Ep. #999 - Improving Fraud Detection
In today’s Startup Hustle episode, learn how to improve fraud detection. Matt Watson talks with Conor Burke about the ins and outs of fraud detection. Our guest is the co-founder and the chief technology officer of Inscribe. Let’s discover new tech being used in fintech, the use of AI in fraud detection, and how people adapt to novel technologies.
Covered In This Episode
Why are financial services backward in many ways? What data should be reviewed for fraud detection? How does AI help with fraud classification?
Matt and Conor’s conversation tackle all these things and more. Aside from these points, they also share insights on how various countries differ regarding tech and consumer habits. And there’s also a bit of what direction Inscribe is taking for its company’s future.
Listen to this Startup Hustle episode now.
- The beginning of Inscribe (01:37)
- Fraudulent applications at the bank (04:29)
- The kind of deceitful things that people do (05:49)
- How Conor found a way to solve the problem (08:36)
- Determining fraudulent applications (10:29)
- Fraud rate when using Inscribe (11:48)
- Fighting against errors or false positives (15:07)
- Adoption of using Inscribe tech (18:05)
- Changes of product use in the future (19:41)
- People adapting to fintech (21:20)
- Using AI for fraud detection (22:32)
- AI usage to make fraud detection processing more efficient (26:04)
- Building the software that helps analyze documents (30:01)
- Inscribe’s customer base (30:53)
- The use of credit cards from country to country (31:49)
- The future of Inscribe (33:04)
- Understanding the problem that needs to be solved (36:52)
We want to create a fair and efficient financial service industry. So it’s not just necessarily about reducing the time it takes to get back to customers. But it’s also trying to make it fair, so trying to take some of that bias out.– Conor Burke
What’s really interesting is that humans are really good at this stuff. And we’re often compared to human performance. But what we want to do is get to a superhuman level. So for us to reach the superhuman level, that’s the challenge.– Conor Burke
Some of those AI models seem like things that a lot of people could take advantage of. Just like analyzing a picture and it tells you what’s in the picture. And you have things like Amazon web services and Azure now that provide different AI models that do some of that stuff out of the box.– Matt Watson
Buy and sell legit items, machinery, furniture, and other things. It’s easy to do that with Equip-Bid Auctions, your Midwest online auction marketplace. They provide a fun liquidation shopping experience to their valued bidders. Visit their website and start growing your independent business now.
Don’t forget to check out all of our Startup Hustle partners. They may have the right services that may answer your business needs.
Following is an auto-generated text transcript of this episode. Apologies for any errors!
Matt Watson 00:00
And we’re back for another episode of Startup Hustle. This is your host today, Matt Watson. I’m very excited to be joined today by Mr. Conor Burke from Inscribe. We’re going to be talking about fraud detection today. I mentioned a lot of different companies where fraud is a big issue, and they have some expertise and solutions to help with that. Excited to learn more about their business. Before we get started today, I do want to remind everybody that today’s episode of Startup Hustle is sponsored by Equip-Bid Auctions – your Midwest online auction marketplace to buy and sell stuff. Equip-Bid provides dedicated support to affiliates in Kansas, Missouri, Nebraska, and Iowa. Join the team and sell everything from heavy machinery to home goods, vehicles, and boats, to tractors and furniture. Go to equip-bid.me/startup for details, or just click the link in the show notes. So, Conor, welcome to the show, man.
Conor Burke 00:59
Thanks, Matt. Good to be here.
Matt Watson 01:01
So I see here you’re one of the co-founders and the CTO, the Chief Technology Officer, of Inscribe. Did I hear earlier that it’s you and your brother who started the company?
Conor Burke 01:11
Yeah, so my twin brother, Ronan, and I founded Inscribe five years ago. Which, as you can imagine, definitely provides for a unique founder relationship. But yeah, it’s been a real superpower for us.
Matt Watson 01:21
Are you guys identical twins?
Conor Burke 01:25
We’re not identical. But many people do have difficulties in detail or distinguishing between us. So yeah, we do have a little guide internally on how to, you know, a few tips and tricks internally to describe how to tell us apart.
