Do we need a purchase-order discipline before starting?

For matching benefits, yes. For extraction and coding, no — those help even in a non-PO environment, and the resulting spend visibility often justifies introducing purchase orders afterwards.

Can AI approve invoices?

It can propose approvals and clear low-value, fully matched items under an agreed threshold. Releasing payment and changing bank details should always require a person.

What about e-invoicing mandates in Europe?

Structured e-invoicing removes much of the extraction problem for the suppliers that use it, but mixed estates persist for years. Design for both, and treat mandate compliance as a separate workstream.

AI use case

AI in procure-to-pay

Purchase requests, supplier invoices, matching and payment: where AI helps in procure-to-pay, what it needs, what it risks and how to measure the return.

Miguel Torres, Founder, Merjora · Updated 4 September 2026

In short

In procure-to-pay, AI is used to read supplier invoices and delivery documents of varying layouts, code them to the right account and cost centre, support two- and three-way matching, and surface duplicates and out-of-policy spend before payment. Approval and payment release stay with people. The measurable effects are a higher touchless-invoice rate, fewer duplicate payments and more early-payment discounts captured because invoices stop sitting in inboxes.

The workflow today

Requisition → approval → purchase order → goods or service receipt → supplier invoice → matching → exception handling → approval → payment → reconciliation.

Where AI helps

Extracting header and line-item data from supplier invoices with no consistent layout
Suggesting general-ledger account, cost centre and VAT treatment based on history
Matching invoices to purchase orders and receipts, including partial deliveries
Detecting probable duplicates across format and reference variations
Flagging invoices without a purchase order and identifying recurring maverick spend
Drafting supplier queries when a document is incomplete

Common use cases

Supplier invoice capture and coding
Two- and three-way matching support
Duplicate and anomaly detection
Non-PO spend identification
Contract-versus-invoice price checking

Expected benefits

Higher share of invoices processed without manual touch
Earlier visibility of accruals
More early-payment discounts captured
Fewer duplicate and overpayments
Cleaner spend data for supplier negotiation

Implementation complexity

Moderate. Extraction is mature; matching against messy purchase-order and receipt data is the hard part. Companies without disciplined goods receipting should expect the matching benefit to be limited until that is fixed.

Data requirements

Historical supplier invoices across the real supplier mix
Purchase order and goods receipt records
Chart of accounts, cost centres and past coding decisions
Supplier master data with tax identifiers and bank details
Contract price lists where applicable

Risks and controls

Payment fraud via altered bank details — bank-detail changes must never be automated
Incorrect VAT treatment creating a compliance exposure, especially cross-border
Over-trusting extraction on high-value invoices; set a review threshold
Automating around a broken receipting process instead of fixing it

Example workflow

Invoice arrives in the shared AP inbox
Data extracted, supplier identified and duplicate check run
Matched against the purchase order and receipt where one exists
Clean matches routed for approval; exceptions routed with the reason stated
Coding suggested for non-PO invoices, confirmed by the approver
Payment proposal assembled, with a person releasing it

KPIs to track

Touchless invoice rate
Cost per invoice processed
Average approval cycle time
Early-payment discounts captured versus available
Duplicate payments detected before release

ROI considerations

Processing cost per invoice is the easiest saving to defend; discounts captured is often the larger one
Value depends heavily on invoice volume — below a few thousand invoices a year, scrutinise the case carefully
Include the cost of maintaining supplier master data in the running cost
Avoided duplicate payments are real but lumpy; do not annualise a single event

Frequently asked questions

Do we need a purchase-order discipline before starting?: For matching benefits, yes. For extraction and coding, no — those help even in a non-PO environment, and the resulting spend visibility often justifies introducing purchase orders afterwards.
Can AI approve invoices?: It can propose approvals and clear low-value, fully matched items under an agreed threshold. Releasing payment and changing bank details should always require a person.
What about e-invoicing mandates in Europe?: Structured e-invoicing removes much of the extraction problem for the suppliers that use it, but mixed estates persist for years. Design for both, and treat mandate compliance as a separate workstream.

Could this apply to your business?

Merjora quantifies what this workflow costs you today and what changing it is realistically worth.

Discover your AI opportunities