Carl Spilker and Matt Shubert Explore AI’s Impact on Finance: Data, Fraud, and Market Adaptation

SPEAKERS

Matt Shubert, Matt Tepper, Rich Alterman, Carl Spilker

Rich Alterman  00:00
You're syncing up and tuning in to the lending link podcast, powered by GDS Link. With a modern day lender can dive deeper into the future of data decisioning and Credit Risk Solutions.

Matt Tepper  00:18
Welcome to The Lending Link, the podcast that connects you with the latest insights in the world of lending. I'm your guest host today Matt Tepper, Vice President of Global Marketing here at GDS Link, filling in for our regular hosts Rich Alterman, And today's special episode we have the privilege of bringing together two exceptional guests who are experts in their fields. First we have Carl Spilker, the Executive Vice President of Global Analytics Advisory at GDS Link, Carl leads our team of talented data scientists who specialize in developing and deploying custom models scorecards and attribute sets as part of GDS Link analytics offering Modelica Pro. With extensive experience in senior leadership roles for prominent companies like Ezcorp, Provident Financial Group and Experian Scorex, Carl brings a wealth of global expertise to the table. He holds a bachelor's degree from the University of Memphis and a Global MBA from Trium awarded jointly through NYU, the London School of Economics and Political Science, and the HEC Paris School of Management. Also joining us today is Matt Shubert, the Vice President of Predictive Analytics and Data Science at West Coast University. Matt's expertise lies in transforming data into advanced analytics solutions that drive growth, enhance student success and improve operational excellence. With an impressive background that includes leadership positions at Experian, Union Bank, SAS Institute and Wells Fargo, Matt has a proven track record in developing and implementing cutting edge analytic products, including machine learning and artificial intelligence solutions. Matt holds a MBA and Masters of Science and finance from Indiana University, as well as a bachelor of arts and economics and econometrics from University of San Diego. Welcome, Carl. Welcome, Matt. In today's episode, we're diving into the revolution AI is bringing to lending practices from fraud detection to data driven risk decisions, AI is reshaping the lending landscape and financial institutions are embracing its potential to optimize their operations. But before we start, join me as we turned back the clock about seven months ago, to November 30th, 2022, the US defeated Iran to advance to the knockout stage of the 2022 World Cup, the entire world was becoming familiar with a gentleman named Sam Bankman-fried, Fried and his company FTX, and Open AI at 375 person organization out of Silicon Valley launched a free preview of its new chatbot called Chat GPT. Now, you'd be hard pressed to talk to anyone who has not heard or tried Chat GPT. And the ripple effect has already up ended education, music and company share prices. I think the term AI has fallen victim to becoming a bit of a marketing buzzword. So perhaps let's level set a bit and I'll start with you, Matt, maybe you can speak to really the true definition of AI and speak to the different types of AI that folks might be familiar with.

Matt Shubert  03:12
Yeah, I think when we look at it, it's such a wide range of applications and things we've seen over time. And obviously church be really is made much more disappointed with people rigid agenda of AI. And that almost replacing with some of the human minds may able to do but on a quantum scale versus what a single brain may be able to do on sound. And it's amazing though, there's other sub applications we've seen over time, and almost that blend from what we call predictive to machine learning, artificial intelligence, kind of that navigation to that journey to get to be replicating what the brain does almost like their lead on developing neural networks and how that has been applied over time and evolved and changed. And what that's turned into some of the more recent releases and improvements have been made, obviously, a neural nets trying to copy similar what the brain is doing, but without having everything visible to everybody. And that's probably some of the fears things that go on with an AI is, is it really, do we understand it? Do we know what it's going to do is the machines going to take us over and it's not that far fetched for people to think sometimes maybe it is the Terminator II with glowing eyes. But I think some of that's a little bit over fear just due to the lack of knowledge and acceptance of change and how these things are really embedded in our everyday life as well.

Matt Tepper  04:19
And Carl, maybe you might be able to glean a little bit, a lot of the advancement that we've seen in AI, I think it all kind of also ties a lot back to the type of code and data that AI ingests that kind of plays a little bit of a big role. Correct.

Carl Spilker  04:33
Really good point that and we've actually taken what Chat GPT does, it's a very sophisticated process. And for the lending side, it's it's really, really dumbed down the benefits of leveraging AI, what are some of the mathematical components that are used inside of it? And we've been able to strip those out down to a lending piece. The other thing that's advanced in our industry is the amount of high velocity huge amounts of transactional data. And that is really where if you kind of think about let's take out a tiny piece of the AI methodology, and let's use it to help discern a better predictive model with high transactional data, which is very difficult to do with a traditional regression techniques on not impossible, but much easier. And what we're finding is it's really starting to squeeze out really large values of benefits. And I'm sort of eager towards where we'll step into what happens if we let this tool go in a true AI sense of living.