Matt Watson 01:37
So the company I work at it’s called Camp Digital. And our CEO, her name is Katie. She has a twin sister named Meg. And Meg is also one of the executives of the company. And yeah, they do some shit to people. Or we’ll be on a zoom call, and one of them will get out of the chair. And then the other one will sit down. You’re like, wait for a second, hold on. It’s like Katie roaming. So I’m sure that always makes for some fun moments. So tell me, how did you guys come to start Inscribe? Did you guys work in an industry that had a lot of issues with fraud? And you guys said, hey, you know what, we need to go solve this problem. Or how did you guys get started with this?
Conor Burke 02:19
Yeah, good question. So it was actually during my college years I was in university studying engineering and had this opportunity to work at one of the national banks in Ireland. And obviously, very young, in my career, I was tasked in the, you know, one of the back office teams trying to try to apply some technology to some back office operations. So as you can imagine, this was, like, a very boring, mundane kind of environment where people were doing a lot of manual work. And it was really this, like, early experience for me that just showed just how, like, I guess how backward a lot of financial institutions still were in, you know, this was 2018 2019. And that was the real nexus for us to really start asking questions, you know, why? Why are financial services so backward in so many ways, and relating this to her some of her own experiences with applying for bank accounts, checking accounts, or credit cards? Anytime you’re told to wait a couple of days or wait a week, we’ll get back to you. There’s usually something going on behind the scenes that say, like people are involved, or you’re in some queue, or you’re in a backlog. And this kind of frustration was really combined with my own kind of experience in this back office team with a bank? Where was the original idea that spurred us on to try sagas with our engineering background and our skills in technology?
Matt Watson 03:51
So at that bank, were they processing, say, 1000s of loan documents a month? And, like, 5% of them would be a fraud? Or, like, how much fraud were they running into?
Conor Burke 04:04
Yeah, exactly. So there are essentially their 2000s of these applications coming in a day for things like credit cards, savings accounts, and so on and so on. And what we, as a team, are really tasked with was freesheets applicants, trying to get one determined are already fraudulent? Or do they actually intend to repay this loan or be a good user, and what’s your ability to repay this loan or, you know, be a good customer? And yeah, as you noted there, around 5%, on average, of these napkins that you process today are fraudulent. And this brings up a really tough challenge for lady’s companies in that it’s really hard to find that 5% out of that 100%. You need to really look at a lot of details. I look at a lot of data. And for a lot of the more traditional financial institutions out there, they haven’t really adopted the same or the adequate level of technology to make it easier. So there was, you know, lots of fraud, but also lots of delays that were still occurring at that time.
Matt Watson 05:09
So why? So why would people be committing fraud to begin with? Is it mostly identity theft, where it’s like I stole my neighbor’s information and applied for a credit card? And then I went and spent all the money on the card and left them with the bill basically, like, is it mostly identity theft? Or, like, what kind of fraud were people trying to do here?
Conor Burke 05:28
Yeah, as you can imagine, people get quite creative here. I think it’s useful to look at fraud in two buckets. One is first-party fraud. And then the second is party fraud. So in the case of first-party fraud, this is where a person is applying on their own identity but just wants to get a slightly bigger loan to buy a slightly more expensive car or that I get a slightly better mortgage. And these people are, let’s say, just inflating their income slightly, or if you’re taking out a loan for your business, inflating or Mister Mister Epson tasting, misrepresenting some pressure business. And this is quite common. And we call this internally like, almost like, opportunistic fraud. So people are taking, or like making, maybe like once-off fraudulent claims on the other side, then you have interpreted fraud. And this is where identities are usually stolen. But sometimes, fake identities are created from scratch. So, for example, if someone could, for example, steal your identity, they might get access to your credit score and then apply for a loan, and then you run away with the money, and then you’d be stuck with that cash. I think our subcategory of third-party fraud is also synthetic identity, where people are essentially creating fake people from scratch, where they take some real details. So they might take, let’s say, your email address plus your where you live, plus maybe your SSN number, and put all those together. And each of those is real, like SSN numbers real, the email addresses real, but they don’t all correspond to the same identity. And that can often pass through a lot of fraud detection systems. And then again, that’s another really common way of committing fraud that these, I guess, more professional fraudsters will be using.
