View Full Library
We are here to talk about AI.
Are we dabbling in AI? Are we doing this thing?
And our panel here, I'll have them introduce themselves, but, what they're going to talk about is some of those practical use cases for AI.
So we'll go just down the line and introduce ourselves.
I'll go really quickly. I'm not the most important person up here.
I'm here to get you guys to give us the real good nuggets.
But really quickly, I'm Crystal Anderson and I am the Chief of Staff at MX.
Prior to that, I headed product at MX for two years.
I've been here for three years.
And prior to that I was at H&R Block where I led the financial services business.
I was vice president of financial services.
We loaned $1.8 billion a year and had $29 billion in deposits through our financial services program.
So I've been in this space for quite some time.
I'll have you go next.
Awesome. Yeah. Thank you. Crystal.
John Sun, co-founder and CEO at Spring Labs.
My background is in FinTech lending.
So before Spring Labs, I was the co-founder and Chief Risk Officer at Avant, where I oversaw credit risk, fraud risk, compliance, risk, you name it, as part of the overall risk stack.
My training is really as a data scientist, that's kind of what I always kind of go back to.
So at Avant, my team and I built one of the first tree-based kind of machine learning algorithms for real time underwriting back in 2012.
I at least haven't heard of another company that's deployed anything earlier than that.
But again, a little out of practice.
I haven't probably built a model since 2016 or reviewed the model since 2018.
So my knowledge, I'm sure is out of date compared to some of these guys on the panel here.
And, today, what we're doing at Spring Labs is making AI accessible to financial institutions.
When I say AI, I mean, in this case, more generative AI.
How do we use the conversational intelligence capabilities of generative AI to help banks and financial institutions with regulatory risk reduction, with operational efficiency, with better kind of staffing and better engagement with customers to kind of create more delightful experiences. Great.
Thank you John. Sam? Hey everybody. I'm Sam Maule.
I started in banking and payments back in ‘94, so I've been doing this for a little while.
Been a banker, been a worked at TSYS, did consulting, did a stint at Google, and now I'm with a payments company called Moov.
So this ought to be fun diving into this. Yeah.
I'm Zach Boyd, Director of Utah's Office of AI Policy, newly created four months ago.
Also a math professor at Brigham Young University, where I've been doing research basically into foundations of machine learning and applications in social science.
Wije. The most important person in the room here, And least knowledgeable about finance for sure.
My name is Wije.
I'm co-founder and CEO of a company called Aliya.
We build AI-powered solutions for banks.
Great. Well, my first question's going to be, tell me a bit more about what your organizations are doing with AI.
So Wije I am actually gonna start with you. What does that mean?
You promised me I would go last.
That's why you sat down there.
That's why. Better I know.
What do we do? In this?
The simplest way of explaining it is we take the most valuable data a bank has, which is the transaction data, and we support lending and risk management using that data.
That's in the simplest way.
We go through the ringer in terms of, well, we've been doing it for six years now.
We've done, you know, close to $7 billion worth of loans.
We work with a very large bank partner.
And it's been a learning process for us in terms of the biggest constraints to deploying AI.
Yeah. We've been through it living it, regulatory, compliance, legal, and all of that is what we do. Great.
Sam, why don't we hear from you next? Sure.
What are we doing? So with Moov, it's everything about money movement.
So money in money out, storing all that stuff.
So basically just think about all that transaction data and, I think I said I worked at Google, so shockingly we're doing a lot with Google on the BigQuery side, Looker, and the other tools that we can tie into.
So, I mean, transaction data is the holy grail when it comes to payments, right?
That's what we all wanna see. So you can, yeah.
Yeah. Use your brain. Think What we're doing. Use your brains, use your brain, everybody.
Zach, I'm really interested in this newly formed, tell us some context around this newly formed organization that you're leading and kinda the impetus to the creation of that.
Yeah, sure. Happy to discuss.
So, Utah's Office of AI Policy is unique in the United States and maybe in the world now.
Utah moved first. The idea was to have an in-house government capability where there's an office with people who understand AI are following the dozens of relevant regulatory issues that might affect the state and are primed and ready to make good recommendations to the legislature if and when it seems right for the state to act on AI policy.
So this is mostly regulatory policy, certainly in finance and banking, in healthcare, schools, you name it.
The legislature specifically wants us to be wearing both the consumer protection hat and the innovation fostering hat, which I think has really been super effective so far because usually those two people wear different hats and it leads to a lot of dysfunctions in various regulatory systems.
So I think that's been really good.
We have a charge to go through topically and do deep dives on matters of importance.
So, for example, we could do a deep dive on AI and fintech and banking and think through, you know, what does the right regulatory structure need to be in the age of AI?
How can we make sure things make sense, and then make reasoned recommendations to the legislature going through topically.
We'll take a lot of time and AI moves fast.
So the legislature also granted us a regulatory mitigation authority, which is basically a very kind of nuanced sandbox authority.
If a company wants to use AI in the state to simplify slightly, I can write a contract to any company that wants to operate in the state, clarifying that the regulations of the state shall apply to you thusly, or that you will be exempted from the following regulations of the state.
Or we'll give you a curing period, or we'll cap fees.
And so this applies across all the regulations of the state.
So we've had some companies approach us already, but the idea is we want companies to be able to explore innovative business models and we want Utahns to reap the benefits of AI as rapidly as possible.
How big is the team? Sorry, I had to ask, ask How big is the Team? Oh, I was actually gonna ask you about your facial expression, Sam.
So no, I'm — How big is the team? Oh, I mean, we're a small government office.
I have six people. It's all right. So, no, I mean, it's actually pretty good. It’s six people doing what you’re doing.
That's awesome. Yeah. I mean, but my philosophy is always, I've tried to pack as much knowledge as possible into this small state office as I can, but we are not now, nor ever will be resourced enough to have deep in-house expertise on every subject.
So we have a model of stakeholder engagement.
I've spent a lot of my time trying to find the smartest people that I possibly can in all the relevant industries and engage with them to rapidly acquire the most important facts.
Sam, so you pointed at John and Wije.
Tell me what makes you say they're the smartest people in using AI in the industry? Because they're buried in it and been doing it quietly, looking at you quietly.
When people work quietly and aren't all over social media talking about how incredible they are, I tend to, they're the people grinding this out.
