Home » News » Inside FIS’ use of AI to weed out false-positive card fraud alerts

Inside FIS’ use of AI to weed out false-positive card fraud alerts

April 16, 2024 | Originally published on American Banker | By: Kate Fitzgerald

Banks rely on highly responsive fraud filters to alert them to potentially unauthorized transactions, but the best payment card fraud-defense tools also generate a lot of false alarms — and extra work for financial institutions, according to bank technology provider FIS.

As card-fraud risk has grown over the years, banks have been inundated with a rising number of notifications of potentially fraudulent card transactions, and sorting out which ones to respond to has become an increasing burden for FIS’ customers, said Eric Kraus, head of fraud services at FIS.

To complicate matters, perpetrators of card fraud are also changing their schemes so rapidly, as new types of scams proliferate, that banks need to constantly adjust their fraud filters to react appropriately.

“Over the last 18 months, we’ve seen a sharp rise in so-called credit card enumeration attacks, along with a spike in credit card account takeover and first-party card fraud, and while some of these card fraud types aren’t new, we needed better ways to analyze and respond to them,” Kraus said.

FIS set out to segment the types of fraud — and fraud alerts — that its card-issuing customers cope with so they can take appropriate action in the moment.

The result is an AI-powered risk model FIS recently developed with Stratyfy, a New York-based firm that uses AI to produce customized data insights that help FIS immediately spot fraud trends as they rise and fall so banks and consumers will receive more accurate warnings, Kraus said.

In tests runs over the last year, the model has also helped certain banks reduce the friction that consumers experience when their banks flag suspicious card transactions due to false positives, he said. 

“The new model has helped banks [prioritize their reactions to fraud alerts] so that one less consumer is impacted for every potential fraud incident, which may not sound like a lot, but on a macro scale with thousands of fraud alerts generated every day at banks, it’s a big deal,” Kraus said.

Stratyfy applied a proprietary “interpretable machine learning” AI approach to create FIS’ new fraud model, said Laura Kornhauser, Stratyfy’s CEO, who co-founded the 25-person firm in 2016. 

“With interpretable machine learning, we build a model with full visibility into how a fraud-detection model is working, so we can go inside and actively update the model to reflect changes in the fraud risk environment,” Kornhauser said.

The process doesn’t necessarily happen in real time, but Stratyfy can adjust FIS’ fraud model as often as needed — currently at least once a week — as different fraud schemes evolve and change, she said.

“Fraudsters have gotten smarter and they have more technology at their fingertips, so they’re always trying out new scams at relatively low cost to themselves, trying to find new vulnerabilities,” she said, noting that interpretable machine learning helps align the model’s fraud rules with the latest trends.

Banks using FIS’ fraud model now receive a more accurate stream of notifications of emerging types of possible fraud, while it also segments key categories where card fraud is spiking higher at any given moment, according to Kraus.

One analyst said the ability to segment different types of card fraud could be a plus for banks that have been applying more broad-spectrum fraud filters in a rapidly changing risk environment. 

“Many of the machine-learning models for card fraud detection are general-purpose models built using fraudulent transactions across a large number of financial institutions and consumers,” said David Mattei, a strategic advisor at Datos Insights’ fraud and AML practice. 

A predictive fraud model that is targeted at a specific segment of a card base will likely be better at finding nuanced instances of fraud than a general-purpose ML model that hasn’t been trained to detect a certain strain of fraud, Mattei said.

“By adding niche ML models focused on specific fraud types or card types, FIS might improve the base performance of its card fraud solution,” he said.

FIS said that in recent months, Stratyfy’s AI technology improved the bank technology firm’s accuracy in identifying and preventing fraudulent transactions by 51%, and it also helped FIS spot rising trends in small-business credit card fraud. 

“When fraudsters started going after more small-business credit card accounts, the model segmented that type of fraud so we were able to see patterns and respond to it with appropriate alerts,” Kraus said.