Matt Watson 07:14
So, you would consider it to be fraud if I just basically lied about my income versus stealing somebody else’s identity. Those are both considered fraud. There’s not really a big differentiation there.
Conor Burke 07:30
Exactly. Of course, there is a, you know, in some cases, the outcome may be different. So if you do, if you are already in first-party fraud, and you do intend on repaying, the effect isn’t as bad on the financial institution, or the fraud losses won’t be as large. But yeah, it’s still considered fraud. And you know, it’s definitely something that these banks want to catch before it gets too big of a problem.
Matt Watson 07:55
So how did you guys set out to solve this problem? Is this kind of a mix of doing background checks and things like that? Like, how do you? How do you know how you even know my income to validate that I put it in the right information on the loan? Like, how did you guys set out to solve this?
Conor Burke 08:14
Yeah, that’s a good question. So I think fraud detection is interesting in that a lot of it is based on what data you are looking at. So we’re all probably very familiar with, you know, looking at IP addresses or looking at device information, or even, let’s say sanctions. Listen to one where it says just list online or non-fraud, fraudulent entities. So our kind of approach is Inscribe, trying to look at data that nobody has really looked at before. And for us, in this case, it was looking at information, or artifacts, in the documents that are usually supplied during applications to financial services companies. See these documents like bank statements, tax documents, or utility bills. These have like certain artifacts in them that can be used to, you know, catch fraud and look for patterns. So instead of, let’s say, a, you know, traditional fraud check system, maybe doing FSA ID verification check, we’re looking at, let’s say, a bank statement and trying to figure out, has this tampered in any way? You know, Was there evidence or Photoshop on this document? Is this document just completely fabricated and actually doesn’t really make any sense at all? To a human, let’s say, these would look all real. But really, behind the scenes, you can start to dig a bit deeper into how this hope, essentially every document you receive, was created. And you know, whether it matches what you expect. And this gave us a new lens into applications, which allowed us to, you know, catch fraud, better customers were not caught before.
Matt Watson 09:51
So your guyses solution isn’t just based on taking the inputs on the applications, but actually looking at the physical, You know, documents that come along with the loan application.
Conor Burke 10:05
Yeah, so we kind of split up into a few different areas. So when you apply online for a loan or financial services product, you will often have to supply documents of some sort and wheeze. We look at these documents, as you know, one as a file itself, so say this digital file, or it’s a PDF, or a PNG or JPEG. And we do a whole series of analyses on those files. So how the documents were created by looking at the pixels and so on, but then also looking at information in the documents. So these documents are usually quite rich sources of information. So you have things like names, addresses, dates, account numbers, and bank statements, and you have a whole series of transactions. And this all paints a really, really accurate picture of, you know, an identity, which we then use to catch fraud. So we look for anomalies with Ufford discrepancies. And we look for identities or entities within these documents that are, you know, unknown lists of fraudsters, and then we alert users to those.
Matt Watson 11:08
So what kind of fraud rate do people get to when they use your product? So if it’s normally a kind of five as an industry, you know, benchmark or whatever, if they use your guys’ solution, what are they able to get their fraud down to?
Conor Burke 11:23
Yeah, well, within thrive, we usually catch on yet 5%, or we see a flat rate around 5%. But on average, you know, depending on the particular sub-sector of financial services, you know, we’ve seen customers get down to, you know, one or one and a half percent in terms of actual losses that they have to take on their balance sheet. Which, I will say it sounds small, but let’s say if you’re giving out maybe a billion dollars of loans every year, as a fashion service company, or multiple, multiple billions, and these kind of fraud losses, as well as credit losses quickly add up to, you know, significant volumes, or, you know, dollar values.
