The Featurespace platform and its adaptive machine learning models are now deployed throughout Verifone gateways, the company announced Wednesday (Nov. 4). The tool provides real-time assessments of each payment, helping merchants and payment facilitators prevent fraud, lost sales and chargebacks.
“Fraud is harder to predict than it was a decade ago,” Verifone said in a news release. “Patterns used to persist long enough for businesses to identify and respond to them. Today, AI gives more people access to advanced techniques, allowing fraud to evolve much faster.”
Verifone’s risk scoring framework examines more than 1,800 attributes of each transaction in real time, while keeping the request processing time under 100 milliseconds. With Visa’s Adaptive Behavioral Analytics, it develops a picture of what qualifies as “normal” for each merchant, card and account.
For example, a large purchase can be business as usual for one merchant but unusual for another. By looking past individual transactions, the framework finds patterns that occur only when signals are combined, and can learn from them as time goes by.
“Great payment security is invisible to the customer and unmistakable to the merchant,” said Fabian Ganster, head of global fraud and tooling at Verifone. ”Fraud is increasingly shaped by AI, so the defense has to be, too. By combining Visa’s adaptive intelligence with our payments expertise, we can stay ahead of threats that change every day and give merchants and partners payments and protection from one provider they can trust.”
PYMNTS examined Featurespace’s efforts to use customer purchase patterns to spot fraud in an interview earlier this year with Aman Cheema, vice president, head of global professional services, risk and security intelligence solutions at Visa.
“We model your behavior on your payment transactions,” Cheema said, adding that systems analyze purchasing patterns to recognize what a customer’s normal behavior looks like.
If a person who regularly purchases groceries from a familiar location suddenly makes a much larger transaction from an unfamiliar device or location, the system might flag that activity for review. However, Cheema stressed that the challenge comes not from collecting enough data but in identifying the signals that truly matter.
“There’s a lot of data out there. There’s also a lot of noise,” he said.
