Purchase-to-pay is the end-to-end process from identifying a need to buy something to paying the supplier for it. Its stages are requisition, approval, purchase order, receipt of goods or services, invoice, matching, payment and reconciliation. It is where most of a business's outgoing money is controlled, so the hand-off between stages is where fraud and error tend to enter.
Why does purchase-to-pay matter once agents act?
Purchase-to-pay has a sharp gradient of risk. Reading a requisition is harmless. Drafting an order is reversible. Matching an invoice is checking. Releasing a payment moves money that may not come back. An agent can work across the whole process, but its depth should rise and fall with that gradient, not stay fixed at one level.
The final step deserves a fixed rule. The agent prepares payments and may schedule them, but approving the payment run belongs to a person who did not prepare it. That separation is the heart of running AI agents for finance safely, and it is examined in can an AI agent pay supplier invoices?
How does purchase-to-pay work in practice?
- Requisition: someone requests goods or services. An agent can check it against policy and catalogue prices.
- Approval: the budget holder approves. A person.
- Purchase order: a purchase order is issued. Drafted by an agent; issued directly only for routine items within limits.
- Receipt: goods are received or services confirmed, recorded by the person who received them.
- Invoice: the supplier invoices. An agent reads and codes it.
- Matching: invoice, order and receipt are compared. An agent matches within tolerance and holds exceptions.
- Payment: payments are scheduled by the agent and approved by a person.
- Reconciliation: payments are reconciled to the bank and supplier statements.
Step 1 is covered in detail in how to process purchase requisitions with an AI agent.
What does that look like in an example?
A logistics firm buys tyres for its fleet under a framework contract. A depot manager requisitions 40 tyres. The agent checks the price against the contract and drafts the order, which is under the routine limit and is issued. The depot records receipt of 38 tyres. The supplier invoices for 40. The agent matches order, receipt and invoice, finds the shortfall of two, and holds the invoice with a note asking whether two more are coming. It does not pay 38 and short-pay the supplier on its own initiative. The accounts payable clerk confirms with the depot and agrees a credit note with the supplier.
How is purchase-to-pay different from procure-to-pay?
The terms are often used interchangeably. Where a distinction is drawn, procure-to-pay starts earlier and includes sourcing: choosing suppliers, running tenders and negotiating contracts. Purchase-to-pay starts once a supplier and price exist. For agents the difference matters because supplier selection and contract negotiation are commercial judgements with their own approvals, while purchase-to-pay is mostly applying terms already agreed.
Purchase-to-pay is also broader than accounts payable. Accounts payable covers invoices and payments; purchase-to-pay includes the requisitions, orders and receipts that come before them.
Where Pryme fits
In Pryme, each step of purchase-to-pay is a separate capability with its own certified depth. An agent can be allowed to match invoices, held on purchase orders above a value, and never permitted to approve the payment run. A supplier invoice arriving by email is an external source, and an external claim cannot authorise a payment. Each verdict is filed with the bound that decided it.
Questions people ask
Where do most purchase-to-pay errors come from?
Most come from the hand-offs: an order raised without approval, a receipt never recorded, an invoice with no order, or supplier bank details changed without verification. An agent that checks every hand-off consistently catches more than a team checking by sample.
Should one agent run the whole purchase-to-pay process?
It can work across the process, but its rights should differ by stage, and no single agent should both create a supplier and approve payments to it. Splitting incompatible duties applies to agents just as it does to people.