Matt Watson 12:07
So, I had somebody on our podcast recently, one of our guests, their last name as default. And I know somebody else whose last name is no, I’m gonna guess your guys’ software does not like them. That’s very
Conor Burke 12:18
unfortunate. Yeah. Even if I was a bank, I’d be sure to have some subconscious bias in me if someone secondary was default. But yeah, that’s yeah.
Matt Watson 12:28
Yeah, I think they said it was a French, it was a French last name that was, you know, not super common. But you know, there were lots of French people that had that last name. So it was interesting. I never have.
Conor Burke 12:41
Yeah, I will say it, that is actually an interesting point in that, you know, something that’s coding Scribe is, you know, we want to create a fair and efficient and fair and efficient professional service industry. So it’s not just necessarily about, you know, reducing the time it takes to get back to customers also, trying to make it fair, so trying to take some of that bias out. So maybe, if you are a person looking at a loan with a second name of default, that may subconsciously bias you. But if you’re a machine, you know, we haven’t, no machine has necessarily been trained to recognize the word default as a name as a suspicious name.
Matt Watson 13:15
So when you guys also process people looking for names are on like known terrorists lists, and you know, all that kind of stuff do though?
Conor Burke 13:24
Yeah. So I guess, if there was an error in the list lower default? Or let’s say no, it was mentioned. Yeah, that will be quite unfortunate.
Matt Watson 13:33
Like I have, I have another friend whose first name is ISIS. And she’s the first one who has had all sorts of problems because her name is ISIS.
Conor Burke 13:44
Yeah, it’s a challenge. And I think you often hear stories to where young people with certain names or certain, let’s say, Darnay, might accidentally be on a list, they have to go through a whole series of, you know, painful bureaucracy to get her name off that list or, you know, to get to all of the attributes like exception handling second airport, when you get asked to do with like, second screening is yeah, this creates like a really bad experience in this financial service industry. So yeah, I will say though, there are a lot of companies including Inscribe that are trying to remove all these exceptions and like, take this out of the way. So yeah, hopefully, hopefully, these people are there. Yeah, seeing this experience improve.
Matt Watson 14:29
So what kind of imagined that the problem you guys, one of the problems you guys are going to have is you’re always fighting against false positives, right? Like, you’re flagging 5%. But like, a portion of those are people with these kinds of names or whatever. And like, your guys’ software, just it’s never gonna be perfect. So what happened? I’m just curious what happens to those people if their loan application gets flagged by your software? Are they just kind of screwed or just ends up in a bunch of manuals further?
Conor Burke 15:01
Further through Good recommendation, we’ve worked really hard over the last number years to try to tackle this. And, you know, initially we have, we’ve used a score essentially. So we can give a score back about just how confident we are that something is either, like 100%, legitimate, or, you know, absolutely fraudulent. And this gives us confidence for certain for large portions of the populations or over applicants that come to our systems about whether they can accept or reject. And in the cases where, let’s say we’re less sure, usually, in the case where these, there’s more likely to be false positives here, we can recommend a manual review to take a look at those. So as you mentioned, false positives are a really important topic. And it’s always a balance between, you know, what level of let’s say, even false negatives are you willing to accept to reduce a false positive rate, and this comes down to, you know, what kind of customer or customer experience you want to have? Would you rather take some fraud losses, to create a better customer experience to avoid any false positives, or maybe you want to avoid all fraud losses, and maybe some portion of your customer scorchers extra manual review stage? And this is usually a conversation we have with many of our, many of our customers, and we take into account things like, what’s the cost or bad business outcome here, you know, if it’s a business over half a million dollars, it’s okay to you know, add a little bit more friction. But if you’re setting up, you know, a new FinTech consumer app, you really don’t want any friction in that process. So you’re willing to take a bit of fraud loss?
Matt Watson 16:31
Yeah, if it’s a $100 microloan, you can’t exactly afford to spend a lot of time trying to manually review it. Like it doesn’t make any sense. Exactly. Yeah. Well, as a reminder, today’s episode of Startup Hustle is brought to you by Equip-Bid Auctions, an online marketplace dedicated to growing small auction businesses. They’re solving problems and providing a fun e-commerce or liquidation shopping experience to devalue bidders, go check out their incredible offers and sign up at equip dash bid.me/startup. And there’s a link in the show notes. So I’m curious as to why it sounds like you guys are solving a real problem. Right. And I’m curious, so how has your guys’ company growth been over the last five years? You know, who? Who is? Who have you seen adopting and using this type of technology? I’m curious, do you have any great case studies? Or, you know, how, how did that go from a growth perspective?
