Inconsistent invoice formats
Every vendor produced invoices differently: PDFs, email-body HTML, free-text notes. Structured extraction had to work reliably across all formats.
Decision Foundry · Case Study
How a mid-market manufacturer used AI to eliminate manual AP matching, automate 94% of invoices, and build an audit trail their auditors could rely on.
The result: AI-powered matching now clears 94% of invoices automatically (847 a month, $2.3M in value), and the average decision takes 3 minutes instead of 2-3 days.
Key Metrics
94%
Auto-match rate
(was 0%)
847
Invoices/month
through the engine
$2.3M
Monthly value
matched automatically
3 min
Avg decision time
(was 2-3 days)
The Customer
The client is a mid-market industrial equipment manufacturer based in Melbourne, Australia. With 280 employees and $40M in annual procurement spend, the AP team of three was processing every invoice manually: matching line items in Excel, chasing PO approvals over email, and signing off month-end in a frantic 3-day sprint.
The Problem
The client's AP team received invoices from 40+ vendors by email and post. Each had to be opened, the PO located, line items compared in Excel, and a sign-off email sent to the relevant manager. With no system to enforce the process, steps were skipped, errors slipped through, and the team had no visibility until month-end reconciliation.
Challenges Faced
Every vendor produced invoices differently: PDFs, email-body HTML, free-text notes. Structured extraction had to work reliably across all formats.
Goods Receipt Notes lived in a separate WMS with no API. Pulling and verifying GRN data against invoices required a custom integration layer.
Auditors required a full, tamper-proof record of every matching decision. A standard database that could be edited after the fact was not acceptable.
The Invoice & PO Matching Automation App ran in dry-run mode for two weeks, showing decisions without acting, so AP staff could verify accuracy before handing over control.
Some vendors had negotiated 1-2% price variance tolerances; others required exact matches. A single global setting would have generated false exceptions.
A single batch could auto-approve 50+ invoices worth hundreds of thousands of dollars. A hard pause at 50 approvals per run was a non-negotiable safety guardrail.
The Solution
The AI-powered Invoice & PO Matching Automation App runs a structured sequence of checks on every invoice, comparing quantities, prices, totals, and receipt data against the purchase order. The AI confidence score tells the AP team exactly why a decision was made. Every decision goes to an immutable audit log before anything is approved.
AI matches each invoice line to its corresponding PO line. Quantity, unit price, and total are compared within configured tolerance bands. The match panel shows both documents side by side with a full comparison table below, so reviewers can see every check at once.
When a goods receipt is available, the Invoice & PO Matching Automation App adds two additional checks: received quantity must equal or exceed the invoiced quantity, and the receipt date must precede the invoice date. All three documents appear simultaneously in the match panel.
Before any result is written to the accounting system, the match outcome is appended to an immutable JSONL audit log. The log cannot be edited or truncated: append only. Every entry records the invoice, PO, score, checks, and decision. Auditors get the full picture on demand.
“In a standard AP system, you can’t see why something was approved or rejected six months ago. With the Invoice & PO Matching Automation App we can pull up any invoice, any decision, and see the exact score, the exact checks, and who touched it. That’s what our auditors needed.”
AP Manager · Mid-market industrial equipment manufacturer
Get In Touch