INTERVIEW: Onfido CEO Husayn Kassai

INTERVIEW: Onfido CEO Husayn Kassai

Having started up in 2012 as the brainchild of a few Oxford graduates, Onfido has in recent years established itself as a leading provider of biometric authentication technology. The company takes a selfie-based approach that has become particularly popular over the last couple of years, enabling identity verification with a combination of facial recognition and document reading technology – and it has been embraced by a number of organizations in areas like financial services, cryptocurrency, and the sharing economy.

Outside of those kinds of regular business activities, Onfido has also seen a couple of big developments this month that have particularly illustrated its high profile in the identity industry: the company announced that it would lead a training course on ID fraud for INTERPOL, and it joined the FIDO Alliance’s Board of Directors. It’s on the heels of these advancements that Onfido CEO and Co-founder Husayn Kassai spoke with FindBiometrics Digital Marketing Director Susan Stover, in an in-depth interview covering the company’s evolution, changes in the market and the culture, Onfido’s AI-driven technology, and more.

Read the full interview with Husayn Kassai, CEO & Co-founder, Onfido:

Susan Stover, Director of Digital Marketing, FindBiometrics: Can you give our readers a brief history of Onfido?

Husayn Kassai, CEO & Co-Founder, Onfido: We started seven years in 2012 after graduating from university. I met my two co-founders at Oxford and we, in many ways, could see that the world was moving online and yet businesses were no longer getting to their customers face-to-face. So, there was going to be a need for a more effective way to verify that customers are who they claim to be in a remote setting.

FB: At FindBiometrics and Mobile ID World we have been reporting on Onfido for quite some time and in 2018 and even so far in 2019 we have seen a lot of growth from your company, can you tell us more about that?

Onfido: Yes, of course and it also goes back to when we founded the company, we could see that there were basically two approaches and we felt that there are no longer fit for purpose and that’s just been proven out over time. So, one is that if you want to verify someone as who they claim to be you currently can use a credit bureau but that credit bureau model is essentially a log of everyone’s date of birth, name, and address as a result of them at some point having gone inside a bank branch for instance, with paper documents and getting verified. The challenge is that a lot of that data is now sort of compromised because it’s on the dark web as well so it doesn’t offer the security that is needed. We could see the early signs of that when we started in 2012 but with every passing year and in fact, every passing month, with more and more breaches happening the less currency these bureau models have and then obviously it’s about a hundred and fifty-year-old model and in our view is a bit out of date.

The alternative is to verify people face-to-face as you would in the bank branch, or at a hotel reception for example. And with the banking onboarding face-to-face instance as the neo-banks in the fintech wave has proven, consumers would much rather have a seamless convenience-based experience so they can verify themselves at home.

Specifically in 2018 and the run-up to that, the reason why we have been fortunate in timing is twofold: partly due to the technology side of things, call it supplier technology but also it is because of consumer demand. If I talk about the consumer demand first that is customers ultimately putting a lot of emphasis on the experience of signing up. They want to sign up from the comfort of their own home using their phone and they no longer want to have a paper-based face-to-face experience. That is on the consumer side and that in our lifetime is going to continue. The thinking is: if I can do everything with my phone nowadays, why would I go and open up a remittance platform face-to-face or send money, and so on. So, everything is becoming this smartphone focused generation and the convenience economy and all that sort of feeds into the demand side of digital and remote onboarding.

On the supply side and the technology front we’ve been very fortunate. In 2012 a few things came together to help us do this for the first time. One was internet connectivity globally increased and the second was the camera phone quality improved so that you could take a high-quality image of not just your face but also of your photographic ID. The third, was the improvement in biometrics proved how machine learning powered biometrics can be done better than the human eye. And then there are other small things that continue to improve and play the contributing factor. I guess servers becoming more accessible and therefore making it economically possible for machine learning models to run and so on.

Back to those trends and the theory behind our sudden company – the hypothesis was consumers are going to continue to want to access services quickly and conveniently using their smartphones in a remote setting and on the supply and technology side, machine learning models can do a much better job of not just government ID checks but also facial biometric checks, and in the past two years we have seen both sides accelerate.

FB: That is fantastic, specifically for Onfido providing mobile biometric technologies for industries like financial services where legacy systems are leaving these organizations vulnerable.

Onfido: Exactly, and financial services has in its own way gone through a transition phase and we are probably still in the middle of that as well. But a decade or so ago a financial institution gained its credibility from its book of assets. So, where you’d go to a mortgage broker or even an insurance broker and they’d derive the potential value from how big their book was and that would give you the confidence to go and bank with them, for instance. That model is completely changed now because consumers first and foremost, when they think of which bank they should sign up to, they don’t often now know or even care about the assets as such. What they want to know is who has the slickest onboarding or the easiest app or the most convenient or who’s available 24/7 and so on. So, we shifted from the prime services, in our view, from focusing on assets and the legacy of a business to a much more customer experience one and that has given rise to the fintech wave that we’ve seen.

FB: Onfido’s identity verification engine was recently awarded three Cybersecurity Excellence Awards – Congratulations! – one of which was for fraud prevention and one of them was for artificial intelligence security. How do you see a hybrid human AI approach help mitigating fraud risks?

