Subscribe to the Liminal Newsletter
Stay updated with the latest news, data and insights from Liminal
How can document fraud detection help fight identity fraud? On this State of Identity podcast, host Cameron D’Ambrosi discusses building AI and machine learning models for a fraud vector with Inscribe Co-Founder and CTO, Conor Burke. This duo breaks down the challenges banks and fintechs face in combatting fraud.
Cameron D'Ambrosi, Senior Principal at Liminal
Conor Burke, Co-Founder and CTO at Inscribe
Cameron D’Ambrosi [00:00:04] Welcome everyone to State of Identity. I’m your host, Cameron Ambrosi. Joining me this week, Conor Burke, co-founder and CTO of Inscribe, a leading fraud detection and risk management solution for fintechs. Conor, welcome to State of identity.
Conor Burke [00:00:21] Thanks for having me, Carmen.
Cameron D’Ambrosi [00:00:22] Tell me a little bit about your background. It says here, you know, in your bio that you have a twin brother, Ronan, and that you two co-founded Inscribe in 2017. What made you decide to co-found a company with your brother in the fraud space?
Conor Burke [00:00:38] Yeah, great question. So this is interesting is actually during a brief stint with Deloitte, which I know you’ve spent some time time with as well, where I was working on a project on a national bank in Ireland, and this is my first exposure to this whole world of digital identity, onboarding, underwriting. And the first time we saw behind the scenes what goes on to on boards and new user interface of services. And this really early experience for me stayed on it in the back of my mind for a while. And then when Ronan I were. Talking about opportunities where we could have an impact on the world. This experience came back in top of mind and says, Yeah, look, given our technical expertise and our unique take on the on this particular problem, we we jumped at the opportunity. And yeah, as you can imagine, founding a company is a challenge. For anybody. But I think. When you have a chance to do it with someone who you trust and know very well, it can often make that a slightly less challenging experience. So yeah, it’s been fun journey so far. The link inscribed with my twin brother on.
Cameron D’Ambrosi [00:01:54] And and so, you know, you said you were on this engagement and that you were kind of seeing firsthand the challenges around onboarding. Well, let’s do this first. You know, tell me a little bit about inscribe and, you know, the features of the platform. Or put another way. What was that particular problem set that you saw, you know, when you were on this engagement and had us inscribed seek to address it?
Conor Burke [00:02:18] So one of the big learnings we had was behind the scenes. Any time essentially documents are involved during the onboarding and writing process. Consumers are usually in for a bad time, usually means long delays, lots of questions and back and forths. And this was what I saw during my experience. At that national bank and what we learned from talking to lots of customers. So I inscribe we really tried to create an API to solve this docking problem. So in this guy, we call ourselves a document fraud section and automation company, where we work with companies like Blue Vine, Petal Trap actions to automate, to review of documents during this onboarding and underwriting process. So every month we tax tens of millions of dollars of fraud, but also save consumers hours of our days of time by enabling faster responses. And then I guess more longer term, our goal really is to really tackle a lot of these bottlenecks that goes on behind the scenes in financial services that create these bad customer experiences and remove a lot of the frustrations that typically plagued this industry. Had some big success so far, but we really see so much more opportunity there to to to improve where we are today.
Cameron D’Ambrosi [00:03:29] I guess, to be blunt, you know, so I’m a I’m a bank. How do I consume in Scribe, you know, what’s the platform like and which steps of the lifecycle do you really see yourself as helping smooth some of those rough edges off of?