And this takes a long time.
I'll speak for Wade Arnold, who I work for at Moov.
When you're building out payment rails from, you know, bare metal connections all the way up, it takes a long time and it's really, really hard.
What we're talking about is transformative, what y'all are doing. Yeah, Thank you. You're welcome.
Hey, Wije, can I have you speak into the mic just so that the recording can hear you?
I thought my voice is loud enough that everybody that no, anything that is as transformational as this is really, really hard.
People completely underestimate how difficult it is, the cost of doing it.
I mean, it's prohibitive, right? Compute costs now.
It's crazy. I mean, we had to build our own stack of Nvidia H100s to mitigate some of that.
The talent, it's very hard to find.
They, the people who do this are really very special people.
They're kind of not normal because they're just, they're not data scientists in a traditional way.
They're kind of artists.
'cause they really just know something that we don't normally.
The other part of it is, you know, the regulatory lift.
But I think in terms of one of the biggest hurdles is finding the right leadership with the right strategy.
Because everybody talks about AI.
In fact, I had my team just run all of the annual reports of everybody here looking at how many times AI machine learning and artificial intelligence I mentioned in the reports.
So you got J.P. Morgan up at the lit top, huge number of times that they talk about it.
And then the vast majority are about one once, you know.
So, but there is a lot of conversation going on about it.
And, but the strategic direction is the really important, it's like everybody thinks of point solutions.
It's not a point solution, it's a curated offering, right?
Even MX, which is one of the best in the business, is beginning to realize that it's a curated solution that has a closed loop to it that makes money.
And banks need money.
Their NII, I mean, I'm speaking for you guys, but you know it better than I do.
Pretty much sucks, can't grow. What are you gonna grow?
And that's the problem AI can solve because you can get deeper into your customer.
You can figure out how to price them more accurately and you can make better loans to more people, right.
As is the single.
Yeah, I mean, I would echo that sentiment, although Sam, you know, the reason we don't do social media is because we're bad at it, not because for lack of trying.
But I think that's actually, you know, Wije, you bring up a really good point, which is, you know, every time there's a new technology that comes about, people are like, oh, AI this, AI that, right?
At the end of the day, a good AI product shouldn't be defined by the AI-ness of it.
It should be defined by its value to your organization, right?
Ultimately, that's what matters. And I feel like at the beginning of every single kind of tech cycle, you see a lot of investment into the AI-ness of products instead of investment into ROI investment into how it moves your organization forward.
So that's something that we try to focus on at Spring Labs is how do we build products that, again, present efficiency, delightful experiences to your customers, and ultimately add your, move your business forward and add value to your customers where the AI of it is kind of a secondary. Can I ask you a question real quick?
Of course you can. Just real quick, I, I like what you said about don't think data scientists think artists and all that.
Yeah. Because I've heard it and you know, stated when we're talking about AI, it's like learning an alien language or can you emote on that a little bit?
Yeah. It's, I, most people think of AI and data science in a, in query form that is so one dimensional, right?
Because it's about, think of it like you're sitting there and you've got a trillion dots around you, around your head, and it's kind of like, how do I connect these dots to make some sense of that?
And that's a very complex thing and that's what these guys do.
If you can imagine that.
And the power of it is like in categorization, you guys are all familiar with transaction data and categorization.
When a category comes out, you can't explain how it got there.
I mean, it's very, very difficult 'cause it's multidimensional.
So it becomes a problem for us vis-a-vis the regulators when the compliance component of it is, well how did it get there?
You know, it's like us trying to explain a multidimensional framework in 2D, and this is where I think the most amount of work needs to be done, is a collaboration of people like us with the regulators, the sandboxes, or we talked about clinical trials, right?
In the medical industry, there is a lot of AI that's used to analyze data and come up with protocols that have to go through clinical trials.
And the FDA process. We should be thinking along those lines and we will.
Yeah. And I'll say that, you know, not just in banking and fintech, but across the board I'm seeing a transformation where before we're so used to humans being able to give us explanation for things.
And, in some sense that, you know, it's a value to the consumer to be able to be given an explanation for things if those explanations are actually valid, right?
Like they derive subjective benefit from this and sometimes hard concrete benefits from it.
But now I, you know, it's gonna take society a little while to realize that with these, with this kind of model, there's a complex trade off between the value of being able to explain things and the capability of being able to do better, right?
Like the best models right now are the least explainable.
And I know some people are trying to get around compliance requirements or to meet compliance requirements by taking the AI and tacking some explanatory output onto it.
And I, you know, I am sympathetic.
I don't think that this is the biggest priority to go after as a regulator at this very moment because it is the state of the technology and people feel like they need an explanation.
But I think scientifically it's actually super unclear that these explanations actually are explaining what the decision making process of the AI actually is.
What I hope society gets to is to the point where we realize there's just a value trade off here, right?
Like we can have enhanced capabilities at the expense of less explanation and we can accept something along this curve of how much explanation we really need.
Talk to me about some of those use cases.
So we've heard lending and pricing and improving the experience.
What are some of the most impactful use cases you're seeing in financial services? We'll start with you John. Yeah, great question.
I mean, practical applications, obviously is the key to success for any technology.
And looking at it from the more generative AI side of things, and I'm sure kind of Wije and the rest of the folks have the opinion on the more traditional machine learning AI, generative AI is uniquely good at understanding human language.
So these are pre-trained models that have such a vast corpus of data.
You can manipulate language questions and kind of constructs without, a ton of additional training.
So I would say look around your organization and just see where there's inefficiencies that involve a bunch of kind of language related tasks.
I mean, the ones that we're seeing that's really kind of interesting.
Compliance is a good one.
It's always one of these areas where, you know, even if you're a tech forward organization, you never say, let's invest in compliance tech.
You say, let's go throw more bodies at it.
So here now you're presented with a technology that can all of a sudden change the game in terms of how you interface with your customers, how you hear what your customers are telling you, how you turn those insights into kind of structured data to better inform technology process and roadmap decisions.
So those types of things are the types of use cases that generative AI is uniquely kind of good at.
I'm gonna give an example that's not banking, if that's okay.
That's okay. So from the Google days, so in Canada, especially in British Columbia, they went through a period of excessive drought, ton of forest fires.
