The transition to the EMV chip in North America has been both a boon and a bit of a headache when it comes to payments security.
On the one hand, Worldpay’s VP of Data Products Nicole Jass noted, EMV has done an awful lot to slam the door shut on card-present fraud. According to Visa, counterfeit fraud at U.S. chip-enabled merchants was down 66 percent in June of 2017 compared to June of 2015.
That’s great news, but sadly it’s not the whole story, as fraudsters worldwide didn’t shrug their shoulders and give up. They just found a window to open, and that window has been card-not-present (CNP) fraud. According to Javelin, CNP fraud is now 81 percent more likely to occur than point-of-sale (POS) fraud — the greatest gap the company has ever observed.
That shift, Jass told PYMNTS, made Worldpay consider what they were doing to fight the building tsunami of CNP fraud, which she said brought them back to data — specifically their data and the massive troves of information at their disposal.
“In 2017, we started to really think about how we could unlock the power of the data to really help our merchants fight that growing fraud problem,” she said.
To that end, Jass noted, Worldpay recently partnered with machine learning company Featurespace to build a better fraud-fighting engine.
Fighting Fraud by Finding the Good Instead of the Bad
The challenge of locking out fraud, Jass and Featurespace CTO and Co-Founder David Excell explained, isn’t just around locking out fraudsters. Making a system mostly inaccessible isn’t that hard. Even if it successfully locked out 100 percent of fraudsters and made sure another “bad” transaction never happened again, it still probably wouldn’t be a very good system.
The reason is that all those restrictions would most likely net false-positive and send good transactions away — which is not what the merchants Worldpay works with want, although they’re certainly worried about CNP fraud. In fact, she noted, as larger merchants move toward online and omnichannel sales, they increasingly want solutions for locking out digital fraudsters.
Merchants are in the business of facilitating transactions — not blocking them — which means depressing their revenue by pushing good customers away is not a price they’re willing to pay to prevent fraud. They want a smooth sales process with as few roadblocks as possible built in.
For Worldpay and Featurespace, constructing bigger walls meant “protecting” merchants from actual commerce. Instead, the goal was to build smarter walls that know who to let through and who to lock out.
“For us, the goal … is about modeling what a good transaction looks like because the merchant cares far more about getting the good transactions through,” Jass said. “The real focus on the product we are building together is making sure we have a good handle on what those good transactions look like.”
This is where the power of a large data set comes in, Excell noted; it gives the learning engine perspective not just into how the customer is acting or buying at this specific merchant during this transaction, but also how this customer has acted across a variety of merchants across the vast Worldpay ecosystem.
“The power behind that is machine learning. What we’re doing within that is building up a picture of what each individual cardholder looks like, what each merchant looks like … keeping an eye on the types of transactions we see going across the channels,” he said.
Beyond Red Light, Green Light
Creating better digital security for CNP transactions ultimately means understanding that digital security won’t always be binary. There will be many transactions that are absolutely in the clear — “green lights.” Others, Jass said, will just look wrong for a variety of reasons — “red lights.”
The challenge is in the “yellow lights” — when it isn’t clearly fraud but some details are unusual. Perhaps the customer is new or is making a larger purchase than normal. A smart system specializes in those yellow light reviews: “The goal is for a system to be able to tap into additional data sources,” Jass explained.
In Worldpay’s case, Excell noted it all comes down to scale and the fact that the entire ecosystem gives the learning machine understanding into consumer behavior across Worldpay’s merchants — which translates into a clearer picture of what their “total retail behavior is.”
However, it will still be a while before that snapshot is available. The system the two firms are building, Jass said, won’t be available until late this year or early next year and will focus on larger omnicommerce retailers — though merchants big and small are the eventual target.
Their project will be ongoing, mostly because it has to be. Fraudsters aren’t going away; in fact, by the numbers, they’re only getting more aggressive and working harder to find ways to crack into digital transactions.
That they must be stopped is not a question. The challenge is making those stops consistent — without shutting down commerce for honest consumers who just want to buy something.