Unlock editor summary for free
Rula Khalaf, editor of the FT, picks her favorite stories in this weekly newsletter.
As much as many of us value human interaction, there are some tasks that AI can do far better than any human. Customer identity verification is one of them, according to online finance professionals.
As anyone who has ever opened a bank account online or in person knows, almost everywhere will require a photo ID as proof of identity.
At a bank branch, an employee will look at you and your passport or driver’s license, for example, to see if they match. Or, if you apply online, you’ll be asked to take a photo of your ID and face, which will then be read by a machine. And the technology is likely to be more accurate at analyzing the images than the bank employee.
Why are machines better at identity verification?
“Humans are not very good at recognizing people,” explains Yuelyn Li, chief product officer at UK-based digital identity authentication provider Onfido. She points to studies that show machines are better than ordinary people at recognizing faces and as good as “professional face checkers,” who are more likely to be employed in law enforcement than banks.
There are many reasons why machines excel at facial recognition. Li says people can cross-check what someone looks like now with ID photos that may be years old, and match a 3D image of a real person with a 2D one. The machines are more precise because they can use measurements, such as the distance between facial features and the size of scars, to tell if the face and photo match.
The machines are also far more experienced than even the most skilled bank clerk. Li says Onfido uses a machine learning model that improves the accuracy of the technology powering its system over time. It can also, at the request of customers, check whether someone has previously tried to connect the same person with a different name.
Onfido also offers verification using other documents, such as utility bills. It looks at the document to confirm an address, checks that it has a valid issue date and that the template looks reasonable. Essentially, though, the company’s advice to financial institutions is that using records like utility bills isn’t the most secure way to verify someone’s identity.
How does online verification improve security?
Li says that online customer onboarding – taking a customer to a banking or investment platform and setting them up – is more secure than more traditional approaches, as it’s now usually done entirely within a bank or payment provider’s app. “One of the things that people are very nervous about is [using] things like call centers,” she says. “A lot of attacks that are happening now — a phishing attack or a text message or something like that — they [criminals] I’ll understand you because you’re exiting the app.’
This helps explain why online verification has become mainstream in recent years, with Li noting that “the majority” of onboarding is now online, among the traditional and new institutions Onfido works with in the UK.
Revolut, a British financial technology company that has no physical branches, has more than 30 million online customers worldwide. Its head of financial crime and fraud, Aaron Elliott-Gross, says the company is using a combination of internal models and systems alongside vendor services to tackle the “persistent threat” from, for example, transmitted images generated by AI and deep fakes out like “selfie” videos of real people.
“They [the vendors] they check if the selfie is live, they check if there is movement and sellers are very good at that. We provide an additional layer [of checks] on top of that because we see so many add-ons that we think we have something to add. . . to see if there is anything unusual we can detect.
What other digital checks can be done?
Elliott-Gross emphasizes that facial recognition and verifying documents such as bills are the “basic building blocks,” but much more sophisticated checks happen behind the scenes. “It’s really what you do beyond that that sets you apart. . . It’s very obvious [to the potential new customer] that we’re checking documents, it’s very obvious that we’re checking data,” he says. “Also, we do a lot of modeling work.
“[We] say, let’s not just look at the data that a customer has given us, let’s look at other signals we have about that customer – say [what the] The IP address of their mobile device looks like the history we have of their phone number or email address [which Revolut can get from data providers] to find out if this client is a real board or a scammer.
Revolut also uses its identity tools for more than just onboarding. For example, it can check that people still have control of their accounts if they change their phone or if they make an unusually large payment. This ensures that if the phone is stolen, the thief will not be able to run a large transaction from the victim’s account.
The fintech is also “very interested” in the emerging field of behavioral biometrics, which involves profiling, for example, the angle at which a user typically holds a device and the speed at which they press buttons or type. All of these idiosyncrasies can help authenticate the user once the user has been logged on.