AI fraud detection: how to spot fake receipts

With the democratization of generative AI, creating a fake receipt has never been easier. Companies must adapt their controls to this new reality.

New forms of fraud

• AI-generated receipts: receipt images created from scratch using tools like ChatGPT or Midjourney
• Amount modification: digital alteration of amounts on real receipts
• Sophisticated duplicates: same expense submitted by different employees

How EVA detects fraud

EVA uses multiple layers of analysis:

• Image analysis: detection of typical artifacts in AI-generated images

• Cross-verification: matching with merchant data (tax ID, address, hours)

• Pattern detection: identification of suspicious behavior over time

• Cross-checking: comparison across all employees' expenses

An ongoing race between generation and detection

AI for generating fake documents and AI for detecting them evolve rapidly — but not necessarily at the same pace. Some fake receipts may temporarily slip through the cracks and get reimbursed.

This is precisely where EVA makes a difference: beyond real-time control, EVA enables retroactive audits on old receipts that have already been reimbursed. Every 6 months or more, as detection models improve, you can run additional checks on your expense report history to identify fraud that went unnoticed at the time of initial reimbursement.

This post-payment audit capability ensures that even if fraud escapes initial control, it will be caught later thanks to continuous advances in detection AI.

Concrete results

Our clients detect on average 3x more anomalies than with manual control, with a false positive rate below 2%.