Matt Shubert  05:38
Yeah it really has to do with it can build attributes and features and that you think about the transactional data, it really is taking the series of transactions over a certain period of time, and how do you turn that into some kind of insight and get that signal or sub signal within there, and it'd be it'd be amazing in the future, that's that time to devote that time that speed to getting to those insights can be augmented and supplemented with some of those, whether the Chat GPT or whatever comes from from a generative sense.

Carl Spilker  06:02
Exactly, he can kind of tease out the things, the nuances, and go down another layer and see some things that you don't see, that does find a few segments that provides additional lift. And from a lending perspective, regardless, if you're super prime, prime, subprime, it's able to tease out some incremental value there, which from a lending perspective, puts money in your pocket and serves more customers in a positive way.

Matt Shubert  06:31
Yeah basis point account accounts have a permanent basis points can make a massive difference in lending. And that's where you get those mentioned sub signals. That's where you really can start capturing some of those. And you mentioned segmentation. And that's all I'll say history to do is just trying to get better predictive lift for those different breakout segments. And some of the models that have built my past and lending would have up to 13/14 different segments. And each of those has its own amount of predictive signal. And what machine learning allows you to do is to get the maximum kind of pro value out of that whether it's X GB or neural nets or whatever application is being used. And I think the speed of development getting there is probably the the biggest hindrance, that's especially these tools that might be able to help with some of that, in some ways. Almost quantum computing is its own layer. But that's really what's the beef and the meat compute that's behind this is really more of a quantum layer that it was a quantum peperonata symptom, the trillions of amplitude that occurs versus the what we can do on a local machine.

Matt Tepper  07:27
And mentally as a child of the 90s, I vividly remember a little bit of the groundswell of concern surrounding the early adoption of the internet, many folks were skeptical and hesitant to plug in their credit card information online. However, over time, this was cited as the internet brought forth valuable benefits that have been integrated into our everyday lives. I think it's without saying we find ourselves in a similar scenario, guys, we're kind of talking a little bit to some of the concerns kind of out there. So Matt, let's start with you. Why should we not be your AI? What reassurances Do you think you can provide that might alleviate some of these concerns about this tech.

Matt Shubert  08:01
Well, you mentioned the internet, internet's great example, ATMs is another great example, I can think back to my grandparents generation that was anti ATM, refused even to have a debit card kind of thing. Because of that fear that the machine was gonna wasn't gonna give them the money they're asking for like the teller inside would, is I think somebody says that building that trust piecing people comfortable, some of the change and seeing things are different. Obviously, I've heard with AI a way to make it more trustworthy is to put a human face on it. Something we do at West Coast University is that we have a chatbot. And I'm not gonna say that chatbots, the most amazing thing that was ever built, but it has very humanistic qualities sounds human looks human, it has facial change reactions and things. And if you could put something with the string of a maybe a Chat GPT or something of that power behind it, it can be that much more powerful, but you make it look safe, and make people feel safe and secure. So I think it's more of a fear of the unknown versus an actual, maybe justifiable fear, in a sense. Now, who knows. And there's some folks out there as well, something's been released in the last week, or some of the folks involved with some of these more generative tools are saying, Oh, we must fear we must worry whiskey regulation around it. I kind of see that. Some It's trust, and then some it's just really understand what it really consists of today. And it's probably not as intricate and advanced as what people think I mean, Chat GPT, I to hate to break it down. But there's a little bit of split. Thomas didn't want that. And so my team is now that it's not that different than what an index view is things are, it's just delivering it in a more concise, direct manner and makes it seem more humanistic. And he read the responses you get from it, it's not that different than what you would get from a humanized chatbot a little more sophisticated for sure. I don't want to downplay it at all but there's a making making people feel safe with and that's probably when people are people like it right because it seems like I was talking to somebody where you're building a relationship with that person. That's why it's what it's doing. I think they're gonna be controls and limits and things in place. You can make sure it doesn't get out of good I guess it doesn't get out of control like people might be fearful worker.

Matt Tepper  09:54
Yeah, I think to your point, and there's definitely a lot of fear of the unknown fear of just kind of where this could potentially head without necessarily really just kind of looking at what it is in the moment and maybe kind of seeing where some of the benefits might lie. So kind of building off that earlier analogy, just surrounding the adoption of the internet, we've seen how embracing technology can bring forth a multitude of benefits. Obviously, digital only businesses emerged after adoption, the internet, internet communication possibilities obviously expanded exponentially, the sharing of information resources all became more accessible than ever before. So I might pose to you Carl's we explore the potential of AI, it's essential to understand the value that it currently yields. So in your experience observations, you kind of were talking a little bit of being able to really kind of start understanding some of the nuance and some of the some of the data that you're utilizing, what other sort of value do you feel like aI brings to the table? How do you kind of see it enhancing business operations, decision making processes.

