Visa May Use Doppelgangers to Fight Payments Fraud

Visa wants to know from the jump if you’re a big spender (or a hacker). 

The company wants to patent a method for generating “behavior profiles for new entities.” Visa’s filing details a system which would understand card holder behavior without having significant activity data. 

“When a new entity is added to a conventional behavior-monitored system, a time period, which is typically at least several days, is needed for the new entity to create a behavior history,” Visa noted. “This time period … can provide a window of opportunity for attackers to target the system.” 

Visa’s method takes “feature data” from a new customer’s profile, such as age, gender, credit rating, region or occupation, and creates a “doppelganger behavior profile” based on pre-existing customers that fall into similar categories. To create these doppelgangers, a machine learning model assigns the pre-existing customers each a similarity score to the new customer, and several of the profiles are combined based on those scores. 

Once that profile is assigned to a new customer’s account and they are monitored. If fraud, anomaly or any other “malicious intent” is detected in the customer’s spending based on the predicted behaviors in the doppelganger profile, the account is automatically suspended. 

Fraud detection isn’t the only use case Visa points out for this tech. The company noted that building these mimic profiles could be used to “generate a baseline for modeling behavior” of new customers, so that they can be “more quickly targeted with accurate messages and incentives.”

It makes sense that Visa is looking into AI-based user tracking capabilities that can be used in multiple ways. A primary strength of machine learning models is discovering patterns in data and making predictions based on them. And, as this patent exemplifies, that can be useful for both tracking down cyber criminals and finding out how you like to spend your money. 

This also isn’t the first time we’ve seen Visa want to analyze spending habits using AI. The company previously sought to patent tech that uses deep learning to analyze “financial device usage” to try and reel customers back into using their Visa as their primary card. 

With more than 4 billion cards worldwide and billions of dollars in transactions handled annually, using AI is vital for a company like Visa to find needles in haystacks. But in terms of preventing fraud, this patent doesn’t go into detail on exactly how it decides what’s deemed to be suspicious activity and what can fly under the radar, said Ali Allage, CEO of BlueSteel Cybersecurity. “Everyone is focused on behavior,” said Allage. “All the AI stuff is starting to sound the same, and I didn’t see anything (in the patent) that was an outlier.” 

Financial institutions have been trying to get smarter about the “predictive aspects” of spending and borrowing, said Allage, often using AI to do so. But AI is only as good as the data it’s trained on and the developer that built it. If a model like this starts to lump together demographics of customers based on those biases, they could be targeted or unfairly treated by fraud tracking systems. 

“Could someone be labeled as a highly risky individual financially, if they’re borrowing a whole ton in the first 30 days because they’re renovating their house? And is that a fair assessment?” he said. “That’s the part that’s not really clear to me.” 

However, these institutions are likely to err on the side of caution when it comes to risk-profiling technology, he said, even if it does lead to false positives. Currently, a lot of these firms take reactive measures when dealing with fraud, rather than predictive ones. “I would almost bet that they’ll make it more cautious on the risk-labeling side,” Allage said.