Onfido: First of all, fraud, as you know particularly well, is a massive problem and it is just getting worse. Online identity theft in particular is in the US the largest crime and the fastest growing crime. In our view all fraud especially if it’s online has just one common denominator and that is identity theft because the perpetrator, the fraudster, doesn’t want to get caught. So, that in some ways is the one chance that a provider like us can work to catching or reducing these bad actors or fraudsters.

The second thing is that the sheer scale of it is mind-boggling. The UN estimates are that between 2 percent to 5 percent of the world GDP is laundered money which is between $800 billion to $2 trillion. A big chunk of that laundered money is used for human trafficking, drug trafficking, terrorist financing and so on.

From the outset we could see that this is a massive problem and that the current processes aren’t fit to reduce or solve these because ultimately, it’s about a dozen companies that fail to catch the lion’s share of these money launderers. I mean, it’s the credit bureaus and a few other of our colleagues in the industry. So, we are all, I guess in some ways, supposed to be the bottleneck for this, and yet starkly very few have been caught. And the last thing that was particularly frustrating or just didn’t make any sense to us was you go to Bank A or you go to let’s say Fintech A and you try to cheat the system and you get caught, that’s not going to make you go home and stop, you are just going to go to Fintech B and Fintech C until you find the weakest link and you make through. So, you can’t really have a convenient easy signup process without getting really good at catching fraud and in fact reducing fraud.

Now, back to our machine learning and hybrid approach. What typically happens is when a government ID or facial biometric for instance comes through and the machine learning models extract the details and determine whether the patterns and so on seem genuine based on the tens of millions of checks that we’ve done, then it’s all fine and it goes through. But if we suspect something, we have human experts double checking and when a human double checks, they use other tools to ascertain as to whether the ID seems like it’s genuine or if it’s just noise and it’s fine. For example, if your driver’s license went through the washing machine, it might be stained and when you show that to us and if it’s flagged the machine learning models won’t know why or the context whereas a human can better determine if this is an anomaly that has been seen in the past or know that it seems like a new one.

What happens as we are powering a greater proportion of especially the Fintech ecosystem, we see repeated highly sophisticated fraudsters using the same techniques. So, when we learn about a fraudulent bad actor from one customer for example, all models learn and that helps benefit all the other customers. By sitting in the middle of all of these platforms our models get better at detecting fraud and that’s essentially why our large enterprise customers have never switched away from us because once you are able to show them that you are able to detect a fake that their own teams would never been able to, then that is what is particularly impressive and it is only possible with machine learning.

Back to your question on why is a human needed; a human is needed in two ways: one if we suspect something, we don’t want to just reject it because that might be a legitimate consumer that might want to gain access to rent a car or to open a remittance accounts for instance. So, with a human double-checking we can pass more checks than we otherwise would be able to and to supplement the models. But the second and more important thing is that they’re offering a feedback loop because whenever the human agents are responding, they are also annotating the models so over time the model learns and become even more effective. So, that’s another key essential part of how the technology works.

Our aim is not necessarily just as a compliance role – we do that as a consequence for what we do – nor is it to exclusively to catch fraud but it’s to balance the trade-off between helping give access to as many people as possible, helping 99 percent get onboarded as seamlessly as possible while preventing the bad one percent, the bad actors. The more effective we become at stopping the 1 percent of bad actors the more effective we become at helping the 99 percent come across. It would be very easy for us to just reject the 5 percent of all ID’s and have zero human touch, that’s very possible, but that would prevent us from obtaining our goal on helping people gain access, which is taking it from say 94 percent or 95 percent and pushing that to 99 percent when you have human review.

FB: In your opinion, where do you see the future of identities? Is the future of digital identity mobile?

Onfido: Mobile is likely to play a big role in the future of identity but ultimately it depends on the way it is looked at. But one aspect, that we at least believe will happen, is that we’re going to move from the centralized credit bureau, which often is sort of a leaking database model, too much more of a decentralized world where the consumer owns and controls their legal identity. Now that in all likelihood will somehow be tokenized on a phone – it doesn’t have to be that or exclusively be that but that is very likely to play a big role in it. The thing that matters and we definitely believe is that the days of centralized credit bureaus are dying and ending and we’re going to move to much more of a decentralized world where the consumer owns their legal identity. That in essence is sort of our company Mission – to help build this open world where identity is the key to access.

FB: Finally, what can we expect from Onfido? I know we will be seeing you at Money20/20 in Amsterdam, but can you let us know what else we can expect from you this year?

Onfido: Yes, of course. So, across three fronts – first, we are continuing to hire very senior team members from Amazon, Cisco and sort of world-renowned companies and we will be announcing more of them over time. We will grow the team and continue to do that especially across the technology sector. Second are the expansion markets that we are focusing on including India and Southeast Asia where we are going to do a lot more there. We have a small team and we are growing it. And the third thing is product releases. We have a whole host of new features that we will be announcing over the coming year.

FB: That’s great and thank you so much for joining me today, it was really great to hear from you.

Onfido: Thank you.