Carl Spilker  10:52
All kinds of leverage off what Matt Shubert mentioned, it's one part trust. But it's also one part replacing a task that you have done under times that you don't need to do 100 times again. So I kind of think of right now, it's really replacing things that we've already seen in the past that we know we're going to do it. Part of the modeling process is you know, it's going to review the trends we've seen in the past and tried to help you replicate and rank water, what you're going to do in the future in a positive way, and give you the deep, that's a piece of that I look at as being a positive outcome from this. And so the word I'm kind of thinking of is we can do more lazier things so we can do more productive things. And this tool will replace a lot of the things we've done. And I truly think just in across the globe, people are going to replace who aren't really productive. It'll take away the nuance. And over time that will make us trusted, every model has a level mistakes. Ideally, you would like to minimize that which at the end of the day are no different than what humans made, and will begin to trust it, which kind of echoing back to what Matt said, it's fine, that level of trust with the right balance of positive and less mistakes. So I've seen some things here that it it's definitely driving incremental value. It's making life dramatically more efficient. But it will offer some things that just don't make sense. That's a piece that just takes more data, more learnings and deeper dive into branders basics laid out more detail.

Matt Shubert  12:27
I have a good example I can share with it from a lending actually model that I worked on the previous employer, where we were building out a predictive model, they say, Okay, what's the likelihood of default within 12 months of origination of software can be any type of lending product. And it was hit around businesses. And when, for example, it was scoring Google and some other ones extremely low, which you would go okay, how can you score that business low? Well, it's because they just might not be the fastest payer, they might not pay their bills as quickly as somebody because they don't have to kind of thing they can wait 30 days, or they have a 60 day clause in their contract, because they're able to basically create those corner cases. And I think the same thing applies with AI in the sense where you have to identify this cork is going to be and then try to put controls limitations around it. I mean, back at bad actors are gonna be hard to remove from the use of some of these tools. But if you're gonna demo this cork is just like we're trying to create the those overlays other controls in place to kind of offset the to keep your bill to be used in ways that maybe it wasn't intended.

Matt Tepper  13:23
Yeah, in preparation for, for this recording, I was reading an article that kind of reminded me that, again, kind of going back to really the early adoption of the internet, that it was really kind of more folks really playing around with the technology that really led to the most, you know, it'll be greatest innovation, the fact that it started as a kid in a dorm room, just trying to rate the attractiveness of coeds, led to obviously Facebook, another guy wanted to just see if he could buy a book out of a bookstore across the country. And now we got Amazon. So I think there's certainly something to be said about, that the malicious actors going to maybe kind of get a little snuffed out, and that the true innovation is maybe yet to be seen, it'll probably just come from really folks just kind of embracing this and really kind of playing around and seeing what's possible out of it. That said, Carl, you kind of started speaking to it. And Matt, I know you have some experience on this as well. So maybe I'll let you answer first. But what might be some of the best practices for organizations that are looking to deliver, build deploy AI driven models.

Matt Shubert  14:23
I think it's really making sure you understand the problem you're trying to address. And that'll give you a lot more visibility to where you're trying to apply the tools. And help me quickly more quickly identify what those corner cases are before they create a adverse outcome type of thing. And that probably fits really well with lending in that aspect that adverse outcomes could be either you're denying people who shouldn't be denied or maybe you're approved, people shouldn't be approved or letting fraudulent folks come through the process. And I think the better you understand your operational process and where it's going to apply, where it's gonna fit, and then maybe some of those manual or less automated process that may replace, it's gonna give you a better sense of where they need to have the right controls in place and then monitoring and visibility. to early signs of either scenarios that are breaking a record cases, unable to adapt to this current exists, or even where it might have a weak point of prediction. Carmen should rank order directional kind of accuracy is great. But that doesn't really help when you're denying somebody who's trying to get credit to go fund your small business to on their child's education to learn those types of things. And I think better you have an understanding of where it applies and is being fit into your operational process within your organization, the better chance you're gonna have to make sure you have the right controls and things in place to make sure minimal adverse impact comes from the application.

Matt Tepper  15:37
Before our call, you sent over a note in preparation for today. I really liked that sprinkler system analogy that you kind of tossed ID, can you maybe kind of walk our audience through that a little .

