Visa Is Rolling Out New Scam Detection Tech. It’s Already Prevented $460 Million in Losses in the U.K. Alone
The new fraud detection system from Visa is ‘invisible to the customer.’ Here’s what it does.
James Mirfin, Visa’s head of risk and security, has spent the past few months helping his father, who has been in and out of the hospital in the U.K., move money around. “I know how hard it is now to add a new payee to your bank account,” he told me. “It’s like, have you met this person? Did they call you? You go through all these 15 questions, and eventually if you get them right, you can press pay.”
Fraudsters also know what those questions are. Coaching a victim through the answers is a standard part of running a scam, which results in those safety checks slowing down honest customers while the people they were built to catch breeze past them.
To fix that experience, Visa announced an upgraded
To fix that experience, Visa announced an upgraded version of a fraud prevention system called A2A Protect today. With this system in place, a customer starts a transfer, and before the money moves, the bank sends the details to Visa through an API—a direct machine-to-machine link that returns an answer in milliseconds. Visa sends back a score from 1 to 99 that grades the likelihood that the payment is a scam, with a plain reason attached, indicating the type of scam it thinks it is looking at. Visa never touches the money, and each bank decides what, if anything, needs to happen next.
Two systems sit behind that fraud score. Visa completed its acquisition of Featurespace, a Cambridge University spinout, in December 2024, buying behavioral analytics that profile how a single account normally behaves. Visa’s own contribution is pattern recognition at network scale. “You get stuff that looks strange for [the individual user] or strange from a network level or both, and that dictates the score,” Mirfin explained.
Weekly roundup of the latest in tech news
An Inc.com Featured Presentation
What the number replaces
Using this new system, one European bank saw a 51 percent uplift in fraud caught when using the tool compared to when not using it. They also saw a 42 percent reduction in false positives for fraud flags compared to when they weren’t using the tool, worth 5.2 million euros across five months. A consortium pilot in Brazil caught 82 percent of fraud by value, the measure that tracks money at risk rather than the count of flagged transactions, identifying $90 million. Visa says a U.K. trial has prevented up to $460 million in losses to date.
The most interesting result came from one bank
But the most interesting result came from one bank in that U.K. pilot, which gave Visa more data than any other participating bank. It mapped its account numbers to card numbers, letting Visa see debit card history sitting behind each bank account—though only for Visa-issued debit cards. Detection by value rose roughly 30 percent above what A2A Protect was already catching on transfer data alone, because the extra layer surfaced a scheme the transfer data would have missed. For example, international students were selling control of their U.K. accounts before leaving the country. The card history showed ordinary student spending—coffees and books and the underground—then the account went dormant for a year or two before reappearing as a destination for transfers from people who had been scammed.
Fewer false positives mean fewer genuine payments held up and fewer analysts working on fraud alerts that go nowhere. Mirfin sees a bigger end result than that. Because the score lands before the bank has to challenge anyone, it can stop interrogating everybody and reserve the questions for what the model has already flagged.
Source: www.inc.com
