Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
For American Express (Amex), using AI and machine learning (ML) to address credit card fraud is nothing new. The company has been using artificial intelligence to automate billions of fraud risk decisions for years, while hundreds of Amex data scientists work on AI and ML models related to fraud risk.
“It is certainly a key focus for us,” James Lee, VP of global fraud risk at Amex, told VentureBeat. “We’re totally vigilant to make sure that we defend against those risks.”
However, account login fraud is a particularly thorny challenge that is only rising in importance. With the advent of chip-pin cards and online one-time passwords, fraudsters are looking at more unconventional ways of committing credit card fraud.
Amex ML model pinpoints account login fraud
One way they do that is to log into a customer’s online account to change key demographic info, order replacement cards, get access to OTPs or disable spend/fraud alerts — and then make fraudulent transactions on the customer’s card. They may even access membership rewards currencies and try to redeem them for digital gift cards.
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
To sniff out login fraud, Amex recently developed an end-to-end ML modeling solution which, at an account login level, can predict if the login is from a genuine customer. Logins with high-risk scores are required to add incremental authentication, while low-risk logins get a seamless online experience. This ensures bad logins are captured in real time while good customers are minimally impacted.
Next-step-up authentication is high in friction for genuine customers, Lee explained. “There was a strong push from our leadership team to make sure that we evaluate the risk of the individual logging in, leveraging the vast amount of data and history we have on that customer’s activities,” he said.
Now, with the iteration of the ML model for real-time prediction of account login risk, fraud rates have been decreasing over time. “With the first iteration versus now, the model is stronger-performing than most other models in the marketplace provided by third-party vendors,” he said.
Stopping login fraudsters in real time
Abhinav Jain, VP of global fraud decision science at Amex, leads a 60-person fraud machine-learning team working globally for Amex on projects related to all kinds of fraud. He says building an ML model to address login fraud risk has been a key project goal over the last few years.
Traditionally, he explained, Amex developed machine learning models that analyze fraud risks at the point-of-sale transaction — when a customer is using a credit card in a store, for example.
But as login fraud activity ramped up with online takeovers and account hacking, Amex saw the need to prevent fraud at the login level, “so that we can stop the bad actors upfront and not wait for them to transact,” he explained.
The first challenge Jain’s team was able to solve was integrating logins into an ML platform which had trained the model on historical customer data. “Each login needs to get scored by the model in real time,” he said.
A second challenge was figuring out how to identify fraudulent logins. “When we build a transaction or point-of-sale model, we reach out to customers, or customers reach out to us, so we know which transactions are fraud or not,” he said. But with account login fraud, “it becomes tricky, because we don’t go back and ask customers.”
Instead, Amex had to develop a logic for the ML model to learn. It uses the customer’s past online login behavior to identify which logins are fraudulent, which are good and which are uncertain.
Amex ML model offers a feedback loop
“It’s really about that feedback loop,” said Lee, who explained that the machine learning model incorporates new information and determines whether certain signals and characteristics translate into false positives or are actually accurate predictions of future fraud behavior.
“There was always a rules-based structure to determine the low versus moderate versus high risk,” he said. But that was more of a static output, whereas the new ML model can assess all of the most recent information in real time and then factor that into the most recent performance as the model calibrates itself.
“That has allowed us to strengthen the hit rate for high-risk prediction,” he added. “It is what enables us to have the industry’s leading fraud reduction rates relative to any networks or competitor issuers in the marketplace.”
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.