Researchers have developed a new artificial intelligence (AI) system to detect accounting fraud within individual companies and across supply chains and industries.

The machine learning technique FraudGCN analyzes patterns in financial data and corporate relationships to identify and predict fraudulent activities. It uses graph theory and machine learning to examine the web of relationships between firms, their auditors and industry peers.

“It’s an unending, mathematical arms race between the authorities and the fraudsters,” Chenxu Wang, lead author of the paper and an associate professor with the School of Software Engineering and the Key Lab of Intelligent Networks and Network Security at Xi’an Jiaotong University, said in a news release.

The development comes as financial markets grapple with the impact of accounting fraud. A PYMNTS report reveals that 62% of financial institutions with over $5 billion in assets report increased financial crimes, exposing growing vulnerabilities in the U.S. banking sector. As methods for perpetrating fraud evolve, including criminals’ potential use of AI, there is ongoing interest in developing more effective detection methods.

Current Methods of Financial Fraud Detection

Traditional fraud detection methods often rely on audits, which can be labor intensive and may struggle to differentiate between genuine business success and manipulated figures. These hurdles means that many firms may go unchecked for extended periods.

Paul Wnek, founder, CEO and principal solutions architect at ExpandAP, told PYMMTS there are many common types of fraud in businesses: “Invoice fraud, such as fictitious invoices for goods or services that were never delivered or legitimate invoices that are altered to divert funds. Vendor fraud, such as setting up fake vendors to receive payments for nonexistent goods or services or kickbacks to award vendors or employees for approving contracts or invoices. Payment fraud, which can occur when fraudsters gain access to payment systems or manipulate approval processes to authorize payments fraudulently.”

These schemes can be challenging to detect using conventional methods.

“What is needed is an effective and accurate algorithm to automatically identify accounting fraud, and leave the days of random auditing behind,” said Mengqin Wang, another researcher involved in the FraudGCN project, according to the release.

FraudGCN attempts to address this by constructing multi-relational graphs representing company connections. This allows the system to analyze patterns across corporate networks.

When tested on data from Chinese listed companies, the researchers found that FraudGCN outperformed current approaches by a margin of 3.15% to 3.86%.

However, the practical implications of these improvements in fraud detection are still unclear.

The Role of AI in Fraud Detection and Perpetration

As AI’s role in fraud detection expands, experts note its potential for both detecting and aiding fraud. Joe Stephenson, director of digital intelligence at Intertel, discusses AI’s dual nature in this context.

“In the insurance industry, we are in the business of selling, and because of this, often overlook potential implications emerging technologies like artificial intelligence may have on claims,” Stephenson told PYMNTS. “While AI is great for underwriting, we are also seeing criminals leverage ChatGPT and AI to advance fraudulent activity, whether it be through the development of synthetic IDs or metadata.”

This introduces new challenges, as Stephenson explained: “Metadata isn’t traditional, and the use of social media makes it easy for any person to exaggerate claims or organize criminal groups.”

However, AI can also be leveraged to parse through large volumes of data.

“Advanced algorithms can scan and analyze social media activity, identifying patterns and anomalies that might go unnoticed by human investigators,” Stephenson said.

The “Financial Fraud Prevention Playbook” by PYMNTS examines how financial institutions can leverage advanced technologies like behavioral analytics and machine learning to combat digital-age fraud tactics, including AI-powered schemes, malicious bots and synthetic identities that evade traditional security measures.

Automation in Fraud Prevention

Alongside AI detection tools, the industry also implements automation in financial processes as a preventive measure.

“Automated accounting systems built with best practice security measures offer built-in fraud detection capabilities, such as anomaly detection and invoice matching algorithms,” Wnek said. “The best platforms are one-stop shops for all AP tasks, leading to a decrease in the number of systems between which data needs to pass through.”

This shift toward automation is an ongoing trend in the finance industry.

“Though traditionally change-averse, accounting teams have started to recognize the value of automation in improving the efficiency and accuracy of processes like accounts payable (AP) and accounts receivable (AR),” Wnek said. “AP automation cuts that line item completely, reducing costs anywhere from 40-95%.”

Additionally, by reducing manual interventions in financial processes, automation may decrease opportunities for certain types of fraud.

However, the adoption of these technologies has its challenges.

“The two biggest barriers to digitizing AP and AR adoption are cost and complexity,” Wnek said. “But these barriers are easily alleviated when businesses consider the return on their investment compared to outsourcing customer support and finance processing.”