Conor Burke 17:27
Yeah, great question. You know, I think startup journeys are all different. And, you know, I think other categories, the first two years or so have been described as it’s really about us learning the problem and working with a few select partners. And really, to work with them and try to figure out is this solvable from a tech perspective. And then you have last last couple of years, we’ve had opportunity to work with some really great, great, great companies across, you know, many sectors, financial services, so some of the top crypto platforms, top payments platforms, and, you know, some of the more recent, you know, corporate credit card businesses, so companies like ramp that have experienced massive growth over the last couple of years, as well as the more traditional they, you know, business lenders like blueline, or cabbage, who have been, you know, longtime users windscribe. So, we don’t start yet, but we do have quite a lot of coverage across us. So if you’re maybe a consumer business, it’s reasonably likely that, you know, you know, a large portion of people in the US would have, you know, applied to financial services, which has used Inscribe at one point or another.
Matt Watson 18:39
Wow, that’s awesome. Huge. I mean, that’s a huge, huge success for you, right? I mean, that’s awesome. Congrats on that. So as your guys’ products changed over time, I mean, do you now have like multiple different products? Are you guys focused on? You know, really like one kind of core product? Or do you kind of see the product suite changing to be doing different things?
Conor Burke 19:03
Yeah. So we have this kind of internal saying of like, we’re trying to make financial services less annoying. And, you know, documents were like, the first part that was like the most, there was a big cause of bottlenecks behind the scenes. And it was just really annoying for consumers. And as a company, we’re, we’re always on the lookout for things like, Okay, what else is causing these really bad consumer experiences or experiences for businesses? And over the last couple of years, we’ve kept on, you know, asking that question to our customers, you know, what’s causing these bottlenecks? And, of course, document fraud detection was one part. And we’ve since expanded to like, So doc and automation, so doing extracting information from the documents and automating that part as well. But a new kind of problem we’ve been tackling lately is this idea of credit analysis, which is more on the underwriting side. So when you say when you apply for a loan, you are given information, data or analysis to be run to determine what size should a loan be, how much credit should they be given. And this is a problem that we’ve learned that we initially thought was quite automated. But actually, behind the scenes, in a lot of financial service companies, this kind of question of how much should be lent to a particular person or business is actually quite manual. So this is another problem that we’re trying to speed up. And just for some rough numbers here, you can think of, if you apply for a business loan, it’s at a very traditional bank, it could take four weeks could take eight weeks, for them to get back to you. And you know, a big part of that is doing an analysis of the data to understand, you know, what’s your creditworthiness, and that’s the next problem we’re trying to create. And we feel, you know, over the next year, so we’re going to keep on looking and years in the future, keep on looking for his problems, keeping looking for his bottlenecks that will hopefully, you know, anytime in, you know, 510 years time when you’re applying for, you know, any kind of fashion service good use no more of these bottlenecks behind the scenes?
Matt Watson 21:01
Well, and you mentioned, I think you mentioned cabbage earlier, right. So I mean, that’s a huge difference of using something like cabbage versus going to a bank of America or whatever, right, you can go on cabbages website and getting like an instant decision, you know, around your loan, which is so different compared to the traditional way you just described, you know, most of exactly.
Conor Burke 21:23
Yeah, that’s a really good point. And what was, what is interesting, you know, majority of our customers today are actually fintechs. So are the actual most technically savvy customers. But yet, we find that there’s still a surprising amount of manual work going on behind the scenes. So for example, cabbage is one of our customers, and they’ve done using a platform for a number of years, and we’ve been able to, you know, help them automate large pressure for fraud detection on documents.
