The biannual Association of Certified Fraud Examiners “Report to the Nations” has repeatedly concluded that the longer fraud schemes remain undetected, the greater the losses for the victimized organizations. According to the 2022 report, the average fraud lasts 12 months and leads to a median loss of $117,000. But schemes that last 25 to 36 months result in a median loss of $300,000, and if fraudsters are able to conceal their theft for five years, the median loss rises to $800,000.

Fortunately, advancements in artificial intelligence (AI) and machine learning (ML) have changed the landscape of fraud detection — and shortened the lifespan of fraud schemes. What do these terms mean, and how does the technology help organizations fight fraud?

Explaining the technologies

AI simulates human intelligence and essentially enables machines to “think” like humans. For example, AI can be taught to analyze travel and expense reimbursement claims by looking for anomalies, similar to how a human analyst would. Excessive expenses, inappropriate charges, missing receipts and the lack of an employee’s signature all can be detected by AI software.

ML is actually a kind of AI. It involves training algorithms to become more accurate as they analyze more data. If you put ML to work on expense reimbursement requests, for example, you would feed it both legitimate and fraudulent travel and expense transactions. The technology would then analyze the data set and develop an understanding of how fraud might appear in a transaction and how to identify submissions that are fraudulent. The solution produces a model so that when it receives a new data file that hasn’t yet been analyzed, it’s capable of marking expenses that are potentially fraudulent.

Faster than humans

Humans aren’t completely absent from the ML process. ML requires a human to provide feedback on the software’s accuracy as it’s learning. However, once the model is trained, it can analyze transactions automatically and continuously. And it generally uncovers fraud quickly that might otherwise take months or years to discover by a person conducting an analysis without the help of AI.

So long as a model has been properly trained, there’s no limit to the types of data it can screen — and, unlike people, the model can stay on the job 24/7. You could use AI to flag suspicious numbers in customer and supplier transactions, accounting activities, inventory counts, banking records and financial statements — all of which will help stop fraud schemes in their infancy.

On the horizon

AI and ML technologies are relatively young and are expected to become more innovative and sophisticated in time. For example, current ML and voice recognition software can analyze customer phone calls and develop algorithms to identify fraudulent ones. In the future, the same tools might be used to analyze employees’ responses when they’re being investigated for fraud.

But even now, these technologies are highly effective at detecting fraud and cutting financial losses. They also reduce the time employees must spend on manual processes and they scale easily when data volume increases.

(This is Blog Post #1329)