Matt Shubert  15:45
Yeah, I wonder if is sprinkler sprinkler time I spent a lot of time with sprinkler system, I moved in that house that had little two foot carve outs about four foot long and two foot away from the house with sprinkler heads on it that would squirt five or 10 feet. So either I'm spraying my house with a whole bunch of water, I'm spraying my paper with a whole bunch of water next to it that doesn't need water and my kinda inherited a mess. And as I go through my trying to figure out what am I gonna do I convert these over a drip in, well, I figured, hey, I'm gonna convert all these things over and put drip. And it's almost like an AI system or a machine learning system where you say I'm going to come in and do this one thing, it's got to fix everything. But if I put a pressure reducer below my sprinkler valves, I'm going to have a problem on unequal pressure on each of the pipe valves. So ultimately, I started out with starting with heads and on top of hoses, and I'm like, okay, reducing, you're reducing their short term fix that worked well. But once you start realizing I'm getting too much water, some places not enough others start having valves going bad bill, the ones didn't have reducers. And I realized I just put a pressure reducer on the main line. For me, it's a system where you have the controls in place, and you can control the getting the right amount of water not too much, not too little right place easily to judge it over time to see for grass turns green. Well, you can do the same thing with when you're say making credit decisions as you're going through. You monitor Am I seeing my denials or approvals going up or down? Am I seeing a population that maybe I used to originate and be able to get booked as a new loan or a new new lending product. And then now I'm not seeing this come through. And it's having that right control set place when I mentioned operational processes, actually, that's what I mean, when we like to go set the timer in the garage and sprinkler system and everything, set it forget it, it runs twice a day for two minutes or run on every cycle and everything works perfect and looks green and lush. And if you just try to leave it like that, that was your seasons change throughout the year and seasonality starts to take effect. I mean, you're gonna be flooding your grass for no reason in the winter, because you get lots of rain. So I think want to try that ties back to AI and applications and these advanced techniques, you have to think about your entire system. And that's what I mean operation process flow is what's going on what's the internal factors, external factors, macro factors, micro effects, you have to take into account to make sure things are controlled and set in a way that they can ultimately turn into pretty flowers in the spring and green grass in the summer throughout the year as much as possible. And not worrying about anything being over under watered along the process. And I think when you look at AI, it's the same same type of mechanism, you have to really understand where it's fitting and enough about what it's doing versus EPA know every single thing is doing internal if you have eyes on something, and you can keep it from doing things you maybe don't intended to do. And that's obviously what can impact the business outcomes if you don't have the right controls in place.

Matt Tepper  18:12
Yeah, Carl, I know that one of the services that we offer here through our analytics offering is model monitoring, which I know can really help simplify the model governance side of things, any other color, kind of want to add on to some of the best practices for organizations.

Carl Spilker  18:29
I was thinking about the legacy models, you built them for stability, you're happy, they weren't razor sharp, but they're razor sharp, they told very quickly, that they're sharp, and they worked and they last for a long time. This way, we kind of think with the AI technology, you can actually create a razor sharp utensil to segment much deeper into the pool. So then you think about what is my infrastructure? Can I execute on this very efficiently with AI technology? That's where I believe our software can handle that. And then number two is how quickly can you adapt to the changes because the sharper things are, the more likely they're gonna get Braille and break down and need some retuning rebuilding. And so our ability to kind of reshape the model, the monitoring, say, hey, look, things are starting to drift. The universe that I bring in through the door has started to change, how quickly could I adapt to it? Then more importantly, how quickly can I redeploy my new change into the environment? So I'd say we're the early days where we're using supervised intervention to reach and redevelop we're using a better platform to be more efficient to deploy and execute. And to me that means that 2024-2025 We can find out models drifting, shifting, return or rebuild or redeploy, instead of monster they turn into AIDS, it's going to be pretty clear that after a point of time, we just turn on the machine that turns into a true AI. It's a Moving Interactive model, it's learning and adjusting to the, to the market, that condition similar to the sprinkler analogy is going to start adjusting. And our ability to kind of step in as an overseer, which still require some human intervention, that makes them monitor modulating adjustments will make that product significantly better. So from a lending perspective, there's a lot of regulatory, describe what's going on, give people feedback, that's gonna start becoming very simple, uncovered explainable. And we're really close within a couple, two to five years.

Matt Tepper  20:56
So I know that we've all heard the jokes about our phones, or our smart TVs, suddenly listening to our conversations displaying relevant ads, which obviously is an example of the presence of AI fueled algorithms in our daily lives. Although I think you both can speak to the fact that AI's reach bar extends just beyond just targeted advertising. So Carl, I might pose this to you, where does AI exist in our everyday lives? That might surprise us? Are there any lesser known applications or areas where AI is kind of is quietly working behind the scenes?