Matt Watson 21:52
So I see my notes here, it says, you guys use AI for your fraud detection? So I feel like everybody says they use AI. And I feel like most of it is just some kind of algorithm somewhere with a few if statements. So I’m just curious, do you guys really use AI? Did you like to learn a lot about AI over the last five years yourself, like actually building real AI algorithms?
Conor Burke 22:14
Yeah, you know, this kind of started with just a problem domain in general, you know, fraud detection is just really well suited to machine learning. And if you think of the problem of lots of data, there are a few anomalous artifacts within the data training algorithm, or training a model to detect those is a really great, great way to to generalize. So the way we’ve used machine learning to describe it is, first of all, really try to understand what problem you’re trying to build this model to solve. So for example, Instagram, we have a whole host of models that all do various different jobs. So for example, when we first receive a piece of data from a customer, or like I said, a document, we first ask, what is this document, so is this you know, bank statement is a utility bill is a tax document. And, you know, this is a problem. Again, that’s really, this is your quintessential classification problem of, here’s, let’s say, if you get a million documents, classify these into buckets of, you know, into one of these 10 types. And, you know, you could go try with, I will say, an if-else, or basically, you know, piece of tech, you’re probably could get you maybe 20, traditional way, but to get your 99% Plus coverage, or really, really accurate and fast results, machine learning really helps there. And, on the fraud side, too, as well, with things like our trust score. Having the ability to, let’s say, if you have like millions of documents, and a large portion, those are fraudulent. And you know what features you want to look for, to determine if it’s fraudulent or not. And then you have your customers essentially labeled data to be able to say, yet, this is fraudulent, this is not, we can relate our customers feedback to our own fraud detectors to essentially determine which are the most effective fraud detectors, which ones are most likely to actually be fraud. And again, this is another really good challenge, or a really good application of machine learning where you can just large labeled data dataset and just apply a model on top of that. And that’s been, you know, another really effective use for us.
Matt Watson 24:31
A lot of your customers do that, too. If you sign somebody up, like cabbage or whatever, and you’re like, Oh, we’ve done 100,000 loans, would they go back and give you all their history and tell you like, Hey, these are all the all these were not fraud? And these are the ones we know that we processed in the past that were fraud. And then you label all those and feed them in to further train your algorithm. That is usually how that works is part of the onboarding.
Conor Burke 24:55
Yeah. So in the ideal case, that’s, that’s how it happens. And I will say, we don’t require all of our customers to do that. And even though we’re actually from our own perspective, we don’t need all of our customers to do that. There is definitely, you know, diminishing returns for like, for the label data, so as long as you have coverage across all the main use cases, and so on, you can get a pretty good sense of what’s fraudulent across each industry.
Matt Watson 25:24
Yeah, so I’ve never done much with AI. And I’ve always wanted to so I think that’s really cool. I think the classification is a rather simple example. And I think that that makes great sense. I think a lot of people don’t understand that, right? It’s like, I’ve got PDF documents, I gotta figure out what kind of PDF documents they are. So I was curious, what else do you guys do with AI that helps do that? In your processes, you know, different models and stuff? That is it? You know, is it getting into, like their income and their credit report versus what they’re asking for a loan? And like, how do you? What kind of AI models help with that kind of stuff? I’m just kind of curious outside of the classification.