Conor Burke [00:03:43] Yeah. So if you if you look at this as a onboarding process, many, many people say having to say an ID verification vendor, which would work really well on like ID documents, like passports, driver’s licenses and so on, Where we really focus our attention on is the non ID documents such as like bank statements, text documents, utility bills and so on, which for a long time have kind of been ignored in terms of fraud detection on these documents, but also automation and building, you know, a really smooth solution there. So where it’s but usually fits in is primarily in an API driven setup. So when a customer applies to a fast service company, their documents will be automatically uploaded and stored in their system as part of the application. And previously these documents would be huge for review by a team of people, whereas within scribes they’d first be sent to our API where we do a series of like go checks. So for example, if you’re trying to prove someone’s address with proof, but a utility bill will classify that utility, be able to extract the name and address from that and to get the identity and do fraud detection on the document to make sure that can actually be trusted and give back an automated response back to our service company. You can then get back to the consumer and, you know, process with your application. And we do this for a whole range of document types. So it’s not just proof address. You can be proof of income, proof, employment and so on across a whole range of documents like. The top bank statements. But as I mentioned, their utility bill is company formation documents and so on.
Cameron D’Ambrosi [00:05:18] And so, you know, as I understand it, I’m not an engineer, but I like to pretend from time to time, you know, the process of building out a machine learning model for fraud vectors like this requires, you know, data, for lack of a better word, like lots of good documents, lots of bad documents, maybe some annotation and kind of manual training to get the model up and running. You know, what was climbing that mountain like? And how much of a challenge was it to kind of. Get past a solution that might require a bunch of manual eyeballs to one that can kind of get that flywheel spinning where you’re seeing more documents and therefore the model is continually strengthening itself.
Conor Burke [00:06:00] Yeah, that’s a really good question. Getting the data was definitely the biggest challenge in the early days of in Scribe, this Costar problem where you can’t really provide value until you have some some solution that works, but you kind of create a solution until you have the data. So we got really lucky Air Force in the early days to be able to partner with some of the leading fintechs who really have a big challenge on their hands with document reviews in terms of the fraud that was coming through that vector, but also the automation and time it took to process those. They were willing to work with us in exchange for they could help us with get some data and we could provide a technical, technical expertise to be able to develop a solution. So these early design partners and scribe really helped us, all of us cool to start problem with the data. And then once we had the initial version of in Scribe, we were able to use our use at the end to get more data from our customers who were able to help us along with layering, understand and annotating that data as well. So we’re now at a point where many of our customers do actually help annotate our data for us to tell us which documents are prevalent and which aren’t, and then we automatically feed it back into our algorithms. So early days, definitely biggest challenge, but now it’s one of our biggest strengths. This kind of data flywheel gets super powerful and that can give us this superhuman level of protection.
Cameron D’Ambrosi [00:07:25] And what are the traditional fraud vectors that you, you know, you typically see? Is it what I guess, you know, I have no experience in this facet of the fraud space, but I have a few hunches. I presume that most of the shenanigans that that people are trying to foist involve fudging a few numbers on their income, tacking a couple zeros on to it, changing their name slightly, changing an address, creating documents from scratch as part of either a full synthetic identity fraud or maybe an attempt to take someone else’s identity over. Is it primarily kind of manipulation or is it also kind of wholesale document creation from scratch that you’re you’re looking to dissect?
Conor Burke [00:08:06] Yeah, you’re spot on there. So really we see both first party interpreting fraud and depending on the particular customer and what demographics are serving, you can see different types. But yeah, as you mentioned, people being opportunistic, let’s say about trying to get a slightly larger loan or get more leverage or get through maybe some criteria that a financial services company has. There’s you fudge or like manipulated. That’s a bank statement to, um, to to foster checks or. On the third party side, as you mentioned, they’re completely fabricating fake identities and using supporting documents to further along their case. So, for example, creating completely fake documents or just simply buying them online. So there’s this. And interestingly, a whole host of economy of like buying and selling foster documents online so you can buy your any set of bank statements, IDs and so on from these online stores that you can go and use to apply to, you know, any financial services products online.
Cameron D’Ambrosi [00:09:14] Interesting. And, you know, these days, I know there’s kind of a mix of CAPTCHA methods. I’ve certainly done something like use my smartphone or use a webcam to capture a document. But then I’ve also maybe just downloaded a bank statement or something directly from the Internet and uploaded, you know, is the solution capable of detecting both manipulation of a photo, as it were, but also maybe something more? I don’t know if I would call it technically savvy, but let’s say my bank statement is a PDF going into the PDF and just manipulating the text at a fundamental level as opposed to, you know, not photoshopping pixels, but changing the bitmap, I guess you might say.