Most of those forest fires were caused by truckers, long haul truckers going because of the speed of what they travel.
We create sparks, we create fire and everything else.
So this gets to your multi-layered data elements all coming together at the same time.
So Google, through satellites and weather patterns and through trackers that they could put on the trucks could actually say, alright, here are the weather conditions for specifically where you're at in real time as you're traveling.
Because what happened was the government came in and said, all right, you have to drive under 45 miles an hour that yeah, I know, but to stop the forest fires because of what we're in.
So just the rule or, or the regulation is gonna be, you gotta drive under 45. In trucking and logistics, that's, you're dead, right? That's killer.
So Google was able to come in and work with them to say, okay, we can go and monitor each individual driver, the trucks themselves, the speed that they're going, where they're at with satellite data in real time and what the weather conditions are.
And we can modulate at what speed you need to go in real time, which now is impacting dollars and everything else.
So if you think about that, that's a ton of different data elements all coming together, which was driving profitability.
And by the way, that's like five years old.
That's not something that's new.
So, you've gotta start thinking in ways you haven't thought before, getting back to the alien type brain. Right?
Yeah. I mean, I think you make a very important point. Good.
That was the only one I'm gonna get for this entire talk.
I've been saving that one up.
Take everything that you've been doing for the last couple of decades because it really has been based on not having to take any risk.
When interest rates are zero, you don't do anything with your customer.
You shouldn't be doing anything with your customer.
You should be taking your cheap dollar cheap funding, deposits and going and buying long duration treasuries.
That's what happened.
And it has not created any innovation.
And then now you're getting overloaded by this thing called AI and it's confusing. A lot of people take a blank sheet of paper, re-engineer the whole thing, and your most valuable piece of data, there are three pieces of data that you need to focus on.
One is the transaction data, your bureau data and public records data.
If you know how to use those effectively to serve your customers, you're going from zero to 60 in no time.
Forget about zero to a hundred.
That's what the zero to 100 is what highly paid management consultants will tell you.
'cause they'll go, oh yeah, you need to create the orchestration layer, the data layer.
You need to create all of these things. It's h******t.
Sorry, promised I wouldn't do that. Keep it simple.
'cause banking — everybody's made banking complex.
If you take away the complexity, it's a pretty simple thing.
Take in deposits from your community, lend it out safely, make a spread.
That's what the social mandate is.
And you need to use the best tools available to you.
And AI is totally transformational because you're gonna be able to predict behavior losses.
If you can do all of that, you can price it more accurately rather than guessing.
There's, if you price it correctly, size it correctly, then you make a lot of money.
Can I ask a question actually of the other panelists?
I know in other industries, I tell people, you know, you've got these weird strategic decisions right now.
'cause AI, especially generative AI, is currently the worst that it will ever be, right?
It will only improve from here, all kinds of hypotheses about how much or how little it will improve.
And, I think for a lot of people, this has a strategic bearing, but I am hearing you guys basically talking about the benefits that are already here.
Yeah. What's your opinion?
Like, how much are the benefits already here versus how much do you think is coming?
Do you guys have like, theories about this?
Sorry, John, but compliance is a bunch of rules.
Well, I mean, it shouldn't be from my perspective.
It shouldn't be interpreted.
Okay. Like it compliance is, you know, So don't — You're saying like, you don't need generative AI to do compliance, in my opinion.
It's a set of rules.
You can build it into ones and zeros.
I hope somebody's not agreeing with me on that one. We'll give you a mic in a little bit and Look, we, yeah, we're doing it.
So with it, with a large OCC regulated bank, it's built in, you can build it into the workflow.
Yeah, I mean, I think we might be talking about two different kinds of compliance.
I mean, there's the type of compliance which is, you know, let's analyze a particular model for a particular outcome or a particular set of inputs.
And then there's the compliance of, you know, I just sent out a marketing mailer to this comply with all of my obligations under, you know, various rules that I have for marketing this product.
I just received a complaint from a customer.
Did the, you know, three pages of like personal life story and like my dog just died and all of that.
Did any of that trigger a UDAP concern? Right?
And again, can you do it with traditional machine learning?
Yes, absolutely.
It would be incredibly time consuming and cost inefficient.
What generative AI has presented us is a set of models where you don't have to start from zero.
So if you think about what GPT stands for, it stands for generative pre-trained transformer models, right?
Everyone focuses on the G, which is generative.
What you should be focused on is the P, which is this pre-trained.
It comes out of the box with a vast corpus of human knowledge.
And you can now use that as a springboard to do way more advanced use cases than you could if you just started with the transformer model from zero.
Yeah. As a former Google, I think the transformer part's very important because of my stock.
Let's just make sure that gets said.
I wanna make sure we hear — I saw lots of Yes.
That latter example is what I align with.
So tell us more about that. Absolutely.
Can you introduce yourself real quickly?
Not a problem. So I'm Doug Milow, senior Director of Credit and Fraud Risk Strategy.
I'm also the fair lending officer for Best Egg.
We're a consumer lending platform.
Here's where I respectfully disagree and agree with John.
It's complaints, right?
We've got so much data we collect to see how we're doing.
We've got online surveys, we've got surveys from our funnel, we've got customers talking to agents.
You've got that agent shorthand that they love to write in to describe what happens.
We've got the tracking of escalations to the management layer, we've got the regulatory complaints coming in.
We've got all kinds of signals and it's just completely unstructured.
And that's where Gen AI can really shine.
We're we're working with the Amazon, you know, platform to kind of figure out where to go with that.
So wasn't talking about, I should have clarified my point.
Got it. I was more about serving the customer, not the complaint side of it, but Yeah, some of, some of those. Good point. Yeah, totally agree with you on that part.
You know, it's, Hey, if your APR is above X and your state says it can only go to Y, you've got a problem.
That's a rule. We agree.
Hey, Wije, can you expand a little bit?
'cause you touched on it, at, Moov, we had Matt Harris from Bain Capital, if you know Matt, he's been doing this forever.
The OG when it comes to this.
And he talked about AI and agents and some other components and how it's just gonna wreck havoc with NIM.
I mean, you mentioned that before on the net, you know, can you expand a little bit on what you were saying On the NIM? Yeah.
