TSYS’ Fraud Prevention Gets Smarter With New Partnership

Machines can’t do everything better than people (not yet, anyway), but one area that might prove the contrary is fraud prevention. Andrew Mathieson, group executive of product strategies at TSYS, spoke with Karen Webster about how his company’s partnership with behavioral analytics firm Featurespace is poised to take the efficiency of machine-learning systems to the next level.

Today (May 19), TSYS announced that it is partnering with behavioral analytics firm Featurespace to provide its clients with machine-learning, real-time fraud detection and prevention capabilities.

Karen Webster connected with Andrew Mathieson, Group Executive of Product Strategies at TSYS, prior to the announcement, for a discussion in which Mathieson shared his perspective on what drove the collaboration with Featurespace, the details of the expanded fraud-combating system and the ongoing battle therein.

“If you look at the processing space holistically,” says Mathieson, “with all of the innovations that have been made, more and more activities across the entire cardholder life cycle are coming with the expectation for things to be instantaneous. That’s going to put pressure on our clients to be positioned to make the best-informed decision in essentially less than a second.”

In its search for a partner to help make such an offering a reality, TSYS was impressed by Featurespace because of its approach of applying machine learning to a variety of business use cases for real-time decision capabilities.

Referring to the number of fraud detection models in existence that rely on population odds to make determinations, Mathieson notes that when a person is analyzing transactions using such a system, were he, for example, to see activity occurring on the same credit card in Oslo one day, Tel Aviv on another and San Jose shortly thereafter, the human analyst would likely be naturally inclined to view the behavior as suspicious and flag it as fraud.

However, Mathieson points out, that customer behavior would align with a “perfectly reasonable use case for a venture capital firm.”

“Normal for the person,” in that sense, while “abnormal for the population at large.”

“The machine-learning capabilities allow us to have insights into what’s anomalous behavior for the individual,” Mathieson explains to Webster. “The unit of analysis is the person: what is normal for them and what is out of band. Taking that approach, we can go beyond just population odds,” although he does express TSYS’ belief that there exists the potential application of using both.

While the machine-learning fraud detection and prevention system that Mathieson describes is intelligent enough to recognize an individual customer conducting a transaction far from her home area in a place that she does visit frequently as legitimate, even more so, it would be smart enough to know that “even if it’s a place she’s never gone to” — based on the system’s understanding that the person is generally a regular traveler — “it’s not unusual.”

To reach that level of intuitive understanding, of course, the fraud detection system needs data.

“It gets smart,” says Mathieson, “by looking at whatever data is available — what’s available sub-second, what’s present now that can be decisioned upon – as well as on feedback loops created by what happens subsequently.”

“It’s in automating those cycles where you get the benefit,” he adds, “and it’s where learning occurs.”

Reducing the number of decisions that are normally needed just to reach the point of determining whether or not activity is suspicious, thus gaining speed in the process, explains Mathieson, can not only help beat fraudsters to the punch in their efforts to circumvent otherwise traditional guards, it can also save companies money.

Whether it’s in the fraud space or somewhere else, “wherever you can make a better decision and preclude the need for a human to get involved and think further about it, there’s likely going to be efficiency savings,” observes Mathieson, citing — for an example outside of fraud — how automation has made the process of credit underwriting more effective, both in the process itself and in its cost.

Such efficiency gains in using a machine-learning fraud prevention system, he notes, “are both financial [for the company] and in the experience for the consumer,” in that the latter doesn’t have to invest time in disputing transactions.

While the fraud prevention capabilities that TSYS has gained from its partnership with Featurespace can already be applied in any type of transaction, physical or digital, Mathieson adds that Featurespace’s ARIC engine machine-learning software platform presents opportunities for use cases beyond fraud.

“From a TSYS standpoint,” he tells Webster, “we’re open to innovation wherever it occurs,” particularly given that a prioritization of speed — be it in fraud detection, eCommerce delivery, check processing or any number of other areas — is “the future for all of us in the payments space.”