Podcast: Approaching AI with a plan

Whitney McDonald 0:02
Hi there and welcome to The Buzz, a financial institution automation information podcast. My title is Whitney McDonald and I’m the editor of financial institution automation Information. Immediately is November 7 2023. Becoming a member of me is Mike Lempner. He’s head of engineering and expertise at FinTech mission lane. He’s right here to debate methods to use the precise sort of AI and underwriting and figuring out innovation and use instances for AI, all whereas approaching the expertise with compliance on the forefront. He labored as a guide earlier than shifting into the FinTech world and has been with Mission lane for about 5 years.

Mike Lempner 0:32
I’m Mike Lempner, I’m the pinnacle of our engineering and expertise at mission lane. Been within the position the place I’ve been main our expertise group and engineers to assist construct completely different expertise options to help our prospects and allow the expansion of mission lane. I’ve been in that position for about 5 years previous to that mission Lane was really spun off from one other fin tech startup, and I used to be with them for a couple of 12 months as an worker previous to that as a guide. And previous to that point, I spent about 28 years in consulting consulting for quite a lot of completely different fortune 500 corporations, startups, however largely all within the monetary companies house.

Whitney McDonald 1:09
And possibly you possibly can stroll us by mission Lane give us a bit background on what you guys do. Certain,

Mike Lempner 1:16
Mission lane is a FinTech that gives credit score merchandise to prospects who’re usually denied entry to completely different monetary companies, largely partly as a consequence of their minimal credit score historical past, in addition to poor credit score historical past prior to now. For probably the most half, our core product that we provide proper now’s we’ve a bank card product that we provide to completely different prospects.

Whitney McDonald 1:39
Properly, thanks once more for being right here. And naturally, with the whole lot happening within the business. Proper now, we’re going to be speaking a couple of subject that you just simply can’t appear to get away from, which is AI and extra particularly ai ai regulation. Let’s let’s form of set the scene right here. Initially, I’d wish to go it over to you, Mike to first form of set the scene on the place AI regulation stands right this moment and why this is a crucial dialog for us to have right this moment.

Mike Lempner 2:08
Yeah, sounds good. As you talked about, Whitney AI has been actually all of the the dialog for in regards to the previous 12 months, since Chechi. Beatty, and others form of got here out with their capabilities. And I believe in consequence, regulators are taking a look at that and making an attempt to determine how can we meet up with that? How can we be ok with what what it does? What it supplies? How does it change something that we do presently right this moment? And I believe for probably the most half, you rules are actually stand the check of time, no matter expertise and information. However I believe there’s at all times form of the lens, okay, the place we’re right this moment with expertise, has something modified the place we’re by way of information sources, and what we’re utilizing to form of make choices from a monetary companies standpoint is that additionally creating any form of considerations and also you’ve bought completely different regulators who have a look at it, you’ve bought some regulators who’re taking a look at it from a shopper safety standpoint, others who’re taking a look at it from the soundness of the banking business, others who’re taking a look at it from an antitrust standpoint, privateness is one other, , massive facet of it and in addition to Homeland Safety. So there’s there’s completely different regulators taking a look at it in several methods and making an attempt to know and and attempt to keep as a lot forward of it as they presumably can. And so I believe numerous instances that they’re taking a look at issues and making an attempt to form of have a look at the present rules, and perceive are there changes that must be made an instance of that CFPB, I believe lately offered some some feedback and suggestions associated to opposed motion notices, and the way these are principally being generated within the gentle of synthetic intelligence, in addition to like new modeling capabilities, and together with, like new information capabilities. So I believe there’s there’s some particular issues in some ways it doesn’t change the core regulatory want. However I do anticipate there’s going to be some fantastic tuning or changes that get me to the rules to form of put in place extra extra protections.

Whitney McDonald 4:10
Now, for this subsequent query, you probably did give the instance of taking a look at present regulation, protecting all of the completely different regulatory our bodies in thoughts what already exists within the house? How else may monetary establishments put together for brand spanking new AI regulation? What may that preparation appear like? And what are you actually listening to out of your companions on that entrance?

Mike Lempner 4:33
Yeah, I believe it’s, it’s not simply particular to AI rules. It’s actually all rules, and simply form of wanting on the panorama of what’s taking place. You understand, the place we’re. I believe the one factor that we all know for certain is regulation modifications will at all times occur and the they’re simply part of doing enterprise and monetary companies. And in order that want shouldn’t be going away. So There are completely different privateness legal guidelines which are being put into place some, in some instances by completely different states. There’s different issues, , as I discussed with AI are rising and development, how do regulators really feel snug with that as nicely? So I believe by way of making ready, identical to you’d with any regulatory actions happening, it’s essential to have the precise individuals inside the group concerned in that in for us, that’s usually our authorized crew or threat crew who’re working each internally in addition to getting exterior counsel, who will assist us perceive like, what are a number of the present regulatory concepts which are on the market being thought-about? How may that impression us as a enterprise and we’re staying on prime of it. After which as issues materialize over time, we work to raised perceive that regulation, after which what it means for us, after which what do we have to do to have the ability to help it. So I believe that’s a greatest a part of it’s getting the precise individuals within the group to remain on prime of it know what’s presently taking place, what is likely to be taking place sooner or later, leveraging exterior assets, as I discussed, is they might have experience on this space, and simply staying on prime of it so that you just’re not stunned after which actually form of reacting to the scenario.

