By Ray Birch
PALO ALTO, Calif.—Artificial intelligence (AI) may someday make mobile banking one of the most secure means for consumers to manage their accounts, asserts one analyst.
Mobile devices bring with them many new data points that can verify a user’s identity. However, it will take AI to crunch all that information to effectively fight fraud and also make the consumers’ mobile banking experience simple and easy, said Rahul Pangam, CEO and cofounder of Simility, a fraud analytics and detection firm.
“When I log into my mobile device, there are 300 to 400 data points about my device that can be used to authenticate the user,” said Pangam. “For example, things like is my preorientation for viewing the screen horizontal or vertical, the gyroscope readings . . . those things and much more.”
Data regarding how a user typically interacts with the FI’s online banking solution can be used, as well.
“If I am a user and it always takes me eight seconds to go from the login screen to viewing my account history, and another 15 seconds to initiate a transaction, those are established data points that can be used to verify my identity,” said Pangam. “So there is information from the device financial institutions can leverage, and there is information from users themselves to authenticate.”
Ongoing Series
Pangam’s observations come as part of an ongoing CUToday.info series that examines how artificial intelligence, often referred to as “AI,” will manifest itself at CUs.
Pangam explained that these types of data points, by themselves, are not means to authenticate the user. They are effective, however, when several of the data points are analyzed together to determine what is normal user behavior or fraudulent behavior.
To assess that data effectively artificial intelligence is needed—and this approach is starting to be used by financial institutions today, he said.
“Unless you have machine learning you are simply not able to make this kind of analysis,” said Pangam. “Only artificial intelligence can crunch all of this data, do it quickly, understand it and make a proper analysis. A person here could be looking at 800 variables at once. You have to throw that data into machine learning.”
The speed at which AI moves is critical not only in detecting instances of fraud and stopping them, but also in not slowing down the user.
Pangam emphasized that as fraud detection becomes more complex, that the complexity cannot be seen or felt by the user.
“You want a frictionless experience,” he said, noting that the opposite, cumbersome user experience is multi-factor authentication, such as when a user logs in and then is sent a text message to verify identity.
“You do not want to ask the user to take extra steps,” Pangam said.
Self-Learning
Pangam said use of AI in the fraud fight, as he previously described, is still based on detecting fraud patterns based on previous user transaction and behavior history—therefore manual intervention is required. The next step is for the machines to self-learn, he said.
“The machines will self-learn as user behavior changes and as fraudsters adjust their tactics,” said Pangam.
As machines become self-learning can they stay a step ahead of crooks?
“I have been in this business for ten years, and I’d like to say that is possible,” said Pangam. “But that may be difficult. As the financial institutions get better at fighting fraud, so too do the criminals improve. They have access to the same technology, they have machine learning on their side, as well. I have always found that the fraudsters find a way, it’s their business. They have to find ways to up their game.”