Overcoming Human Error in Payment Fraud: Can AI Help?

Human error is a major contributor to payments fraud, but only about 5% of organizations have fully automated their payment processes to reduce mistakes, according to a recent survey. Experts say artificial intelligence-enabled automation will help reduce risks, but the benefits of this technology are still a distant reality.

See Also: OnDemand | Everything You Can Do to Fight Social Engineering and Phishing

A recent Trustmi survey of 516 organizations found that half of those companies have been victimized by business email compromise and other forms of payment fraud related from human error. The survey found that 41% of respondents automate some aspects of their workflow, but nearly 27% still rely on manual processes. Despite incorporating fraud prevention tools in their technology stack, 14% of organizations do not use them, leaving their payment processes vulnerable to attacks.

With global fraud costs reaching a staggering $485 billion last year, many organizations are looking to AI to defend their enterprises. In fact, 70% of respondents in a NASAQ survey expect their organizations to increase spending on AI or machine learning in the next couple of years, saying they plan to use AI tools to reduce the complexity of daily processes, such as examining beneficial ownership information, enabling financial crime analyst co-pilots, for alert explanation and narrative generation, and for analyzing customer risk profiles.

“As compliance pressure mounts and financial crime evolves, organizations that do not plan to increase spending on AI will need to identify alternative means to strengthen their financial crime management programs,” the NASDAQ report said.

The Challenges of AI Adoption

But the potential benefits of AI-based automation are limited by the costs, time and resources needed for integration with complex payment ecosystems. Overhauling legacy systems can be challenging, and most organizations lack the technical skillsets to take advantage of the new technology, said Shai Gabay, CEO of Trustmi.

The cost of hiring the necessary IT support to lead the digital transformation, invest in new software and train employees are major hurdles for most companies. Other major concerns include integration issues with existing ERP systems and accounting software, Gabay said.

Central banks can raise their game with AI, but the technology carries risks as well, with its tendency to hallucinate and vulnerability to hacking.

In fact, the International Monetary Fund warned that the evolving nature of AI and its applications in finance means that neither developers nor regulators fully understand its strengths and weaknesses. “Hence, there may be many unexpected pitfalls that are yet to materialize, and countries will need to strengthen their monitoring and prudential oversight,” according to the IMF.

One of the main obstacles facing banks is the complexity of their existing environments. B2B payments processing includes systems for vendor onboarding and management, invoice processing and legacy system management. All these processes include a myriad of people and the use multiple insecure communication methods. With so many moving parts, bad actors can exploit numerous weak links in payment processes.

For example, large businesses on average process more than 20,000 invoices per month. Manually scoping for fraud is not just challenging, but an impossible task at scale. Also, finance teams often work in siloes, making it difficult for companies to achieve end-to-end visibility over payment processes. Many organizations rely on outdated legacy systems and manual processes that lack the ability to provide real-time data and comprehensive reporting, making timely fraud detection difficult.

Scammers usually target accounts payable departments, which processes payments to suppliers and vendors. They typically pose as an existing supplier and send fraudulent invoices to an organization or even digitally gain access to a company’s AP processes to authorize large payments, said Infosys. “Usually, by the time such a fraud is detected, it’s already too late; the money has been siphoned off, and the perpetrators are long gone,” Infosys said.

Human error also enables age-old scams that use social engineering and phishing to trick employees into revealing sensitive information or authorizing fraudulent transactions. Sloppy data management, such as sending information to the wrong recipient, can lead to data leaks and financial fraud.

How AI Can Help

Accounts payable automation solutions can flag minute discrepancies in invoices, such as a new address or new bank account details, that manual process might miss. Alerts can prompt companies to follow up with their vendors to verify the legitimacy of invoices before processing payments. “For example, if it has been the norm for an organization to receive an average of six invoices a month from a supplier, and the number suddenly spikes, this inconsistency will be immediately flagged. Modern automation tools employ AI and machine learning algorithms, which is impossible for fraudsters to bypass and deceive and are an invaluable tool in AP fraud detection,” Infosys said.

Businesses see the potential for AI to reduce fraud losses in B2B payments. Companies can use AI to examine historical data to identify patterns, detect anomalies and automate routine tasks such as data entry and calculations. They can use crowdsourced data from vendors to streamline processes and enhance trust. Technologies that provide end-to-end visibility of the entire B2B payment ecosystem offer a comprehensive view, helping detect and prevent issues arising from human errors.

Some organizations have launched AI-based initiatives to fight fraud, but the it’s too soon to see results.

For examples, the member-owned payment cooperative Swift, which has a global network of over 11,500 institutions, announced two AI-based experiments earlier this year.

One initiative hopes to boost Swift’s existing tool for helping financial institutions detect anomalies indicative of fraud. The other project will train an AI model on historical patterns of activity on the Swift network – including the banks’ live traffic data – to create a more accurate picture of potential fraud activity, allowing financial institutions to flag anomalous payments before they are executed.