Look, the problem, the problem is in order to generate the name you or NII you, you need to be able to lend, most organizations are struggling to figure out how to lend one of the spaces is in the Best Egg space.
We should be going after all of the Best Egg customers because they're taking your customers by the way.
And the reason is they're really good at predicting losses, pricing, and getting to the customer.
And the banking system isn't, and because of that, you've left a lot of room available to the fintechs to thrive.
SoFi who, you know, I was on the board of for five years, they did $11.5 billion of unsecured personal loan originations last year.
Okay? I don't know how much you guys did, but you've done billions and like total of 30 billion, if I'm not mistaken in your lifetime.
That's business that belongs to the banking system.
I mean, just bank because you already have the customers by the way, they had to go get their customers.
And that's, I mean, that's money, right?
So that's one aspect of, the other aspect of it is in, Because CAC is a big deal.
It's a big deal. I mean, that's huge.
The other aspect of it is your losses.
You haven't invested in what companies like Best Egg have invested in.
So your losses are higher.
Then the other part of it is your production costs.
I mean, 27 screens, and going to the branch to close a loan And on legacy tech that was built pre-internet for your course not to go down that path too much. But we know that, I don't mean, I'm being, matter of fact here, this is not a criticism of anybody.
It's just lay of the land and we're experiencing something right now in terms of AI that is totally transformational.
That's why I think it needs a blank sheet of paper approach.
First principle thinking, but that's hard to do.
So can it be transformational without the blank slate, right?
If the regulatory structure doesn't change, what happens?
I think this is a Google / Microsoft type situation where the big guys are gonna win.
I was gonna ask you guys this question.
How much do you think Chase and B of A are spending on tech this year?
10 billion, right? Who's at 10 billion? Come on.
This is interactive. Somebody stick their hand up.
10 billion. Good. 10 high end. Anybody more?
Okay, how much? Go for it.
5X that more? Not too much.
You, I mean, you went to the extreme.
It's 30 billion between the two of them. 17 and 12.
So 17 and 13 And 4 million of B of A is just on AI.
Yeah. Out of the 12 billion, 4 billion of that just for Bank of America is on AI And J.P. Morgan, I don't know what the exact number is, but I guarantee you it's a large number.
So let's assume they waste half of it, which they do.
Optimistic. It's still a large number.
And now you're competing against those guys who now have the chat bots, the products.
They're able to turn products around really quickly because this is how the AI works, is you start with the data, you create these dynamic products through the algorithm, and then you're feeding the product to the customer who's generating more data.
And it's a circle, right?
It's a virtuous, and they have, they're touching too many homes.
There's so many homes, they're getting so much data and they're constantly changing their products.
Now, I have a J.P. Morgan account, so I'm constantly getting, Hey, your car lease is up, or something like that.
And I don't even have a car lease with them.
How the hell did they find that out?
Well, there's insurance data, they connect to the dots, these dots around my head. That's what they're doing.
You know, it's a really interesting point.
You know, when, when I started Avant back in 2011, 2012, I think the feeling in the market was very much, Hey, like fintechs are gonna eat the banks' lunch, right?
Like, eventually we will own this product category.
And at the time there were a few major fintechs, right?
It was LendingClub, Prosper.
We were probably the third kind of major, you know, fintech player in the space.
And that didn't end up happening.
I think gradually we all realized that wasn't gonna happen.
So my thesis became sooner or later the banks will win because they own all of the customers.
They have access to all of the, you know, capital.
And that kind of didn't win either, because I mean, again, SoFi's been around for years and Yeah.
But so, a lot of companies have been around for years, right?
So I think the question is like, I'd love to hear your perspective, you know, given they've had at this point, you know, a decade and a half to react, why has the, you know, reaction been so lukewarm?
I, listen, I was an early believer in SoFi and a substantial investor.
I couldn't, I thought exactly what you were thinking.
I bought into Mike Cagney's, Kool-Aid.
I, you can have a great customer experience.
You can have some tech, but the funding is a big problem.
Absolutely. That constrains growth.
And then, you know, you just, it's a really difficult thing to build a bank.
Okay? You know, you need a license, you need to do all of that.
I know of only one company that has nailed it and they're coming to the U.S.
and that's a company called Newbank.
They're listed on the New York Stock Exchange.
Most people haven't heard of them, but it's a $60 billion market cap.
It's $5 billion less than PNCs. They're 12 years old.
They went from zero.
This is not a pitch for Newbank, but they went from zero customers to 105 million.
And this is the interesting stat.
Their revenue per customer per month is $11.
The cost to serve that customer per month is 90 cents.
Yeah.
The average cost to serve is for the banks in Brazil is compared to that.
Oh, it's outrageous. It's outrageous. It's not even close Because it's an oligopoly.
Yeah. It's four banks. Right.
And to be able to do that, I think it's scary that they're coming here at the time, Open Banking is happening.
Forget about J.P. Morgan.
I think J.P. Morgan and B of A and the likes are coming after your deposits.
Yeah. Right. 'cause J.P. Morgan has already said publicly.
They're going from 11% market share to 15.
And it's only a time, once it gets to 15, they're going to 20.
They're not coming. It's not, those deposits aren't coming from B of A, they're coming from everybody else.
Newbank.
I mean, listen, I don't know what their plans are for the U.S.
or anything like that, but what they've been able to do in Brazil, that is the role.
That's the model, that's the new AI bank.
And they went about it exactly the way in the opposite way to SoFi.
They went and bought a license.
They got a license, and they started from scratch. It can happen here And, in Mexico, by the way, they're really close.
They're in Mexico now And Columbia now. Yeah.
We have about 10 minutes left.
We definitely want to hear more, but I want to just pause and see if there are any questions or comments from the room.
Can you see him? Go ahead. Yeah. I was gonna say, the thing that strikes me the most about this, 'cause we've all been talking about it, how fast it's coming.
And I think most of us don't realize how fast it's coming.
And the story that I give back to that is, Yolande Piazza is the CEO of Google fintech, who, I went and followed her, or I'm sorry, yes — Citi FinTech. She left to go to Google now.
She has a payments for PayPal.
When she was there, remember when Alexa came out and everybody told us we're gonna go to voice banking.
Do you remember that nonsense? Oh yeah, by the way, yeah.
There was people on stage here telling you you're gonna use Alexa for voice banking.
