Using AI to Detect Fraud and Anomalies: A Real Case From an Internal Audit Team

Internal auditors are no strangers to detective work. But one healthcare organization’s audit team discovered how dramatically AI could improve their investigations—especially when the clues were buried in thousands of transactions.

The Background

A regional healthcare group was concerned about rising operating expenses, particularly in facilities management. Nothing seemed blatantly off, but leadership wanted a deeper review. The internal audit department suspected minor issues in procurement: duplicate vendor invoices, unnecessary service fees, and inconsistent approvals.

The problem?
There were 32,000 transactions to review manually.

Introducing AI into the Audit Process

The audit team implemented an AI anomaly detection model trained on:

• Prior years’ expense patterns
• Vendor invoice history
• Typical approval workflows
• Standard price ranges for common services

They fed in three years of data and let the AI surface anything that deviated from expected patterns.

What came back changed everything.

AI’s Key Findings

1. Repeated unexplained price increases from two vendors
   The same HVAC contractor issued invoices with subtle monthly price bumps—none of which were approved under contract terms.
2. Duplicate invoices paid weeks apart
   Humans missed them because the invoice numbers differed slightly.
3. Unusual approvals after hours
   One manager approved 78% of their invoices at odd times—11pm, 2am—indicating possible rubber-stamping.
4. Suspicious rounding patterns
   Many invoices ended perfectly in “.00,” which is unusual for hourly service work.

These weren’t smoking guns, but they were leads—exactly what internal auditors need.

The Investigation

With AI’s flagged anomalies, the audit team dug deeper and conducted interviews. They found:

• A vendor had changed pricing without formal amendments
• A new AP clerk had misinterpreted exceptions and paid duplicates
• A department manager was overwhelmed and approving invoices without review
• The organization was overpaying an estimated $140,000 annually

None of this was malicious—but it was costly.

What Changed After AI

The organization implemented:

• Automated duplicate invoice checks
• Real-time price deviation alerts
• AI-assisted vendor performance dashboards
• Approval workflow reinforcement

One year later, excess spending dropped significantly, and the audit team established a continuous monitoring program.