Conor Burke 26:07
Yeah. So on the, let’s say, credit side or income side, we don’t actually ask customers for, you know, what was the income that was, you asked for, let’s say, but what we do instead is try to tell our customers what income this particular applicant was. So let’s say in the case of a bank statement, what we do first here is, let’s say you just imagine you’re getting a bank statement. And you’d extract all the information from it. So what we’ve done there is we’ve developed a few models that look for things like, let’s say, the opening balance, and the balance and all the transactions. So you can imagine for a human it’s a read of XM, that’s quite easy. You can see, here’s a date, here’s the description, here’s the amount. But to train a machine learning model due to the sound like all bank statements in the United States, across all formats automatically within seconds, is really, really difficult. And this has been a really great application. And last, I’d say maybe two years, there’s just been this proliferation of, I guess, new models that have made this now possible, actually even say, two years ago, for features we developed, the instructors wouldn’t have been possible because the technology and the knowledge just wasn’t there. So models, or transformer based models, big natural language models that have been, you know, released by the public or like, I guess, research labs, like Facebook, or Facebook, or Google or open AI, I really, I guess, enabled, I guess, new generation of models that now can do these tasks, like extract the name, description, balance, outre bank, Sandra automatically. And yeah, in terms of, you know, other models, then we also have Yelich ways to, let’s say, determine if you get it, let’s say a business bank statement. And you’ve always like debits and our incomes coming in, how do you determine actually, which is revenue worksheet versus, which is I’d say, just like other money coming into business, and that’s, again, a good challenge for a machine learning model, or some kind of algorithm to determine that. And again, with all these challenges, what’s really interesting is that humans are really good at this stuff. And we’re often compared against human performance. But what we really want to do is actually get to a superhuman level. So for us to reach the human level, being a superhuman level. And that’s really the challenge our engineering team has, you know, week in week out is, how do we get the right data? How do we get the right models? And how can we keep on pushing ourselves to get better than this human level performance?
Matt Watson 28:44
So some of those AI models seem like things that a lot of people could take advantage of, just like, you know, analyzing a picture, and it tells you like, what’s in the picture? And, you know, things like, Amazon Web Services, and Azure now that provide different AI models that kind of do some of that stuff out of the box, right? Give it a picture and tell you, Oh, there’s a chair in it, or there’s a car in it, or there’s a human in it or whatever. Right? But so for some of the stuff you’re describing, is any of that off the shelf, like, oh, I can give it a document, it’ll tell me the type of document or it’ll know what’s a bank statement? And we’ll automatically read that out, or do you guys have to like invent all of that?
Conor Burke 29:21
Yeah, we’ve definitely had some open source frameworks. But yeah, all the models we’ve developed are built internally. And this is really based on just a requirement for high performance from our customers, we often find that a lot of off the shelf models, just they just don’t get that level of accuracy. So maybe, you know, 50-60% accurate, but our customers require much higher levels of 9599 and so on. And yeah, that’s just required us to build an internally specialized sculptor problem better as well. That’s often the thing that a lot of official platforms do. struggle with is like trying to scope it to a particular use case and It just is understandably a much harder problem if you don’t have some of those restrictions. But yet in Australia, we’ve got a limited scope of some of our models to contain the problem, and that has allowed us to get much higher performance rates.
Matt Watson 30:14
So is your guys’ business primarily in the US? Or do you have a lot of customers outside the United States?
Conor Burke 30:21
It’s primarily the US, but I’d say you know, 20, about 20% for customer bases outside the US. So it is definitely a global problem. And we are, you know, excited about helping customers elsewhere in the world.
Matt Watson 30:35
Well, most people don’t really think about this, but there are other countries around the world where there’s no, not even such a thing as a credit report. You know, our company Full Scale, we have like 300 employees in the Philippines. And there’s no such thing as a credit report where you can get somebody’s social security number and even check a credit report. And like, I think like 70% of the population there doesn’t even have a bank account. Right. So there’s, I can only imagine doing fraud detection there would have to be so different. And so, so much more complicated to do, compared to here, like, hey, at least you have credit reports and other a lot of online databases you can get information from.
Conor Burke 31:16
Yeah, exactly. I think financial services is particularly unique in that. There are such differences and contrasts between countries in terms of regulation, consumer habits, technology, culture, even credit courses, a great example. Even with credit cards, I say, you know, I’m from Ireland, and the idea of credit cards just isn’t really a big deal. I personally have never had a credit card or taken out a loan. So even though I’m in FinTech, and yeah, I think that cultural differences are interesting, even just between, you know, Ireland, us. But yeah, if you go further afield to, you know, Asia and Africa, there’s definitely, like different cultural differences there as well.
Matt Watson 32:02
So you said in Ireland, a lot of people don’t use credit cards, or they only use credit cards?
Conor Burke 32:06
Don’t. So it’s not very common. There is definitely, you know, definitely credit cards in Ireland, but it’s just not like a big cultural thing. So, and if, you know, for people who do have credit cards are often, you know, paid off pretty quickly, and they don’t, you know, keep a balanced running, etc.