Conor Burke [00:09:53] Yeah, that’s a great question. We always strongly recommend our customers to or to encourage end users to upload, let’s say, PDF documents directly from the bank’s website. It’s a much higher fidelity, higher information density file format that allows us to provide better fraud detection and offer better defenses. Having said that, there are some there are a smaller volume of consumers who do still want to upload photos into one of the physical documents which we do provide for our section as well. But in general, it’s less secure methods. And of course, as with any fraud detection system, it’s important to have to extract that right balance between friction about what your consumers want to do and the defenses against us. So that’s usually a conversation that we have at our customers.
Cameron D’Ambrosi [00:10:46] I understand you also have. A Credit Insight product that’s kind of moving beyond the documents themselves and into. Right, you know, OCR extraction, but also, I guess what you call intelligent parsing and analysis of what’s contained in those documents. You know, can you talk a little bit about the future of the platform and some of those new product areas that you’re exploring that are maybe less about the documents themselves and more about, I guess, you know, what they truly mean from an identity perspective?
Conor Burke [00:11:18] Exactly. Yeah. Which is, you know, few exciting, exciting projects in Stripe currently on and all credit analysis, every one of them. So we really asked and you know, why why are our customers there? Why why do companies ask for documents in the first place? And what our bigger customer base is, large majority of our customers are underwriters, so they’re giving out loans or extending credit in some fashion. And when you get back to them to pay their tax form after checking whether they’re legitimate are not to extract information from them, you’re usually. Decision on them in some way. So, for example, understanding someone’s income or revenue for a business, understanding what kind of obligations they have in order to make some type of credit decision. So this was also an area of work that our customers were often struggling with in terms of combining reviews and it taking a lot of time to get back to customers. So we’ve been spending a lot of time on trying to automate that process process as well. So we now have to give back our customers. Essentially this kind of credit analysis of is this will this customer be a good customer in terms of them to repay their loans and so on? And yeah, going forward, we really see this as a opportunity to create an even better experience for an end user. So again, this was often a bottleneck, just like anytime documents are involved. In underwriting our onboarding process. Any time I say someone, let’s just have a credit score or doesn’t meet certain requirements. Usually they’re either rejected outright or does a big money review process. But with this credit analysis feature, we hope to automate a lot of that and gives us a response back to our customers.
Cameron D’Ambrosi [00:13:07] It sounds like the future is is anchored in digital identity to some degree for you. Where do you see this landscape evolving? Obviously, you know, there’s a lot of financial data sharing networks that are emerging now, whether it’s FDX, whether it’s Sequoia. You know, open banking is on the rise, which I think from a layman’s perspective, you might think, well, that means that, you know, documents are are not going to play a role. I would maybe take a contrarian perspective there. And don’t ever underestimate the degree to which paper is going to persist and continue to play a role in our lives, you know? How are you thinking about this evolution of the space and the intersectionality between, you know, these physical credentials, like physical documents and pieces of paper, and then that digital side of the coin, whether it’s open banking or other mechanisms for kind of digitally making these transactions work?
Conor Burke [00:14:02] Yeah, This is something we spend a lot of time on in Scribe. And our stance on this is really not trying to pick sides. We just want to serve where the world is today and where of the future. So today we do really see a balance between, you know, all these different data points coming from different sources and different mediums and media. And we’ve primarily focused on like the documents everyone say, but as are kind of hitting out with do like credit analysis, that’s a feature that is agnostic, let’s say. So you can provide data to API sources like open banking to feature like credit analysis. And this all comes back down to us for really just finding the bottlenecks to what’s causing the manual reviews, the slow experiences for end users and putting our resources there in terms of, you know, where I see this going in the future, I think there will be a balance or for like certain tasks, documents will play a bigger role than others and there’ll be like faster innovation in certain types of documents at API sources. But on the whole, I think documents will be playing like a big part for the foreseeable future, but same time, like many new sources. So I think open banking is one, but I think open finance in general and just having more alternative data is going to continue to accelerate and create more types of analysis needs to be done on this and next couple of years, trying to ingest a lot and make sense of that is going to be an interesting challenge for a lot of these fintechs.