Whitney McDonald 6:14
Now, as AI regulation does begin coming down the pipeline, there’s undoubtedly not been a a ready interval, relating to investing in AI implementing AI and innovating inside AI. Possibly you’ll be able to speak us by the way you’re navigating all of these whereas protecting compliance in thoughts, forward of additional regulation that does come down. Yeah,

Mike Lempner 6:39
completely. The, , for for us in AI is is a extremely form of broad form of space. So it represents, , generative AI like chat GPT. It additionally entails machine studying and different statistical sorts of algorithms that may be utilized. And we function in an area the place we’re taking over threat by giving individuals bank cards and credit score. And so for us, there’s a core a part of what we do the underwriting of credit score. That’s is difficult entails threat. And so for us, it’s essential to have actually good fashions that assist us perceive that threat and assist us perceive like who we wish to give credit score to. We’ve been ever since we bought began, we’ve been utilizing AI and machine studying fairly a bit in our our fashions. For us, one of many essential issues is to actually have a look at and the place we could have many fashions that help our enterprise. A few of them are credit score underwriting fashions, a few of them are fraud fashions, a few of them could also be different fashions, we’ve dozens of various fashions that we’ve is ensuring that we’re making use of the precise AI expertise to satisfy each the enterprise wants, but in addition taking into consideration regulation. So for example, for credit score underwriting, it’s tremendous essential for us to have the ability to clarify the outcomes of a given underwriting mannequin to regulators for example. And so when you’re utilizing one thing like generative API, AI or chat GPT, the place accuracy shouldn’t be 100%. And there’s the idea of hallucinations. And whereas hallucinations might need been cool for a small group of individuals within the 60s, it’s not very cool once you speak about regulators and making an attempt to clarify why you made a monetary resolution to provide any individual a bank card or not. So I believe it’s actually essential for us to make use of the precise sort of AI and machine studying fashions for our credit score underwriting choices in order that we do have the explainability have it. And we have been very exact by way of the result that we’re anticipating, versus different kinds of fashions. And it could possibly be advertising and marketing fashions, there could possibly be, as I discussed, fraud fashions or funds fashions that we could have as nicely that help our enterprise. And there, we’d have the ability to use extra superior modeling methods to help that.

Whitney McDonald 8:57
No nice examples. And I like what you mentioned about explainability as nicely. I imply, that’s large. And that comes up again and again, when it does come to sustaining compliance whereas utilizing AI. You’ll be able to have it in so many alternative areas of an establishment, however you have to clarify the selections it’s making, particularly with what you’re doing with with the credit score decisioning. I’m shifting in to one thing that you just had already talked about a bit bit about, however possibly we are able to get into this a bit bit additional. is prepping your crew for AI funding implementation. I do know that you just talked about having the precise groups in place. How can monetary establishments look to what you guys have carried out and possibly take away a finest follow right here? For actually prepping your crew? What do you have to have in place? How do you alter that tradition as AI because the AI ball retains rolling?