Matt Watson 32:24
So they just use debit cards instead, or, or they choose cash, debit cards, debit cards, okay. Okay. Well, so what do you see as the future for you guys and your company? Where do you see this going over the next few years? Are you? Where do you see this going?
Conor Burke 32:43
Yeah, like I was mentioning earlier, I think the big opportunity here is, how can we? How can we create how financial service companies create these great customer experiences? So you can think of all these companies like let’s say, chime and Coinbase. And even on the business side, like cabbage, really providing excellent experiences from, let’s say the front end of production services to stack, or we’re really interested in is like front and back end? How can we enable some of these great customer experiences, and tackle these bottlenecks behind the scenes. So for the next couple of years, we’re super excited about continuing to help our current customer base, just solve these bottlenecks, but also expand to slightly adjacent markets, and expand geographically as well. So we’re seeing, you know, I guess, like three axis of growth for us, you know, one across like, the features, like adding more functionality, but within fraud detection, but also outside of fraud detection, to tackle these bottlenecks, and then to also some industry, so we’ve been primarily focused on SMB lending and personal lending. We’re also here we’re looking to go much deeper in areas like, you know, payments, crypto and, and so on. And then lastly, it was kind of hinted that they are primarily in the United States. But yeah, we see massive opportunities in Europe as well.
Matt Watson 34:06
So you mentioned crypto and Coinbase and stuff. So do you guys also get involved with KYC like know, your customer kind of stuff?
Conor Burke 34:15
Yep, exactly. So there’s contrary primary use cases. One as we’ve been talking about is like underwriting, but yeah, KYC and Kyp is the other big use case. So we do know, many customers using actually, actually for both, so both KYC, Kelby, and underwriting are able to do many customers just using us for KYC and Kyp as well. So you can think here you know that’s a proof of address documents, proof of business ownership. Anytime you’re really applying documents to a financial services company we can usually help out there.
Matt Watson 34:44
Okay, well, once again, this episode of Startup Hustle was sponsored by our friends over at Equip-Bid Auctions, join, sell and earn. It’s that easy with Equip-Bid Auctions. Become an affiliate and start or grow your independent business by visiting equip, dash bid dot Me slash startup today. Even easier to Startup Hustle dot XYZ and click on the partner’s page. Look for a quick bid, and everything you need to start your own business is there and ready to go? Well, as we round out the show today, I was curious, do you have any other final tips or words of wisdom for listeners that don’t have to be about fraud but just about, you know, being an entrepreneur and business or, or anything you’d love to share?
Conor Burke 35:30
Yeah, that’s a good question. I wanted to have always come back to, and I often say to our team, as well as really try to build great relations, relationships with your, with your users and your customers, you know, they have so many insights that they that love to share and you know, if you’re able to help them in any way, I think that’s just a great way. A great way of building business. And that’s something we’ve really kind of cherished Inscribe as, like, trying to really get to know our users. And that can make you know a lot of other parts of your business so much easier. And yeah, that’s kind of like one tip. I definitely recommend it to our listeners. I definitely appreciate it’s a little bit cliched. But if you really take it to heart, it can go a long way.
Matt Watson 36:14
Well, honestly, I think a lot of companies struggle with that where they don’t actually spend enough time talking to their customers and their users to really understand their problems. And they, they just act as they know them, but they don’t really truly understand them, you know?
Conor Burke 36:29
Yeah, you know, I think complacency when you, especially if you initially understand the problem you’re solving. Complacency can quickly creep in. So yeah, constantly fighting against, like discovering new problems your customers have discovered, you know, new insights they have is super important.
Matt Watson 36:47
All right. Well, thank you so much for being on the show today. Again, this was Conor Burke with Inscribe, and if you want to learn more about Inscribe, you can go to Inscribe.ai. That’s Inscibe.ai. Well, Conor, thank you so much for being on the show today.
Conor Burke 37:06
Thanks for having me. It was fun. All right. Take care.