Citi panicked, they brought all of their executives and business line owners to meet with her.
And she ran the meeting in New York, flew them all in, sat down and for 15 minutes they all argued about how Alexa was gonna change all their business models with small business banking and everything else.
And after 10 minutes Yo stood up and said, how many of you own an Alexa?
Anybody want to guess how many hands went up? Zero. Yeah.
Zero. Not a one.
She stopped the meeting, sent everybody home, bought everybody an Alexa. They came back a month later.
Well, they set the meeting for a month later, she sent an email out and said, does anybody think we need a meeting?
And the answer was no. This is completely different.
This will transform.
I think five years is probably a long horizon.
It is going to be ridiculous.
I don't think you understand the tsunami that's headed this way for our industry.
It's, you know, the, I think the problem is that there is a lot lost in translation when it comes to AI.
That's true. And it's like the blind leading the blind.
We need to figure out where we want to go first.
Okay? So if you want to go and say, Hey, I want to have a 360 degree view of my customer at all times, then set it as that and then figure out how to get there.
Okay. And the source of truth for that is your transaction data.
'cause every transaction in the world begins and ends with a bank pretty much.
And you have a very detailed understanding of behavior, income, volatility, spending behavior. Risk can be managed because a good borrower is somebody who's a responsible spender.
You can figure all that out.
This is so, you know, the way I think about it is if you want to be in the 360 area, you gotta figure out how to gain their trust and get more data from them.
How do you gain their trust? You do more business with them.
So everybody talks about product penetration. Do it.
Don't just sit on a deposit account and say, that's good enough because you'll lose that deposit account to Chase.
'cause they will take the auto loan and when they take the auto loan, they will know what that deposit account is doing and they'll market the s**t out of it.
That's how it's gonna work.
And they can do it at scale at much lower cost because of the IP.
So go back to the Newbank example that I was telling you.
Every new customer per month, cost to serve per customer month.
That is the new metric. That's the new KPI.
No bank uses it.
And I think that's kind of where, you know, traditional machine learning, AI is kind of meeting during the AI is one of the unique capabilities.
Again, I kind of keep going back to its ability to parse unstructured human language data.
Well, one of the values of that capability is structuring unstructured data.
So now all of a sudden you can kind of take data from a variety of different sources and create structured, usable data out of it.
I think transaction data is probably one of the most exciting kind of areas in new data development.
I know back in, you know, 2016, again, when I built my last model in 2016, we were just starting to kind of tap the beginnings of kind of transaction data, as a way to kind of add value to the underwriting fraud, loan assignment, line assignment process.
And I'm sure kind of the space has evolved, you know, since then.
But one of the challenges was always how do you take this kind of pseudo structured, unstructured data set and make useful variables and useful attributes and, and kind of do variable creation from it?
And that's the type of thing that generative AI can help you do except to like the nth degree.
You can now take completely unstructured data points, like voice data from your calls with customers.
Maybe somebody in passing mentions something like, Hey, I just lost my job.
Well, it's really hard to get your customer service agents to create a structured data point out of that conversation.
AI can do that across your entire database in a matter of kind of, you know, minutes or hours I, on transaction data.
No way can generative AI do that job and do it accurately because it's not trained on 267 characters.
Okay. Generative AI is trained on sequential language.
So anybody who thinks that ChatGPT can do categorization, I think, well, you can do it, but I would argue accuracy is not gonna be much better than 80%.
I definitely wouldn't use a GPT model for that.
Right? I'd start with like some sort of like small language model and, you know, transformer model and train from there.
You would be, you were talking about transaction data, but the other part of it is these GPT models, there's an open source issue and some of the best ones aren't.
So what are you gonna do? Send them your data.
That's gonna be another thing to deal with.
And I don't think you can do that with bank data.
I mean, banks are paranoid about everything and I would be the most paranoid about data given all the cybersecurity issues and so on and running a bank.
That's my biggest risk.
Yeah. I mean, I will say one of the hardest things for regulators, big picture, and I think most people are basically asleep at the wheel on this, is we are gonna digitize all this data.
It is going to deliver immense values to customers.
It is also probably true that in 15 years, most of it will have leaked somewhere or a lot of it will have leaked out, leaked somewhere.
It's really a hard thing to deal with, right?
I mean, and so, you know, we're buying this future for ourselves and I don't know that there's really any way to avert it, but it's also probably true that this is happening.
Yeah. I mean, wait till your supercomputers come in.
I mean Yeah, we didn't even talk about quantum computing supercomputers.
Right? And what is the, I mean, we're hearing noise about mid 2030, 2035, that's 10 years out.
But I have a lot of faith in humanity and the intelligence, and we are problem solvers.
You put a problem in front of us and we'll figure out how to solve it.
And I think every problem is solvable.
It takes time. Just somehow society is still running despite all the data leaks.
Right? Like, it's kind of a, despite us trying not to do our best.
So we have just a few minutes left.
Any questions in the room before I have one last question.
I'm gonna ask the panel.
I have a question for the audience. I would love that.
How many of you guys are running banks and working at banks?
Sorry, working what? No, running banks.
Because I mean, you, I wanna ask you how you guys are making money?
They're all out there Skiing. Yeah.
You see the most important question is how do you make money?
And this is an interesting discussion on how to make money.
Where is everybody by the way?
Spa. Spa. Enjoying the weather.
Petting that dog. D**n dog. Oh, there he is. D**n dog is here. Don't talk bad about the dog. So we need a lamer location next time.
Is that what I'm hearing? Oh God.
My question is, where do y'all, where does this group see AI playing in the decentralization environment and space?
In what way, Brett?
Well, as the consumer begins to, own their data. Own their data. Oh, okay.
I think my personal view on that is, you need to own your data.
And that data is yours. It's an asset. It's your asset.
I fundamentally believe in that.
I just, you know, don't know how it's actually gonna happen.
I mean, obviously it's gonna be on a block and things like that, but how we monetize that is gonna be really the key.
I actually think that in the future state, we're gonna have data hedge funds because it's an asset, a really valuable asset of yours.
I think it's, I think part of this is gonna come faster than you think and it's gonna come sideways.
So the example is, how many know about the United Healthcare hack that took place in February?
I love that. Most of the room hasn't heard about this.