Mike Lempner 9:52
Yeah, I believe for us, it’s just like what we do for any new or rising expertise usually. which is, , we’ve bought a an total form of framework or course of that we’ve like one is simply establish the chance and the use instances. So we’re actually understanding like, what are the enterprise outcomes that we’ve? How can we apply expertise like AI or extra information sources to unravel for that exact enterprise problem or consequence. After which in order that’s one is simply having that stock of the place all of the locations that we may use it, then to love actually taking a look at it and understanding the dangers, as I discussed, credit score threat is one factor. And that we could wish to have a sure method to how we try this, versus advertising and marketing or fraud or different actions could have a barely completely different threat profile. So understanding these issues. And even once we speak about generative AI, for us, utilizing it for inside use instances of engineers writing code and utilizing it to assist write the code is one space the place it is likely to be decrease threat for us, and even within the operations house, the place you’ve bought customer support, who possibly we are able to automate various completely different features. So I believe understanding the use instances understanding the dangers, then additionally having a governance mannequin, and that’s, I believe, a mixture of getting a crew of individuals which are cross purposeful to incorporate authorized threat, and and different members of the management crew who can actually have a look at it and say, right here’s our plan. And what we wish to do with this expertise, can we all really feel snug shifting ahead? Can we totally perceive the danger? Are we taking a look at it like holistically, then additionally, governance? Like for us, we have already got mannequin governance that we’ve for that actually establish what are the fashions we’ve in place? What kinds of expertise can we use? Can we be ok with that? What different form of controls do we have to have in place. So I believe having an excellent governance framework is one other key piece of it. Investing in coaching is a one other key factor to do. So significantly within the case of rising generative AI capabilities, it’s quick evolving, it’s actually essential to form of guarantee that individuals simply aren’t enamored by the expertise, however actually understanding it, understanding the way it works, understanding the implications, there’s a distinction as to whether we’re going to make use of a public dealing with software and supply information like Chet GPT, or whether or not we’re going to make use of inside AI platforms utilizing our inside information, and use it, , for extra proprietary functions. So there’s a distinction, I believe, in some ways, and having individuals perceive a few of these variations and what we are able to do there, it’s essential. I believe, lastly, the opposite key factor from an total method standpoint, is to actually iterate and begin small, and get a number of the expertise on a few of these low threat areas. In for us the low threat areas, like we’ve recognized various completely different areas that we’ve already constructed out some options round customer support. And engineering, as I discussed, you should utilize a number of the instruments to assist write code, and it is probably not the completed product, nevertheless it’s no less than a primary draft of code which you could, you can begin with that. So that you’re not principally beginning with a clean sheet of paper.

Whitney McDonald 13:09
Yeah, and I imply, thanks for breaking out these these decrease threat use instances which you could put in motion right this moment. I believe we’ve seen numerous examples recently of AI, that’s an motion that is ready to be launched and used and leveraged right this moment. Talking of possibly extra of a future look, generative AI was one factor that you just had talked about, however even past that, would simply like to get your perspective on potential future use instances that that you just’re enthusiastic about inside AI, the place regulation is headed. However nevertheless you wish to take that future look, query of what’s coming for AI, whether or not within the close to time period, or close to time period or the long run? Certain.

Mike Lempner 13:53
Yeah, it’s I believe it’s a really thrilling time and insane, thrilling house. And to me, it’s exceptional simply the capabilities that existed a 12 months in the past the place you possibly can form of go and and put in textual content or audio or video and have the ability to work together and and get like, , fascinating content material that might assist you to simply extra whether or not it was simply private searches or no matter be productive, and to now the place it’s accessible extra internally for various organizations. And even what we’ve seen internally is making an attempt to make use of the expertise six months in the past, could have concerned eight steps and numerous what I’ll name information wrangling to form of get the information in the precise format, and to feed it in to now it’s extra like there is likely to be 4 steps concerned in so you’ll be able to very, you’ll be able to rather more simply combine information and get to the outcomes and so it’s change into rather a lot less complicated to implement. And I believe that’s going to be the longer term is that it’ll proceed to get simpler, a lot simpler for individuals to use it to their use instances and to make use of it for quite a lot of completely different use instances. And I believe completely different distributors We’ll begin to perceive some patterns the place, , there is likely to be a name heart use case that, , at all times happens, , one instance I at all times consider is, I can’t consider a time prior to now 10 plus years the place you known as customer support and get transferred to an agent, the place they don’t say, this name could also be recorded for high quality assurance functions, with high quality assurance of a telephone name often entails individuals manually listening to it and taking notes and form of filling out a scorecard. Properly, now with , AI capabilities that may all be carried out in a way more automated method. So there’s, there’s numerous various things that like that form of use case, that sample that I’m guessing there are gonna be distributors who will now put that sort of answer on the market and make it very straightforward for individuals to eat nearly just like the AWS method, the place issues that AWS did are actually form of uncovered as companies that different corporations can form of plug into very simply. That’s an instance the place I believe the expertise is headed, and also you’ll begin to see some level options that can emerge in that house. from a regulatory standpoint, I believe it’s going to be fascinating, , just like dying and taxes, I believe, , regulate regulation is at all times going to be there, significantly in monetary companies. And it’s to do the issues that we talked about earlier than defending prospects defending the banking system defending, , completely different areas which are essential. So I believe that’s, that’s a certainty. And for us, , I believe it’s, there’s more likely to be completely different, completely different modifications that can happen on account of the expertise and the information that’s accessible. I don’t see it as being drastic modifications to the rules. However extra wanting again at a number of the present rules and saying, given the brand new expertise, given the brand new information units that exist on the market, are there issues we have to change about a few of these present rules to guarantee that they’re, they’re nonetheless controlling for the precise issues?

Whitney McDonald 16:59
You’ve been listening to the excitement, a financial institution automation information podcast, please comply with us on LinkedIn. And as a reminder, you’ll be able to price this podcast in your platform of selection. Thanks on your time, and you’ll want to go to us at Financial institution automation information.com For extra automation information,

Transcribed by https://otter.ai