Carl Spilker  21:29
Yeah, I think it's going to be on pattern recognition, using some of the natural language processing items that come from it. So when they start associating events in your life, and you get presented with a solution, I think that's going to be a bit of the I would not be so crazy, your tire being blown, and the system tells you somebody is on its way, we'll be there in five minutes, probably pretty darn close today. Whereas now I was growing up, she's, you're lucky to put the car to the side, stick a sign out and walk five miles to a gas station, totally different. And even five years ago, he called AAA somebody like, now it's going to actually see, understand that a bit make start those processes, which is kind of going back to my earlier event, it's going to take the processes that we've seen and have been repeated and make them efficient for us and take that trouble away. So I see a lot of so then I put it onto the lending platform. I had to consciously think gosh, you know, Matt is going to lend me money at one 1% interest and trusting I'm going to invest in so and so stop for 10% out easily arbitrage. You know, you can see the future gone. Let's puts 10,000 in this account, and it finds its way to do that for you. The arbitrage is those that would be kind of really a cool future state. My daughter was complaining about insurance, when it's insurance gonna tell you that they've actually met the peak of being the most efficient driver ever give you the savings until you reach out to you it'd be nice to have somebody adjust that.

Matt Shubert  23:13
We think about how like loan aggregators work and what if when you're shopping and it could be a mortgage or new insurance guarantee they handle two and what if something could deliver to you like you pushed offers to you now from individuals warps? Can we say here's how you get the best deal? Hey, you're up, ruin knows you're up for renewal. Some insurance companies do that, too. They'll send you that hey, yup, right. I think that kind of time when they think you're up for renewal and try to advertise to you. But what if somebody did it on your behalf and kind of said, Okay, now here's the competitive and almost negotiates for you in a way. Here's the offers this person getting for a comp policy, who wants to give them the best offer who's most assured and getting that new business type of thing. And, like I like to shop and barter and things like that for annual expenses to car to give me these lows? Boscov's kind of thing. And I think it'd be great to have a tool to help me with that. Almost like one of the banks has a shopping assistant you can download on your browser that tells you if a product is cheaper somewhere else. So that's an example. It's that's really just AI at work and away on going and doing that for you. I know you mentioned that listen to headphones. And I always tell them my wife's shopping for something, or maybe clothes or something. And then I started getting strollers popping up in my recommendations from Amazon to go in there and make a purchase type of thing.

Matt Tepper  24:20
So shifting gears a little bit, Matt, I was hoping you might be able to kind of speak to how exactly AI is transforming lending industry maybe what are some advantages financial institutions and borrowers can expect when utilizing AI and specifically within lending operations? Are there specific improvements or efficiencies that stand out?

Matt Shubert  24:39
Yeah, I think some are doing I've been doing it for few years, or handful years to improve the document collection process, underwriting processes. Some of those things are very manual where you might have to spend different trips to the bank or different trips to make that happen. While it's kind of lower electronic approach. I think there's still some opportunity there. There's also all the behind Back Office efforts that are happening in your in your letters of origination of the loan collections. I'm always kind of things that are involved and getting them quicker to decision, I think you're seeing a lot of banks going into actually fits the fair lending requirements that have been a little bit better by using a predictive model that you can explain and have accepted by your internal external regulators. And being able to say this model is being used as sort of things where they like, oh, well, this person's a lawyer, let's give them a great loan. Okay, because then we got job stability. Has that changed over time? 30 years ago, that was pretty pretty common, and are one of the methods I was taught back. And I when I first got into credit risk, the director ERISA workforce said one thing you learned above all else case, credibility, stability and equity, do you understand case you understand how to write credit response? And it was interesting, as I move for that, then moving the modeling team and I said, Well, gosh, how do you capture occation case into a predictable? Assignment? How do you get more of that, beyond just the credit components to get into the ability stability, equity pieces? How do you get those data elements and I think it's really more advanced techniques can help augment some of that and make it a little bit more predictive. That's all really all an X GB approach is doing is basically taking that gradient descent. And going from a 2D view, almost like a 3D view prediction. And let you get splices amongst those 3D angles versus at least a 2D angle of it. And then if you segment you can get segments of those different 3D splices or dices that are happening, and to get that sharp razor's edge, like Carl mentioned earlier, that knife is just your SUV. And those things just give me another tool to get deeper and more precise, almost like a surgeon does when they're performing a really integrated brain surgery to remove an aneurysm. I personally still think it change the way lending is approach and probably make it more inclusive, more equitable, be able to make sure those folks are getting credit that deserve credit across some of the maybe more marginalized groups as well. Because there's a lot of those are probably going to turn into approvals rather than denials by having these are layers of angle and consideration versus somebody back 3034 years ago, we go on to make branching talk to a loan officer and pay your loan out great, okay, great, you're making this you get this type of job looks like you've paid your bills before approved, open for bias versus the machine actually, likely has less bias than the human. Because that human might not just be looking at the application, they might be looking at the person they might be looking at the way they're dressed, the type of car they driven, and all kinds of other factors that she isn't gonna look at, aside from she's gonna say, Did you pay your bills on time looks like you have an acceptable income looks like he has steady employment. Looks like you're doing all these things, right? And take out a while that bias that might be inherently tossed in by humans. So I think people fear AI, but AI might make things more fair for the average folks out there as well.