So do you know how much money moves in healthcare payments in the U.S? We're talking payments.
This is what we do. How much moves a year?
3.4 trillion?
4.5 trillion. Do you know I’m ex-military?
Do you know how much we spend on our military?
It's 3X what we spend on our military moves.
Do you know how much money was moving in February?
Not a fricking dime. Zero. Nothing.
Remember when you went to the pharmacy and you couldn't get your stuff?
All right. You know, you're like, what's going on?
And they were faxing in all of this.
So I know you think banking and payments is bad.
Oh my God. You are gonna have that.
That's national infrastructure. That's national security.
By the way, Russian hack everybody just so you know, black cat and hardly got any attention.
And what if that happened in banking? Seriously.
Yeah, Seriously.
I mean, you see what I'm saying?
And that's just digitized again. Identity. Yeah.
Everybody's stolen everything about that.
And it basically shut down one of the most important industries in the U.S.
I think part of the problem is that 70% of the market is dominated by Epic.
And the DOJ should be investigating them for antitrust. Well, it should have been. And it yeah, it's a whole, Hey, how you doing? Well, We're digressing a little. It's the government guy. Well, no, I mean it's, it is true that one of the consequences of AI is introducing more single points of failure across the economy.
I mean, there are lots of forces making that happen, but traditional antitrust is all about harm to consumers, not about national security.
And so there's this, I've been in lots of discussions about how do we develop a new antitrust doctrine. Yep.
That actually accounts for national security, not just, you know, price harms to consumer. Yeah. Because the New War is digital. Yeah.
Thank You. All right, well, we are at time.
Thank you all so much. This has been fascinating.
I, we had lunch together and I said, a really great panel, the moderator doesn't have a job at all because they, do the job for you.
So I hope you all appreciated that discussion among this group.
Thank you for moderating. Thank you So much for your time. Yeah, thank you. And we have, happy hour this afternoon.
We have dinner tonight, so there'll be lots of opportunities for you to grab one of these gentlemen.
One-on-one and to ask questions. So Thank you all for coming and thank you to MX for hosting this.
Really appreciate it.
Are We Dabbling with or Deploying AI in Banking?
We’ve all seen, heard, or dabbled with artificial intelligence by now. In fact, a recent report says 91% of financial services organizations are assessing or using AI today. But, what does that look like? Let’s explore where the rubber meets the road when it comes to AI in banking — use cases, practical applications, and more.
Explore MX Products
Learn More About MX
Watch Next
Adrienne Harris, New York State Dept. of FInancial Services
Fireside Chat with Adrienne Harris
Dr. Leda Glyptis, PhD
The Digital Transformation of Financial Services
Rebel
Finances, Feelings, and Fears
Transcript
Are we dabbling in AI? Are we doing this thing?
And our panel here, I'll have them introduce themselves, but, what they're going to talk about is some of those practical use cases for AI.
So we'll go just down the line and introduce ourselves.
I'll go really quickly. I'm not the most important person up here.
I'm here to get you guys to give us the real good nuggets.
But really quickly, I'm Crystal Anderson and I am the Chief of Staff at MX.
Prior to that, I headed product at MX for two years.
I've been here for three years.
And prior to that I was at H&R Block where I led the financial services business.
I was vice president of financial services.
We loaned $1.8 billion a year and had $29 billion in deposits through our financial services program.
So I've been in this space for quite some time.
I'll have you go next.
Awesome. Yeah. Thank you. Crystal.
John Sun, co-founder and CEO at Spring Labs.
My background is in FinTech lending.
So before Spring Labs, I was the co-founder and Chief Risk Officer at Avant, where I oversaw credit risk, fraud risk, compliance, risk, you name it, as part of the overall risk stack.
My training is really as a data scientist, that's kind of what I always kind of go back to.
So at Avant, my team and I built one of the first tree-based kind of machine learning algorithms for real time underwriting back in 2012.
I at least haven't heard of another company that's deployed anything earlier than that.
But again, a little out of practice.
I haven't probably built a model since 2016 or reviewed the model since 2018.
So my knowledge, I'm sure is out of date compared to some of these guys on the panel here.
And, today, what we're doing at Spring Labs is making AI accessible to financial institutions.
When I say AI, I mean, in this case, more generative AI.
How do we use the conversational intelligence capabilities of generative AI to help banks and financial institutions with regulatory risk reduction, with operational efficiency, with better kind of staffing and better engagement with customers to kind of create more delightful experiences. Great.
Thank you John. Sam? Hey everybody. I'm Sam Maule.
I started in banking and payments back in ‘94, so I've been doing this for a little while.
Been a banker, been a worked at TSYS, did consulting, did a stint at Google, and now I'm with a payments company called Moov.
So this ought to be fun diving into this. Yeah.
I'm Zach Boyd, Director of Utah's Office of AI Policy, newly created four months ago.
Also a math professor at Brigham Young University, where I've been doing research basically into foundations of machine learning and applications in social science.
Wije. The most important person in the room here, And least knowledgeable about finance for sure.
My name is Wije.
I'm co-founder and CEO of a company called Aliya.
We build AI-powered solutions for banks.
Great. Well, my first question's going to be, tell me a bit more about what your organizations are doing with AI.
So Wije I am actually gonna start with you. What does that mean?
You promised me I would go last.
That's why you sat down there.
That's why. Better I know.
What do we do? In this?
The simplest way of explaining it is we take the most valuable data a bank has, which is the transaction data, and we support lending and risk management using that data.
That's in the simplest way.
We go through the ringer in terms of, well, we've been doing it for six years now.
We've done, you know, close to $7 billion worth of loans.
We work with a very large bank partner.
And it's been a learning process for us in terms of the biggest constraints to deploying AI.
Yeah. We've been through it living it, regulatory, compliance, legal, and all of that is what we do. Great.
Sam, why don't we hear from you next? Sure.
What are we doing? So with Moov, it's everything about money movement.
So money in money out, storing all that stuff.
So basically just think about all that transaction data and, I think I said I worked at Google, so shockingly we're doing a lot with Google on the BigQuery side, Looker, and the other tools that we can tie into.
So, I mean, transaction data is the holy grail when it comes to payments, right?
That's what we all wanna see. So you can, yeah.
