How a Mid-Market Finance Team Cut Invoice Processing to Under 10 Seconds
A mid-market accounts payable team processed thousands of vendors invoices a month, everyone keyed in by hand and matched against purchase orders in spreadsheets.
Each invoice took 6-10 minutes. Errors crept into GST numbers, bank details and totals, and a changed bank account or duplicate invoice was only caught by luck.
OnPoint Insights built Invoice AI on Azure OpenAI – reading every invoice with two OCR engines, checking it against master data, and routing it to a reviewer in seconds.
Finance Operations / Accounts Payable
10 Weeks
- Generative AI & AI Agents
- Intelligent Document Processing
- Validation & Fraud Controls
- Approval Workflow Automation
- Analytics & AI Assistant
The Challenge
The team was skilled and the process was defined but every control that mattered depended on someone noticing. Four gaps stood between an invoice arriving and a safe payment going out.
6-10 minutes of typing per invoice
Every invoice was keyed in by hand, and OCR errors on GST numbers, bank details and totals had to be found and corrected manually.
Purchase order matching by spreadsheet
Checking each invoice against its purchase order and goods receipt meant cross-referencing master data by hand, one line at a time.
Fraud caught by chance
Blacklisted vendors, changed bank details and resubmitted duplicates were only spotted if the right person happened to notice.
Approvals with no record
Sign-off happened verbally or over messages, leaving no reliable trail of who approved what, or when.
Our Approach
We didn’t just automate data entry. We rebuilt the controls, so every invoice is read, checked and routed the same way, every time, with the evidence attached.
GPT-4o Vision reads every invoice, PDF, scan or text — while Azure Document Intelligence adds a second read with per-field confidence scores on the fields that cost money when they’re wrong: GST numbers, bank details and totals.
Specialised AI agents validate each invoice against purchase orders, goods receipts, master data and business rules, catching duplicates, calculation errors, price variances, blacklisted vendors and bank-detail changes automatically.
Reviewers see the original file, the extracted data and every flag in a single queue, with one-click approve or reject. Every action is logged with who, when and why, and an AI assistant answers questions across the full invoice history.
We deliberately stopped short of auto-approving invoices. The system does the reading, matching and fraud screening; a reviewer makes the final call with the evidence in front of them. It keeps a named person accountable for every payment, which is what auditors and finance leaders actually need.
The Impact
The team stopped keying data and started reviewing exceptions.
6-10 min → under 10 sec
Around 95% less handling time per invoice, from arrival to a reviewer-ready record.
4-6× more invoices per clerk
Each AP clerk now processes four to six times more invoices a day.
Every variance caught
Price variances and quantity over-billing are flagged automatically on every invoice, not left to spot checks.
Fraud stopped before payment
Real attempts changed bank details, blacklisted vendors and duplicate resubmissions, were flagged before any money left the account.
Explore Other Case Studies
Next Case Study
Digitizing Continuous Improvement for a Global Medical Manufacturer
Real-Time Equipment Visibility for a Global Medical Manufacturer
Enhancing Financial Insights for a Leading Philanthropic Fund
Collaborate with us
We're here to answer your questions and help you find the right solution.
"*" indicates required fields