Cameron D’Ambrosi [00:15:43] What do you hope to see in the space? You know, we’ve talked a lot about what we kind of expect to see. You know, do you see your customers kind of changing their posture and tactics to evolve to this, you know, new fraud landscape? And, you know, in some ways, how have you seen the customers run with your product and maybe ways that you you wouldn’t have expected and demonstrate value that maybe wasn’t apparent on the label for you?
Conor Burke [00:16:08] Yeah, This is interesting. And that’s. I think fraud has often been seen as a gatekeeper to innovation or growth. But what we’ve seen over the last while is really using fraud detection on products that can scribe as a way to achieve growth, get greater. You know, better products. And this was kind of interesting during that we’ve had over the last while and perhaps understanding the importance of so, for example, you know, with we have products like inscribe, we’ve been able to some of our customers to increase the and increase the demographic by like say 30% of their address book customers just by with things like better understanding your risk profile of certain demographics or they have serve customers who don’t have credit scores. And this is enabled I guess, interesting use cases there. And I think looking to the future, I think we’ll see this trend increasing where. FinTechs and banks are in their effort to achieve growth in the tougher times will be looking at alternatives and different ways of doing things to be able to achieve that growth. And we’ve seen early signs of that, and I definitely hope to see that in the future and expect to see that, and we’d like to see that as well.
Cameron D’Ambrosi [00:17:31] Where, you know, are are fraudsters holding serve, for lack of a better word? You know, we’ve seen a ton of talk about the kind of COVID related surge in fraud, people taking advantage of of kind of government benefits and programs. And now we’re heading into what looks to be a recession, which I think oftentimes is a leading indicator of maybe folks. Dipping a toe into fraud or or taking advantage of a moment to try and make a few bucks. Like, are you seeing any any trend lines emerge in percentages of excuse me, percentages of fraud in fraud applications or or from your perspective, is it kind of just a a steady baseline that maybe is going to keep humming along regardless of macroeconomic conditions?
Conor Burke [00:18:20] Yeah, you’re going to hint. Something really interesting here that we did observe was during the pandemic, we saw a massive spike in fraud. As you mentioned there, the government assisted loan program, the PGP scheme was quite an interesting scheme where there was a lot of fraud going on. But yeah, we saw a continuation of some of that fraud in other parts of the financial service ecosystem after those schemes were were closed down. In terms of what we’re seeing then as we move into this kind of I guess next next economic storm is is is two things. So one on the fraud side is we’re seeing slightly elevated. Fraud rates on you know, on the kind of the credit card industry. So we work with a lot of personal credit cards and corporate credit cards, and we’re seeing obviously increased volumes in those products, but also increased fraud at an absolute level, which is an interesting trend that we’re going to keep an eye on in terms of what we’re seeing from a customer’s perspective and what they’re doing, we’re seeing. Our customers and companies in general are taking a closer eye at our fraud defenses and really trying to, one, ask where they can get a better or increase the demographics that they can serve, but also get our get a better handle on where the fraud losses are occurring. So during the last couple of years where as a capital is cheaper, they have looser fraud budgets that say so how much you lose every month to the higher budget. But no company, no matter how big, likes to have, you know, big write offs on their balance sheet every year. So we’re seeing companies really trying to tighten their grip on on those losses and yet are looking at statistic and Scribd are also looking at the whole fraud stack to see if they can tighten their grip on.
