How Artificial Intelligence Uncovered Evidence of Fraud

digital folders.jpg


Gary Krausz, CPA/CFF, Gursey Schneider LLP
John Colthart, General Manager, Audit & Assurance, V.P. Growth, MindBridge Ai

While many industries have adopted artificial intelligence (AI), mostly in the tech sector, its adoption rate in financial services has been relatively slow. Even though AI is far from a pipe dream, 76% of banking CXOs agree that adopting AI will be critical to differentiate in the market, the technology tends to conjure up images of sci-fi movies, with ambitious computers replacing humans. The reality of AI is much more collaborative and effective. The story of how our audit services firm in California analyzed more than $2.8 million in fraudulent transactions provides the world’s first case of an AI-enhanced investigation of a real human CPA.

Dismantling the fraud
We were contacted by a consumer products manufacturing client to write an expert report on fraud they discovered that was committed by a CPA. While we do have the investigative expertise, there was no way any team could go through the amount of transactions they had in any reasonable amount of time. AI was the perfect alpha patient for this investigation.

The client had $120 million in annual sales and more than $6 million EBITDA without taking the fraud loss into account. They had suspected fraud to have occurred across three years and requested an examination of their general ledger for the period between 2014 and Q2 2018, to the tune of 6.2 million transactions. The evidence provided was to be used to prosecute their case.

It was very clear that the old ways of internal audit with sampling didn’t make sense here, as once you’ve exceeded the capabilities of Microsoft Excel or traditional CAAT tools, your auditor quickly runs into a wall. AI gave us the ability to take very big data sets and make them manageable, as it took all the data and knew how to put different thoughts together and infer relationships between items.

Unlike the traditional methods in our industry of relying on hunches and instinct to pick an account and start digging, AI offered up its own “intuition” based on the current data set and the learnings from prior analysis. In this case, without any tuning by us, the AI analyzed the entire data set from the client’s general ledger and flagged the fraudulent transactions in a very short amount of time, including identifying items of large amounts posted into unusual accounts and items posted by the same person multiple times.

Redefining fraud investigation
While AI removes the laborious tasks of data ingestion and identifying samples, we still needed to perform the investigation and write the reports. It’s similar to when the bomb squad comes in and the first thing they do is identify the signature of the device. AI gave us the fraud signature and evidence, then it was up to us to do the actual dismantling. Understanding the signature of the fraud allowed us to select a large sample of all other transactions that had the same common characteristics.

This case has cemented the need for artificial intelligence-based auditing to handle current and future amounts of transaction data. Firms would be well advised to start planning their adoption strategies now. With organizations amassing enormous amounts of data across different data sources, it’s difficult to investigate a reasonable number of transactions to inspire confidence. No audit firm would ever sign up to do so. AI offers the opportunity to ingest and inspect 100% of a client’s accounting transactions and go beyond simple rules reporting (sorting and filtering). It allows examiners to dig deeper into insights based on the behavior of the data and to augment professional judgement with insights that replicate what an army of the world’s best experts can do.

This is a leap forward from current data analytics and even predictive analytics, as these techniques are limited to reporting on what has happened before or predicting future “what if” scenarios. AI, and associated machine learning methodologies, analyze the data, make assumptions and learn how the data behaves by testing and reassessing without human intervention and at incalculable speed. This provides audit and forensics experts with insights and predictions at a far greater scale and depth than ever before.