Nearly two and a half billion people around the world remain unbanked, and this isn’t for lack of financial institutions wanting to serve them.
One of the greatest barriers to servicing many of these people is the inability to verify customer identities. For the longest time, banks have relied on customers visiting brick and mortar locations to make deposits and transactions—it’s a lot easier to verify that you are who you say you are if you have to present yourself in-person.
“The world is moving online but the means by which you carry out identity verification is stuck in time in an offline world,” Hussayn Kassai, Co-founder and CEO of image recognition startup Onfido, told CNBC.
Kassai’s startup uses machine learning and artificial intelligence to provide a secure service for institutions and organizations to allow customers to prove who they are online. Applied in banking, Onfido’s product could make it easier for the previously unbanked to gain access to basic financial services from anywhere in the world, digitally.
Onfido requires users to submit photos of both the front and back of their identification documents as well as an image of themself to set up an account. To gain access in the future, users take a series of selfies. Artificial intelligence and machine learning algorithms then cross reference the identification documents with the selfies to assess user validity. Onfido can accept around 600 types of identity documents. What’s covered? All passports, a majority of driving licenses and national identity cards.
The primary challenge in using photos to verify identity is preventing forgery. To make the service accessible, Kassai told the Center for Data Innovation they couldn’t use the 3M scanners found in airports that use natural light, ultraviolet light, and infrared light to catch any forgery or tampering of IDs—the process needs to be able to rely on regular webcams to verify user identity.
To catch when someone might have tampered with the digital image of an ID before uploading it, Onfido uses machine learning to “generate heat maps to identify signs of digital tampering,” Kassai said. “For example, a pixel from another part of the document may have been copied from one part of the image to another, which a human would not spot but the algorithm can. A human can then read the heat map and see that the signs of tampering appear in sensitive areas, indicating a fake.”
Additionally, Onfido uses a “liveness” test to ensure someone can’t print a copy of the photo in your ID or another image out of you, use either of those and gain access to your account.
In cross reference to the ID, each selfie submitted thereafter generates a confidence score between zero and one. Attempting to use a print out of your ID photo would register as a perfect one—”[suggesting it’s not just the same person but the same photo,” Kassai said. “Such a score is an outlier: a different photo of the same person should generate a high score, but not a perfect score.”
Users are also must respond to instructions for two or three different actions, like blink, smile, say a sequence of numbers or look over your shoulder. “The user won’t know exactly what they will have to do beforehand,” Kassai said. “That’s much harder to cheat.”
And, going off his interview with the Center for Data Innovation, it would seem Kassai believes financial services could be just the beginning of possible applications of online identity verifications in society.
“As consumers become used to verifying their identities online, we’re going to eventually unlock higher-risk services, like healthcare information, or a full bank account with access to credit,” Kassai said. “Right at the top would be online voting—we should all be able to vote on our smartphones on election day, instead of having to queue up, sometimes for hours.”