have a purchase order?

No. Raising one for every small purchase costs more time than it saves and encourages workarounds. Require purchase orders above a deliberate threshold, for new suppliers, for recurring spend, and where delivery must be confirmed later. Use catalogues for low-value routine items and let the contract be the control for things like rent and utilities.

What matching tolerance should we set?

Enough that ordinary rounding and small price differences pass automatically — for example the lower of 5% or a fixed cash amount. Without tolerances every minor difference becomes a human decision and the matching process costs more than it prevents. Set the tolerance deliberately, write it down, and review it.

How quickly should a requisition become a purchase order?

Procurify's mid-market data shows about 58 hours in companies of 100 to 500 people and 65 hours in those of 500 to 2,000, with their guidance that beyond 72 hours the waiting becomes the bottleneck. The Hackett Group's leading organisations run this step 58% faster than peers. Under 24 hours is a strong target for routine requests.

Why do people raise purchase orders after the invoice arrives?

Almost always because the proper route is too slow to use when the need is real. A retrospective purchase order proves nothing, since it was written to match the invoice it is meant to check. Treat a high rate of these as evidence that approval thresholds or cycle times need fixing, not as a discipline problem.

Is a purchase order process a fraud control?

Yes, particularly the matching step. The ACFE's 2024 study found billing schemes in 31% of cases at organisations with fewer than 100 employees against 22% at larger ones, and payment tampering in 23% against 9%. Smaller teams struggle to separate who orders, receives, approves and pays. A genuine three-way match still works in that situation because it requires three separate records to agree.

What share of invoices should have a purchase order?

There is no published universal benchmark, but as a practical rule, if fewer than about 70% of invoices arrive with a valid purchase order number, the process is not really operating. Measure it before trying to improve anything else, and look at which suppliers and departments account for the gap.

What should we automate in the purchase order process?

In order: fix approval thresholds first, since that costs nothing and usually saves the most time; then automate purchase order creation from approved requisitions; then automate matching with sensible tolerances; then move routine items into a catalogue so most orders need no individual handling. Hackett's leading organisations process requisitions, purchase orders and change orders entirely electronically.

Sources

  1. Procurify, 2026 Mid-Market Procurement Benchmark Report, 250+ organisations, 2023–2025 platform data. Requisition-to-PO 58 hours (100–500 employees) and 65 hours (500–2,000); the 72-hour guidance; rejection rates 3.6%–5.5%.
  2. Ardent Partners, AP Metrics That Matter in 2025, Cost per invoice $9.40 and $2.78 vs $12.88 (pp.18, 25); cycle time 9.2 days, 3.1 vs 17.4 (pp.16, 27); touchless 32.6% and 49.2% (pp.19, 28).
  3. The Hackett Group, 2025 Digital World Class Procurement research, 14 July 2025. 58% shorter requisition-to-purchase-order cycle times.
  4. The Hackett Group, What's the Digital World Class Procurement Advantage?, 24 October 2023. 100% electronic processing of requisitions, POs and change orders; 27% more requisitions processed electronically than peers.
  5. Association of Certified Fraud Examiners, Occupational Fraud 2024: A Report to the Nations, Figure 24, p.34: billing schemes 31% of cases at organisations under 100 employees vs 22% at 100+; payment tampering 23% vs 9%.
  6. Deloitte, 2025 Global Chief Procurement Officer Survey, Purchasing among the top three areas of next-generation technology adoption at 75% (p.11).

Want this run for you?

We take on the transactional half of procurement — invoices, purchase orders, supplier data and indirect spend — inside your own systems and under your approval rules. Start with a free spend audit: we measure your volumes, cycle times and exception rates, and the report is yours whether or not you go further.

Book a free spend audit