Yeah. Use your brain. Think What we're doing. Use your brains, use your brain, everybody.
Zach, I'm really interested in this newly formed, tell us some context around this newly formed organization that you're leading and kinda the impetus to the creation of that.
Yeah, sure. Happy to discuss.
So, Utah's Office of AI Policy is unique in the United States and maybe in the world now.
Utah moved first. The idea was to have an in-house government capability where there's an office with people who understand AI are following the dozens of relevant regulatory issues that might affect the state and are primed and ready to make good recommendations to the legislature if and when it seems right for the state to act on AI policy.
So this is mostly regulatory policy, certainly in finance and banking, in healthcare, schools, you name it.
The legislature specifically wants us to be wearing both the consumer protection hat and the innovation fostering hat, which I think has really been super effective so far because usually those two people wear different hats and it leads to a lot of dysfunctions in various regulatory systems.
So I think that's been really good.
We have a charge to go through topically and do deep dives on matters of importance.
So, for example, we could do a deep dive on AI and fintech and banking and think through, you know, what does the right regulatory structure need to be in the age of AI?
How can we make sure things make sense, and then make reasoned recommendations to the legislature going through topically.
We'll take a lot of time and AI moves fast.
So the legislature also granted us a regulatory mitigation authority, which is basically a very kind of nuanced sandbox authority.
If a company wants to use AI in the state to simplify slightly, I can write a contract to any company that wants to operate in the state, clarifying that the regulations of the state shall apply to you thusly, or that you will be exempted from the following regulations of the state.
Or we'll give you a curing period, or we'll cap fees.
And so this applies across all the regulations of the state.
So we've had some companies approach us already, but the idea is we want companies to be able to explore innovative business models and we want Utahns to reap the benefits of AI as rapidly as possible.
How big is the team? Sorry, I had to ask, ask How big is the Team? Oh, I was actually gonna ask you about your facial expression, Sam.
So no, I'm — How big is the team? Oh, I mean, we're a small government office.
I have six people. It's all right. So, no, I mean, it's actually pretty good. It’s six people doing what you’re doing.
That's awesome. Yeah. I mean, but my philosophy is always, I've tried to pack as much knowledge as possible into this small state office as I can, but we are not now, nor ever will be resourced enough to have deep in-house expertise on every subject.
So we have a model of stakeholder engagement.
I've spent a lot of my time trying to find the smartest people that I possibly can in all the relevant industries and engage with them to rapidly acquire the most important facts.
Sam, so you pointed at John and Wije.
Tell me what makes you say they're the smartest people in using AI in the industry? Because they're buried in it and been doing it quietly, looking at you quietly.
When people work quietly and aren't all over social media talking about how incredible they are, I tend to, they're the people grinding this out.
And this takes a long time.
I'll speak for Wade Arnold, who I work for at Moov.
When you're building out payment rails from, you know, bare metal connections all the way up, it takes a long time and it's really, really hard.
What we're talking about is transformative, what y'all are doing. Yeah, Thank you. You're welcome.
Hey, Wije, can I have you speak into the mic just so that the recording can hear you?
I thought my voice is loud enough that everybody that no, anything that is as transformational as this is really, really hard.
People completely underestimate how difficult it is, the cost of doing it.
I mean, it's prohibitive, right? Compute costs now.
It's crazy. I mean, we had to build our own stack of Nvidia H100s to mitigate some of that.
The talent, it's very hard to find.
They, the people who do this are really very special people.
They're kind of not normal because they're just, they're not data scientists in a traditional way.
They're kind of artists.
'cause they really just know something that we don't normally.
The other part of it is, you know, the regulatory lift.
But I think in terms of one of the biggest hurdles is finding the right leadership with the right strategy.
Because everybody talks about AI.
In fact, I had my team just run all of the annual reports of everybody here looking at how many times AI machine learning and artificial intelligence I mentioned in the reports.
So you got J.P. Morgan up at the lit top, huge number of times that they talk about it.
And then the vast majority are about one once, you know.
So, but there is a lot of conversation going on about it.
And, but the strategic direction is the really important, it's like everybody thinks of point solutions.
It's not a point solution, it's a curated offering, right?
Even MX, which is one of the best in the business, is beginning to realize that it's a curated solution that has a closed loop to it that makes money.
And banks need money.
Their NII, I mean, I'm speaking for you guys, but you know it better than I do.
Pretty much sucks, can't grow. What are you gonna grow?
And that's the problem AI can solve because you can get deeper into your customer.
You can figure out how to price them more accurately and you can make better loans to more people, right.
As is the single.
Yeah, I mean, I would echo that sentiment, although Sam, you know, the reason we don't do social media is because we're bad at it, not because for lack of trying.
But I think that's actually, you know, Wije, you bring up a really good point, which is, you know, every time there's a new technology that comes about, people are like, oh, AI this, AI that, right?
At the end of the day, a good AI product shouldn't be defined by the AI-ness of it.
It should be defined by its value to your organization, right?
Ultimately, that's what matters. And I feel like at the beginning of every single kind of tech cycle, you see a lot of investment into the AI-ness of products instead of investment into ROI investment into how it moves your organization forward.
So that's something that we try to focus on at Spring Labs is how do we build products that, again, present efficiency, delightful experiences to your customers, and ultimately add your, move your business forward and add value to your customers where the AI of it is kind of a secondary. Can I ask you a question real quick?
Of course you can. Just real quick, I, I like what you said about don't think data scientists think artists and all that.
Yeah. Because I've heard it and you know, stated when we're talking about AI, it's like learning an alien language or can you emote on that a little bit?
Yeah. It's, I, most people think of AI and data science in a, in query form that is so one dimensional, right?
Because it's about, think of it like you're sitting there and you've got a trillion dots around you, around your head, and it's kind of like, how do I connect these dots to make some sense of that?
And that's a very complex thing and that's what these guys do.
If you can imagine that.
And the power of it is like in categorization, you guys are all familiar with transaction data and categorization.
When a category comes out, you can't explain how it got there.
I mean, it's very, very difficult 'cause it's multidimensional.
So it becomes a problem for us vis-a-vis the regulators when the compliance component of it is, well how did it get there?
You know, it's like us trying to explain a multidimensional framework in 2D, and this is where I think the most amount of work needs to be done, is a collaboration of people like us with the regulators, the sandboxes, or we talked about clinical trials, right?