Cameron D’Ambrosi [00:20:19] Are you seeing the the arguments resonate, you know, with customers around thinking about your platform in growth terms as opposed to fraud terms? I think that’s always one of the fundamental challenges in the risk and fraud space is not being seen purely as kind of a cost center item that you are part of the growth equation and that you can demonstrate that you’re actually being additive to PNL as opposed to just, you know, stopping bad things from happening. Would you say minds are kind of shifting across boardrooms in financial services and fintechs about the role that these technologies can play in kind of delighting consumers and driving conversion? Or is there still a lot of talk around, you know, how much fraud solution do we really need?
Conor Burke [00:21:02] Yeah, I think we’re learning we’ve had is I think this depends a lot on the type of company in the stage of company. So I because we’re on a growth trajectory are so for example with a lot of earlier stage fintechs, we’re seeing a larger focus on their practice and growth because they have growth and we see fraud as just trying to reduce the losses that they see there. But on the larger, more established side where, you know, a merchant bank, they they often have their fraud defenses really in place and that they like where they have like where they are in terms of the products they see and would really like to expand their addressable market by having better risk risk controls. And that’s where we see, you know, fraud often being used as a as a way to our solutions that can show up as a way to achieve growth at a later stage. So yeah, that would be like a common next to common ways we see companies approach our problem. Having said that, we would say approaching or viewing fraud detection as a growth lever earlier rather than later is always better. Designing fraud detection into your products, Creating features based on that is you can reap rewards earlier ideas.
Cameron D’Ambrosi [00:22:27] I love it. Well, last thing here. I like to call this the shameless plug opportunity. ConorFor people, for people who are listening and want to learn more about inscribe, get in touch with you. What is the best place for them to go and do that?
Conor Burke [00:22:45] Yeah. Cheers. Yeah. Please visit us on our website in Scribe Dot II and you can contact us there. And if you want to tag this podcast, I’d be happy to chat with you personally or someone from a team we can talk to.
Cameron D’Ambrosi [00:22:58] So amazing. Well, Conor, thank you again for your time. Really, really great chatting and congratulations on all your success and hopefully looking forward to sharing a pint with you sometime soon in in fiscal space.
Conor Burke [00:23:12] I appreciate this. Great chatting.
Onfido CEO Mike Tuchen shares his insights on the digital identity space, and the challenges businesses and consumers face. Tuchen discusses the need for a privacy-first approach, the growing demand for reusable digital identities, and the shift towards user control of personal information.
Secfense Chief Technology Officer, Marcin Szary, joins host Cameron D’Ambrosi to explore the current authentication landscape. They discuss why FIDO Alliance has been a truly transformative moment for the death of the password, how Secfense sets itself apart in a crowded and competitive landscape, and Marcin’s predictions for the future.
Measuring the reach of digital advertising and smartphone app performance is a difficult task made more challenging by tightening data privacy regulations. Edik Mitelman, SVP & GM of Privacy Cloud at AppsFlyer joins host Cameron D’Ambrosi to discuss the current state of the consumer data landscape, how platforms must balance first- and third-party data usage, and why the death of cookies is a tremendous opportunity.
John Bambenek, Principal Threat Hunter at Netenrich, joins host Cameron D’Ambrosi for a deep dive into the current trends across the cybersecurity landscape, from ChatGPT and deepfake offensive threats to leveraging data analytics across your XDR, SIEM and SOAR technology stacks for improved defenses.
Vyacheslav Zholudev, Chief Technology Officer of Sumsub, discusses the current state of the identity verification market with podcast host Cameron D’Ambrosi. They explore the factors driving platforms to move beyond basic identity verification and into other aspects of the digital identity lifecycle. They also discuss the challenges of implementing artificial intelligence in regulated use cases such as anti-money laundering (AML) transaction monitoring.
Host Cameron D’Ambrosi is joined by guest Marcus Bartram, General Partner and founding team member at Telstra Ventures, to dive into his company’s digital identity investment thesis, its transition from corporate VC to an independent fund, Strata Identity’s right to win, and the expanding role of identity in the cybersecurity landscape.