Long ago, financial institutions relied on human discernment to calibrate systems that help identify potentially risky transactions and customers. Now, Google Cloud wants them to give its AI technology increased control over this process.
Wednesday, Alphabet’s cloud business announced the launch of a new anti-money-laundering product powered by artificial intelligence. The company’s technology utilizes machine learning to assist customers in the financial industry in complying with regulations requiring them to screen for and report potentially suspicious activity.
Google Cloud intends to differentiate itself by omitting the rules-based programming that is typically an integral part of establishing and maintaining an anti-money-laundering surveillance program—a design decision that goes against the prevailing approach to such tools and could be met with skepticism from certain industry segments.
The product, an application programming interface termed Anti Money Laundering AI, has already attracted notable clients, including HSBC of London, Banco Bradesco of Brazil, and Lunar, a digital bank based in Denmark.
As a result of the success of the generative AI app ChatGPT, leading U.S. tech companies are demonstrating their artificial intelligence capabilities, and many corporations are racing to integrate this technology across a wide variety of businesses and industries.
Some financial institutions have relied for years on more conventional forms of artificial intelligence to assist them in sorting through the billions of daily transactions they facilitate. Typically, the process begins with a series of human judgment calls, followed by the application of machine learning technology to create a system that enables banks to identify and evaluate activity that may need to be reported to regulators for further investigation.
Google Cloud’s decision to eliminate rules-based inputs for its surveillance tool is a bet on AI’s ability to address a problem that has plagued the financial industry for years.
Depending on how they are calibrated, the anti-money-laundering instruments of a financial institution may flag too little or too much activity. Too few alerts can result in inquiries from regulators, or worse. Too many can overwhelm a bank’s compliance team, which is responsible for evaluating each strike and determining whether to file a report with regulators.
Google Cloud executives contend that manually input rules drive up those numbers. A user could, for instance, instruct the program to flag customers who deposit more than $10,000 or transmit multiple identical transactions to more than 10 accounts.
Consequently, the number of system-generated alerts that turn out to be false leads, or “false positives,” tends to be high. According to research conducted by Thomson Reuters Regulatory Intelligence, up to 95% of false positives are generated by such systems.
Users will not be able to input standards with Google Cloud’s product, but they will be able to customize the tool using their own risk indicators or typologies, according to executives.
Google Cloud claims its technology reduced the number of alerts HSBC received by as much as 60 percent while increasing their accuracy by utilizing an AI-first strategy. HSBC’s “true positives” increased by two to four times, according to Google-cited data.
Jennifer Shasky Calvery, group head of financial crime risk and compliance at HSBC and former chief U.S. anti-money-laundering official, stated that the technology developed by Google Cloud represented a “fundamental paradigm shift in how we detect unusual activity in our customers’ accounts.”
For many financial institutions, convincing them to cede control to a machine-learning model may be challenging. For starters, regulators expect institutions to be able to articulate the reasoning behind the design of their compliance program, including how they calibrated their alert systems. The standard line of reasoning among banks and their regulators is that such systems should be tailored to the institution’s risk profile and specific characteristics.
And while compliance experts assert that machine-learning-driven anti-money-laundering tools have improved over time, their limitations have led some in the industry to doubt their ability to replace a human’s ability to identify where the actual risks reside.
Sarah Beth Felix, a consultant who assists banks in vetting and calibrating their anti-money-laundering tools, stated, “There is so much contextual information that is not taken into account by these systems.” “AI is only as good as the humans who train it.”
Executives from Google Cloud stated that they hope to allay these concerns through a combination of improved results and their product’s “explainability.”
Instead of concentrating on providing transaction alerts, the company’s product utilizes a variety of data to identify retail and commercial customers who pose a high risk. Zac Maufe, global chief of regulated industries solutions at Google Cloud, stated that whenever the product flags a particular customer, it also provides information about the transactions and contextual factors that led to the high-risk score.
“We spent a great deal of time ensuring that the language the model provided to analysts accurately reflected their vocabulary,” said Maufe. “It’s not just ‘give them the answer,’ but also’show them the homework.'”
According to Calvery, the acceptance of HSBC’s new approach by regulators was achieved through testing and validation of the new instrument.
“As soon as we noticed that [Google Anti-Money-Laundering AI] was discovering more and doing so with significantly less noise, we began to ask, ‘What’s not the case for using it?'” she said.
