Solving Digital ID Verification’s ‘Existential Crisis’

There’s a tug of war when it comes to digital ID verification — a tug of war between the customer experience and reducing fraud. Socure Co-founder Johnny Ayers tells PYMNTS that AI-based identity risk scoring can balance those demands by improving acceptance rates, cutting down on manual processes, slashing fraud and reducing false-positives.

Call it the existential crisis of digital identity (ID) verification. It’s the trade-off between financial institutions (FIs) easing the consumer’s journey across banking and transactions done online, and the need to prevent fraud during new customer onboarding.

To that end, Socure and Alloy said last week that they have partnered to offer financial firms a joint solution that focuses on risk scoring. The combined efforts revolve around Socure’s predictive analytics platform and Alloy’s decisioning technology. Among the early users has been Radius Bank, which Socure and Alloy said has used the authentication offering to cut fraud in half and boost new account conversions by 30 percent.

In an interview with PYMNTS, Socure Co-founder and SVP Johnny Ayers explained the rationale and larger trends underpinning the partnership. “A major use case that we are solving for,” he said, “is digital account openings.” Those are, of course, the accounts that are created when a customer presents their identity to a new institution via the web in hopes of taking out credit cards, obtaining loans or opening bank accounts — all the while looking to make transactions online.

“There’s an existential problem,” he said, in “how do you balance an easy customer experience that is very passive and fast, and minimizing or reducing the amount of fraud risk to which you are exposed. … There is always a kind of teeter totter balancing act between the two” for FIs, he said.

Given the backdrop of the more than 1,200 data breaches seen in 2017 and the rise in commerce done by mobile means, there is a ton of stolen data out there, wielded by bad actors who in turn seek to commit fraud. That same data, now floating around on the dark web and elsewhere, had been traditionally used by FIs when crafting modeling techniques to onboard customers. There’s an added challenge in the fact that, for the younger generation of would-be customers, there just isn’t the same “footprint” that might be gleaned from traditional data sources, said Ayers.

“And so,” Ayers said, “as you start to have to look at larger amounts of information” that spans “not five or 10 data points, but hundreds or even thousands of data points, the only way that you can do that is by applying advanced predictive analytics and machine learning to read over that much data and come to conclusive decisions.”

An optimized, machine learning and artificial intelligence (AI)-driven solution, he said, will improve onboarding activities.

In talking about accuracy in those machine-driven models, he said one key aim is to reduce the number of false-positives. Many incumbent players in this market have models in place with false-positive to true-positive capture ratio of four-to-one, six-to-one or sometimes higher — meaning incidences of false-positives far outweigh true-positives when accounts are classified as suspicious. A major way to correct that imbalance, Ayers said, is through feature engineering and algorithms — as a result, the Socure/Alloy solution has seen ratios of one-to-one or better.

When asked about how machine learning can help with real-time digital ID verification, the benefit lies in teaching machines to do repeatable tasks, he explained. In classifying identity fraud, machines are always going to outperform humans. “The idea for us [is] basically training a machine to test new data sources that we assign values to — things like accuracy and coverage and predictions” — and assigning value to new data, he stated.

In the end, the machine can separate signal from noise, and it is the clean signals that are used to train the models themselves to help predict risk. “So, when a customer is opening an account as we are provided information” Ayers said, “that information is exploded into thousands of data points. Those thousands of data points are fed into predictors. Those predictors are then fed into trained models.” The models then return scores to the Socure/Alloy FI customers. Machine learning can also uncover patterns tied to synthetic IDs, third-party fraud or chargeback patterns.

The banks then benefit from a boost in accuracy, as well as a reduction in both fraud and manual review. Ayers stated that digital accounts can be opened in minutes as risk is calculated in milliseconds.

“The end result,” he said, for FIs and, as a byproduct of more accurate models and more accurate risk scoring in real time, “is saying yes to more good people because your models are more accurately predicting good people.”