Matt Tepper  27:37
Yeah, and implementing AI in any industry is going to come with its own set of hurdles. Carl, when it comes to lending, what are some of the challenges that arise when incorporating AI into decision making processes risk assessments, fraud detection? How do we navigate these potential obstacles while maximizing the potential of AI?

Carl Spilker  27:56
The first couple of regulatory that I think, frankly, we could probably haven't thought through by itself, but that's literally an error you want to be open and concern to? The other one is explainability. People need to be told why, why was this event occurring, so they can rectify it, it's something unusual. And that's probably one of the bigger challenges because given the sophistication, the deeper layers you go, it can be kind of an unusual a reason why you were denied, or also the usual why you were approved, on the inclusivity side, that's going to be one of the things that will start bubbling, today is going to be the movement, in the old days of FICO score could stick with you the same rage for a right. And now your score can actually move quite a bit over time. And that's part of the plus. But that also turns into a bit of the negative on the predictive utility, these types of things. Just basically summarize math, the ability to explain, I think it's gonna be the big one, explaining from a regulatory standpoint, but it's also explaining from is lending is going to ask a lot of people that drop hundreds of millions of dollars into something, they're gonna want a good explanation and a good track record of history. I think we're at a point, leveraging AI, typically, traditionally leveraging AI methodologies has been in place for 10 years. It's not old tech from a lending perspective. It's old tech, because nothing it's truly AI where it's alive and executing in real time. But I believe, as I mentioned earlier with it a couple of years that that learning engine is going to be let loose. And the minute you could explain why I believe it's going to make that Pandora's box opened a bit. You kind of touched on something it was kind of leading to my next question in terms of now if we're able to get a little bit more transparency into credit scoring. What sort of specific techniques and methodologies Do you think AI brings to the table that can lead to more accurate and insightful credit scoring? Two things, opening up the parameter space of which it's selecting deeper insights into consumers behavior, historically. The second one is the market is adding more variables into that consumer lending decision. And as adding huge amount of transactional type of detail, picking one, open banking, for example, we've taken credit data, it's someone's file that's reported by loads of other institutions, then augmented with your daily transactional banking behavior, cash in cash out stuff of that nature, the benefits of these new techniques is it's able to tease out and combine those features to drive better insights. And that common very broad way opens things up in a inclusive way. It opens things up in a better predictive way, gives people more it gives some less, but it gives the lender the guy who's investing in that outcome, a much better tool.

Matt Tepper  31:12
Obviously, fraud detection and prevention are crucial components of the lending industry. Matt, how does AI contribute to these areas, particularly in detecting and mitigating fraudulent activities? Likewise, you could provide us any sort of real world examples, you've kind of seen on successful implementations that arena would be curious to hear your thoughts.

Matt Shubert  31:32
Yep, I work with lending and lending fraud solutions for probably often on six, seven years now. And fraud z is very unique use case. And what you're trying to do is really minimize minimal impact your your good population and capture as much of your bad population and a really small percentage. And you probably can do a cost benefit to say, Okay, if 50% of this 1% of my overall population is going to be fraudulent, that's easy decision right, one out of every two is going to be a fraudster. So you're not, you're not causing that much pain for the rest of the that the 80 or 90% have a 1% or 2% population to say these people just get either completely rejected or maybe the quarantine for a manual review, to try and weed out some of the true fosters false positives, true positives kind of thing, to win some of those out. And it's the idea that what more advanced techniques allow you to do let you take in a lot more factors and elements and promise some traditional predictive modeling techniques is your limit on number of variables, just because there's some rules you have to follow, and you can't break those rules or the simplest more approaches won't work. So that's nice thing with like, say more advanced methods in machine learning neural nets, X GB whichever, whichever you choose to use support vector machines, look at these different angles, it's you can put a lot more complex segmentation on those, you can make them explainable, a lot of ways even fraud, you still cannot explain like why this person is being denied. You don't want to tell the fraudster you're being denied because your fraudster not though they know they've been caught. So they make their profile look different next time they come through and not be classified as frauster. The worst thing you do is tell fraudster, they're fraudster it's more to say, oh, no, here's declined. And here's some reasons why. And there's a reason the profile and it makes a little bit harder for them to create that profile that sneaks through outside of that one or 2% population. So I think summarize, it's really just getting that more, you're gonna have a more accurate capture of this frog project populations, and probably be able to give them less targeted reason why they're being denied. So they can't try to sneak back to the door and other way. Because a fraudster will keep attempting to come find that break that open window, that open door, that open problems process they can sneak through, and then get through what they want to do. Because once they're in, it's really tough to capture them. Once they're in, they look like everybody else and they're dressed same, they're, they look the same, they talk the same. So you got to catch them before they get inside the inside of the house or inside the lending institution, I guess for origination and funding and all those kinds of things. So it's having that that kind of capture that improvement in accuracy, as well as the generic approach to how you're explaining why they are maybe not getting the approval they're expecting, or the experience of processors when they get to talk to human being. So the more he forces in the manual review, you can almost see that accuracy allow you to force more of them to be kind of flag and say maybe this isn't the bank I'm gonna pick on I'm gonna go pick up the one who doesn't have a manual process in place because of extra controls and put some controls in, make sure you minimize your false positives and maximize your true positives.

