- Vendor: Northwind Cloud Services Inc.
- Invoice date: 28 September 2026
- PO reference: PO-1042 (Engineering)
- Cloud credits ($10k): 5 × $9,480.00 = $47,400.00
- Goods receipt: GRN-5502 · 5 of 5 received
- PO-1042 unit price: $9,200.00 (contracted)
- Payment terms: Net 30
Invoice Agent
Three-way match on every line, with holds when price or quantity drifts from the PO and goods receipt.
- Unit price and quantity vs PO and receipt within your tolerance
- Payment blocked when variance exceeds policy until a buyer reviews
Watch it catch a real case.
Synthetic documents, the same steps the product runs. Pause or replay any time.
ProcureAI · Invoice Agent
Invoice price creep on a matched PO
The document arrives. Nothing has been read yet.
Outcome
Payment blocked pending buyer review
A person makes the final callWhat it checks.
AI invoice processing agent.
- Line-level match of quantity, unit price, and extended amount to PO-1042 and the linked goods receipt
- Tolerance check against your policy (for example 2% price variance) with the exact delta shown
- Contract clause cross-check when a price change should require a written amendment
- Duplicate invoice detection on vendor number, amount, and invoice date window
Try it on your document.
- 1
Open the ProcureAI trial and choose invoice, or upload a dummy PDF from your AP team.
- 2
Watch rules run first, then Stellitron Intelligence quote the PO line, receipt, and invoice that triggered each finding.
- 3
Approve or release the hold yourself — nothing pays without your team.
Your team stays in control.
Rules execute before any generative step, so obvious mismatches surface without waiting on a model.
Every narrative finding must cite a document span; ungrounded claims are removed automatically.
Holds, approvals, and payment timing stay in your ERP workflow — the agent prepares the case file.
Why it matters.
Accounts payable teams lose days reconciling invoices that almost match: a penny on unit price, a rounded quantity, a freight line coded differently. An AI invoice processing agent should do the tedious matching, surface only the exceptions, and leave payment authority with people who understand the supplier relationship.
ProcureAI’s Invoice Agent is built for that loop. It reads the PO, receipt, and invoice together, applies your tolerances, and explains each break in plain language with the line it relied on. When price creeps beyond policy — as on NW-2026-09-114 against PO-1042 — it holds payment and shows the buyer exactly what changed.
Stellitron Intelligence coordinates specialist steps, but rules run first. A reviewer subagent must quote the document for every finding; a verifier removes anything it cannot trace. That pattern is the same across ProcureAI and FinanceAI: propose with evidence, then wait for a person.
Worked example on demo data
On synthetic Summit Regional Group data, invoice NW-2026-09-114 arrives for PO-1042. Line 7 bills 140 units at $10.00 each while the PO still shows $9.72 — a 3.0% unit-price increase with no matching amendment on file.
The Invoice Agent places a payment hold, attaches the PO excerpt and receipt quantity, and routes the case to the buyer named on the PO. Stellitron Intelligence’s reviewer subagent must quote those three lines; the verifier drops any claim that is not visible on the uploaded documents.
Your AP lead releases or escalates the hold. The agent never schedules payment on its own.
Questions buyers ask.
Does the invoice agent pay suppliers automatically?
No. It prepares match results, explains variances, and can place holds. Your team approves release to payment.
What documents do I need for a trial?
A purchase order, goods receipt or delivery proof, and the supplier invoice — dummy or anonymised files are fine.
How is this different from a basic OCR tool?
OCR extracts fields. This agent runs your match rules, ties exceptions to clause language, and shows the evidence chain for audit.
Can it run in our cloud?
Enterprise deployments keep documents in your environment. The browser trial uses synthetic Summit Regional Group data or files you upload to the trial workspace.
A GOOD PLACE TO START
Bring the ambition.
We’ll build the way forward.
Let’s talk about your project
A conversation about your goals, constraints, and next step.