SAN FRANCISCO — JPMorgan Chase is investing in artificial intelligence to combat fraud and financial crimes as the $2.5 trillion bank looks to increase its precision and capture rate.
“There’s really a couple things that we’re laser-focused on,” JPMorgan Payments Global Head of Trust & Safety & Payments CDO Ryan Schmiedl said Wednesday at Finovate Spring 2023 in San Francisco.
“One is capture rate and precision. We’re looking for ways in which I can capture a larger percentage of these malicious events and [do] so with better accuracy [and] less false positives,” Schmiedl said.
Earlier this week, the bank hosted its 2023 Investor Day and broke out its latest AI research efforts.
The bank is committed to delivering $1.5 billion in business value from AI by the end of 2023, Lori Beer, chief information officer at JPMorgan Chase, said during her presentation. “This value is driven by more than 300 AI use cases in product today for risk, prospecting, marketing, customer experience and fraud prevention. … These gains are tied to investments and actions we’ve taken in the way we deliver software and our modernization efforts.”
JPMorgan employs “supervised tightening techniques” to train its AI models and reduce the volume of false positives, Schmiedl said.
“If we have something that looks suspicious, we have senior investigators going through and confirming this [and] getting us good strong labels,” Schmiedl said. “We can then use these labels to train to do things such as help our screen engine to become more accurate and precise by automatically dispositioning a false positive as opposed to having to reinvestigate these types of things.”
Learning the context
JPMorgan plans to continue to invest in large language models to help AI understand the context in unstructured text, he said.
“There’s a lot of opportunity in this space to find unsupervised techniques — looking at context,” Schmiedl said. “In the fraud or financial crimes space, specifically sanctions, most of the time you get a list and very little context to what you’re looking for.”
Old systems were unable to identify differences in data points, he said. “These things basically lead to a lot of unnecessary friction.”
Editor’s note: This article first appeared on Auto Finance News, a sister publication to Bank Automation News.