Matt Tepper  34:22
Matt, I'll stay with you on this one, as you're kind of walking through that scenario kind of seems that obviously fraud mitigation and financial inclusion are a little bit of the ying and the yang. There's you're kind of walking a little bit of a fine line between the two. So while you're obviously putting in methodologies and processes to kind of catch these bad actors, how is AI kind of working on the other side of the coin? How is it helping contribute to the expansion of the access to credit and promoting financial inclusion?

Matt Shubert  34:49
I think it lets you separate those truth rosters from false fraudsters is kind of easy way of putting it you could just use a year and have the ability to see see more work with more and especially if they could find a way to make it generative AI, I almost think like an underwriter. I don't know that exists today, maybe some of the banks are working towards I know some of the large banks have announced they're getting their own generative pools. And there may be a way to build that, with these more advanced techniques, it really comes down to compute. And one thing that enables is is is having access to ease of access to quantum compute to mass stores of data, the more signals and elements you can use to make your predictions, the more valuable you're going to have to be able to kind of separate the the true goods from the true Bad's and those type of scenarios and it does become more inclusive, because then you're not, you're not denying you're not having that false denial that they should have been approved, but they looked a little bit like they were a potential fraudster coming through the process. It ties back to accuracy. Like I mentioned, if you can go from a 50% capture and a 1% population to an 80% capture. That's a huge difference. You're talking for bad cert for true fraudsters, every one false fraudster type of thing. I think it ties back into bill to make those better for the overall group group and more inclusive, as you mentioned, give people an opportunity and not have them quarantined off in a group, maybe they shouldn't be quarantined off the first bills. Maybe our last financial related kind of getting a little granular here, Carl, loan portfolio management, predictive analytics obviously play a vital role in shaping lending strategies for any organization. Considering the impact of AI, how does it transform this approach? As pertains to loan portfolio management and predictive analytics.

Carl Spilker  36:24
Everything we've encapsulated here is ripe for benefit on the low portfolio, the account management side, it ends up being how easily can your platform consume more, if not frequent, vast amounts of data? And then how does your analytics your AI infrastructure, read and consume that great predictive, positive, negative moderate outcome? And fraud is kind of left a little bit the first time I've seen you? Yeah, there happened to be the first time and get in there for a little bit of time. And then end up later on. So I think, I think this tool is still very immature in the management portfolio management side of the world. But I actually believe the tools, right, clients in their own transactional, amen, call center collections history is just got so much information, which is ideal for this space, I think finding a way to monetize that for benefit. And then also to kind of weed out a lot of this later stage fraud activity. Later state bad, they identify or transactional decline of some of the borrowings or positive transactional improvement is kind of that sweet spot.

Matt Tepper  37:41
And last one for you guys. I don't want to let your imagination run wild here with theoretical innovations do you now envision AI bringing to fruition within lending and finance, for example, I read an article speculating that AI could bring revolutionary change to the world of medical care, take advantage of some of the existing biomedical technology that's already been adapted. So instead of typing your symptoms into a search engine, and getting every prognosis between, you'll be fine and an hour to head to the hospital immediately, AI can now deliver a little more personalized diagnosis. They didn't say advancements and not only care but also just major efficiencies, you envision a similar path and finance and lending where AI expands upon concepts like open banking and alternative data. Carl, your example earlier of maybe the deposit now kind of AI now routes it according to your spending habits. Do you see some like that taking hold and anything else you've kind of been noodling around with.

