Data Drivers

Digital Breadcrumb By Digital Breadcrumb: The Trail To Verified IDs

The rise of the machines?

Not the latest Transformers or Terminator retread. The machines that we speak of here are the ones that act as the velvet rope, the bouncers if you will, of the financial industry. The ones who use data gleaned from thousands of sources to make sure that you are who you say you are when you want to transact.

In an interview with PYMNTS’ Karen Webster and as part of the latest Data Drivers, Johnny Ayers, co-founder and senior vice president of Socure, said that the old static models deployed by financial institutions are no longer robust enough to suss out who’s who – and that a sea change, rendered in bits and bytes, is afoot.


The genesis of Socure, then: “We really started from the idea that all of us as consumers lead very digital lives,” said Ayers. “We email, we text, we tweet. We leave large digital breadcrumbs about ourselves across our natural daily lives.

“The thought was that we’d harness the power of digital consumer identity and combine it with traditional offline data and – as opposed to using rules – use machine learning, [so that] we can actually develop the most robust and comprehensive identity verification and predictive solution.”

The goal has been to learn patterns and behaviors in a world where we, as individuals and consumers, are anything but static.

Previously, credit bureaus attempted to solve the problem of verification, said the executive, with an emphasis on traditional credit application information and DMV data points – and this, of course, was based on Social Security numbers. No one was using machine learning or AI, relying instead on what are known as “if, then” rules written around that data, said Ayers.

The shift cometh, however: “Look at the consumer across their entire identity,” said Ayers, with the idea of feeding the disparate data points tied to that examination into a machine that could tie in with hundreds or even thousands of variables in a way that humans could never tackle manually.

Data Point Number One: 60 Percent Fraud Reduction

This is the percentage of fraud reduction in the customers who are using the approach in the top 3 percent of their fraud models. Said Ayers: “In any fraud prevention model, you want to push as many as the bad actors into the top couple of percentiles. In a perfect world, you would capture 100 percent of fraud in the top 1 percent of a model.”

In this case, he elaborated, the company is able to capture 60 percent of the firm’s customers’ uncut fraud in just the top 3 percent. “For context, many of the legacy solutions are able to capture 30 percent of fraud in a 5 percent review rate bucket. So we are able to, in a much smaller population, capture multiples higher in terms of fraud.” That 60 percent fraud reduction rate would prove especially attractive to credit card issuers who are losing hundreds of millions of dollars annually to fraud – without changing operational or manual review processes.

“That’s a massive ROI,” said Ayers, to the tune of $50 million to $100 million in fraud reduction for some tier-one Socure clients.

To capture that efficiency, he said, “we have abstracted away from the kind of raw data assets to create a Socure proprietary variable base, or predictor base. We used a variety of unsupervised machine learning and clustering algorithms, a variety of semantic ontology and inferencing techniques to create a base [from which] we can then build machine learning models.”

The only human intervention needed, said Ayers, “is for the customer to tell us what they want to predict.” The client defines the target variable, but the machine does all the work.

Data Point Number Two: 30 Percent Approval Increase

This is the percentage of auto acceptance rate increases over bureau-based solutions. Explained Ayers, any Socure customer opening a financial services account has PATRIOT Act requirements. That means delivering and verifying certain types of financial data in order to serve different types of consumers. Historically, FIs would wade through the process of verifying names, addresses, dates of birth and Social Security numbers.

But in the digital age, “we’ve seen a huge gap in their ability to verify thin file individuals and their ability to verify millennials,” Ayers told Webster.

By taking into account everything from phone data to utility data – the aforementioned minutiae of everyday financial life – Socure is able to verify 30 percent more “good” consumers and create a “CIP scorecard” that has been approved by banks.

So why might folks balk at such a robust process? Ayers said there may be hesitation based on using machines in general, and added that it is simpler, perhaps easier, to understand a rules-based model versus one that ties together 80 or 90 variables.

Education remains a key Socure endeavor, said Ayers, in getting clients to understand “how the machine is reasoning over a much larger set of data to come to a known empirical outcome. I think there is a changing of the guard that is taking place right now in financial services.” He cited the emergence of chief innovation officers and chief data scientists among that vanguard.

Stopping 60 percent of the fraud by pushing it to the top 3 percent of a model, said Ayers, means that a financial firm can offer “a really good customer experience and automated customer experience to the other 97 percent of consumers who are opening accounts.

“It’s not about just capturing for fraud or just verifying good consumers – it’s about solving for both,” he told Webster.

The FI that has only a 60 to 65 percent auto approval rate has a tough operational burden to handle, said Ayers, especially when it comes to scaling digitally.

Data Point Number Three: 70 Percent KBA Reduction

This is the reduction in knowledge-based authentication (KBA) processes that comes through using predictive analytics, with models such as those from Socure. Ayers said that the FI need not ask which street you grew up on, for example, in order to let customers proceed across activities or transactions. Many Socure customers have come to Ayers’ firm seeking to get rid of KBA, due to the expense and, of course, the friction that turns customers away.

As to where we’re going from here: Ayers said that the movement to machine-based learning is widespread and will gather steam, making inroads into everything from credit cards to banking to … payments, of course.

Socure will, for its part, look at “new views of the consumer, new views of data points that will allow us to obtain a more comprehensive view of who the consumer is.” The company will continue to invest in a platform that does everything from digital to physical to offline identity verification through a single API. And, he said, ultimately the sea change will be driven by the FIs themselves, who will in turn educate the regulators – not vice versa. So the idea of a federated identity may still be a ways off.

“I think there’s been a lot of problems that have to be solved on the business side,” said Ayers, “before there’s this transformational shift to a tokenized version of identity.”



The pressure on banks to modernize their payments capabilities to support initiatives such as ISO 20022 and instant/real time payments has been exacerbated by the emergence of COVID-19 and the compelling need to quickly scale operations due to the rapid growth of contactless payments, and subsequent increase in digitization. Given this new normal, the need for agility and optimization across the payments processing value chain is imperative.