Carl Spilker  38:39
Exactly that something a little more prescriptive. We started this conversation a couple of weeks ago, I was thinking what I call the inefficient money that sits in my account that I haven't made a decision yet. And it may not be huge amounts, but over 10 years, it can be significant. And if I could have just had a marginal player, allocating it with some relatively positive cash flow, that would have benefited me that market. So that's one area. I think of all the simple things like recipes and cooking dinner. Yeah, that'd be beautiful to say, Hey, dude, how do I make law and it gives you a little more precise activity. Or I need to get to point A to point B, I start and something happens, you know, how can I make that more efficient, or it's something proactive? And by the way, I don't think I could find my way to the grocery store. Now Google Apps, right. I could actually see it replacing a lot of day to day activities, as I mentioned earlier, make it more efficient. I think that's a plus.

Matt Shubert  39:47
Well maybe it's too soon, I purchased our first home about two years ago so around this time it a little bit first time a little over two years guys going through the mortgage process and it was hectic and I thought they didn't bake. I work with a really good job about it. automating as much as possible, but I could see that being a great opportunity to streamline it and some already are in some ways but I think if you could streamline that document collection process down to the signing, I think what the most painful things I had to do which was is a great it was fun because it from my wife and I it was saver Senator documents, we had to put what signatures and all of our loan documents Oh, my gosh, I did everything your DocuSign up to the very end and out of sight out 183 pages doing 25 or 20, what signatures and initials and everything else. And I think there could be opportunity to apply there in some way just didn't make that process that much more streamlined. Why does it have to be three days before you're close? Like there's got to be a way to add AI in there to see why can't we make this a little bit more conducive?

Carl Spilker  40:41
Yeah. And by the way, Matt sorry to interrupt. Exactly. He had something do at a counter legal review as well, because I kind of laughed myself. I have zero idea what I am talking about.

Matt Shubert  40:55
Wouldn't it be nice to get somebody advises on your behalf versus the really especially if you think about the the marginalized populations that have been had have struggled to get lending over time, right? This might be the first time it was my first time sign and all that type of stuff. But I can only imagine that's going to apply for other areas. And I had been in auto lending for years. So I was pretty familiar with an auto lending contract. But gosh, mortgages the whole world I think is operating from there. You mentioned Google map that one that caught my ear where you think what if you could take that you went the wrong way to tell you where to go to get back to where you really want to go with that thing, what direction to go. There's got to be such a great opportunity with that from, from how you enable when you make lending more inclusive and equitable for those populations that need that kind of push along the way I'm pretty good at I'm pretty financially savvy, because I've worked for banks in the past, I understand how to get the best rate and get the best account and shot my CDs and my high yield accounts, all that kind of stuff. I don't know that people know how to do that. And they're going to depend on an aggregator that's probably getting paid for that. What if you had a tool that would help us folks find the best thing for them? Even internal so you bank with a with a bank, and they know you're going to have about 20 grand in your bank account? Well, heck, you always have carrying this 20 grand What if we put that in a CD for three months and get you 3% Instead of letting it sit there getting point 2% Those are always just enable and make I think banking warfare for a wider population. Obviously, there's a profitability piece of that because why pay you three when they can pay you at point two, but I think it would, it would go a long way with folks saying you're making the banking crisis more inclusive, equitable, fair, reasonable.

Matt Tepper  42:18
I know that we are only scratching the surface on this topic. And before we wrap up, we have an exciting announcement for our listeners. GDS Link in collaboration with our partners at Equifax we'll be hosting a webinar, ride the all data wave or get swept away how alternative data and custom decisioning solutions are revolutionizing lending. This insightful discussion will take place on Thursday, June 22. And we invite all of you to register for the webinar by visiting the link provided in our show notes. It's an excellent opportunity to delve deeper into the topic we've been exploring today. This is Matt Tepper and we've been syncing up with Carl Spilker, Executive Vice President of Global Analytics and Advisory at GDS Link and Matt Shubert, Vice President of Predictive Analytics and Data Science at West Coast University. Thank you for tuning in and learning more about AI and its impact on the future of lending. Matt, Carl, appreciate both of your time today.

Carl Spilker  43:08
Yeah, thanks, man.

Matt Shubert  43:09
Thank you very much  both, really appreciate it.

Matt Tepper  43:12
If you would like to learn more about the work of Matt and his team are working on visit "westcoastuniversity.edu". Stay connected with GDS Link and The Lending Link to listen to future podcasts and catch up on the ones you missed. If you enjoyed today's episode, please be sure to subscribe on Apple podcast, Spotify, or wherever you listen to your podcast. And be sure to leave us a five star review. Follow us on LinkedIn and connect with us on Twitter at GDS Link that's at  GDSLINK on Twitter. Have a question for the show or a specific topic you want us to cover. Hit the link in the description to drop us a note. again want to thank both Carl and Matt for their time today. I want to thank you all for lending us part of your day. Make it a great one.