In the medical industry, there is a lot of AI that's used to analyze data and come up with protocols that have to go through clinical trials.
And the FDA process. We should be thinking along those lines and we will.
Yeah. And I'll say that, you know, not just in banking and fintech, but across the board I'm seeing a transformation where before we're so used to humans being able to give us explanation for things.
And, in some sense that, you know, it's a value to the consumer to be able to be given an explanation for things if those explanations are actually valid, right?
Like they derive subjective benefit from this and sometimes hard concrete benefits from it.
But now I, you know, it's gonna take society a little while to realize that with these, with this kind of model, there's a complex trade off between the value of being able to explain things and the capability of being able to do better, right?
Like the best models right now are the least explainable.
And I know some people are trying to get around compliance requirements or to meet compliance requirements by taking the AI and tacking some explanatory output onto it.
And I, you know, I am sympathetic.
I don't think that this is the biggest priority to go after as a regulator at this very moment because it is the state of the technology and people feel like they need an explanation.
But I think scientifically it's actually super unclear that these explanations actually are explaining what the decision making process of the AI actually is.
What I hope society gets to is to the point where we realize there's just a value trade off here, right?
Like we can have enhanced capabilities at the expense of less explanation and we can accept something along this curve of how much explanation we really need.
Talk to me about some of those use cases.
So we've heard lending and pricing and improving the experience.
What are some of the most impactful use cases you're seeing in financial services? We'll start with you John. Yeah, great question.
I mean, practical applications, obviously is the key to success for any technology.
And looking at it from the more generative AI side of things, and I'm sure kind of Wije and the rest of the folks have the opinion on the more traditional machine learning AI, generative AI is uniquely good at understanding human language.
So these are pre-trained models that have such a vast corpus of data.
You can manipulate language questions and kind of constructs without, a ton of additional training.
So I would say look around your organization and just see where there's inefficiencies that involve a bunch of kind of language related tasks.
I mean, the ones that we're seeing that's really kind of interesting.
Compliance is a good one.
It's always one of these areas where, you know, even if you're a tech forward organization, you never say, let's invest in compliance tech.
You say, let's go throw more bodies at it.
So here now you're presented with a technology that can all of a sudden change the game in terms of how you interface with your customers, how you hear what your customers are telling you, how you turn those insights into kind of structured data to better inform technology process and roadmap decisions.
So those types of things are the types of use cases that generative AI is uniquely kind of good at.
I'm gonna give an example that's not banking, if that's okay.
That's okay. So from the Google days, so in Canada, especially in British Columbia, they went through a period of excessive drought, ton of forest fires.
Most of those forest fires were caused by truckers, long haul truckers going because of the speed of what they travel.
We create sparks, we create fire and everything else.
So this gets to your multi-layered data elements all coming together at the same time.
So Google, through satellites and weather patterns and through trackers that they could put on the trucks could actually say, alright, here are the weather conditions for specifically where you're at in real time as you're traveling.
Because what happened was the government came in and said, all right, you have to drive under 45 miles an hour that yeah, I know, but to stop the forest fires because of what we're in.
So just the rule or, or the regulation is gonna be, you gotta drive under 45. In trucking and logistics, that's, you're dead, right? That's killer.
So Google was able to come in and work with them to say, okay, we can go and monitor each individual driver, the trucks themselves, the speed that they're going, where they're at with satellite data in real time and what the weather conditions are.
And we can modulate at what speed you need to go in real time, which now is impacting dollars and everything else.
So if you think about that, that's a ton of different data elements all coming together, which was driving profitability.
And by the way, that's like five years old.
That's not something that's new.
So, you've gotta start thinking in ways you haven't thought before, getting back to the alien type brain. Right?
Yeah. I mean, I think you make a very important point. Good.
That was the only one I'm gonna get for this entire talk.
I've been saving that one up.
Take everything that you've been doing for the last couple of decades because it really has been based on not having to take any risk.
When interest rates are zero, you don't do anything with your customer.
You shouldn't be doing anything with your customer.
You should be taking your cheap dollar cheap funding, deposits and going and buying long duration treasuries.
That's what happened.
And it has not created any innovation.
And then now you're getting overloaded by this thing called AI and it's confusing. A lot of people take a blank sheet of paper, re-engineer the whole thing, and your most valuable piece of data, there are three pieces of data that you need to focus on.
One is the transaction data, your bureau data and public records data.
If you know how to use those effectively to serve your customers, you're going from zero to 60 in no time.
Forget about zero to a hundred.
That's what the zero to 100 is what highly paid management consultants will tell you.
'cause they'll go, oh yeah, you need to create the orchestration layer, the data layer.
You need to create all of these things. It's h******t.
Sorry, promised I wouldn't do that. Keep it simple.
'cause banking — everybody's made banking complex.
If you take away the complexity, it's a pretty simple thing.
Take in deposits from your community, lend it out safely, make a spread.
That's what the social mandate is.
And you need to use the best tools available to you.
And AI is totally transformational because you're gonna be able to predict behavior losses.
If you can do all of that, you can price it more accurately rather than guessing.
There's, if you price it correctly, size it correctly, then you make a lot of money.
Can I ask a question actually of the other panelists?
I know in other industries, I tell people, you know, you've got these weird strategic decisions right now.
'cause AI, especially generative AI, is currently the worst that it will ever be, right?
It will only improve from here, all kinds of hypotheses about how much or how little it will improve.
And, I think for a lot of people, this has a strategic bearing, but I am hearing you guys basically talking about the benefits that are already here.
Yeah. What's your opinion?
Like, how much are the benefits already here versus how much do you think is coming?
Do you guys have like, theories about this?
Sorry, John, but compliance is a bunch of rules.
Well, I mean, it shouldn't be from my perspective.
It shouldn't be interpreted.
Okay. Like it compliance is, you know, So don't — You're saying like, you don't need generative AI to do compliance, in my opinion.
It's a set of rules.
You can build it into ones and zeros.
I hope somebody's not agreeing with me on that one. We'll give you a mic in a little bit and Look, we, yeah, we're doing it.
So with it, with a large OCC regulated bank, it's built in, you can build it into the workflow.
Yeah, I mean, I think we might be talking about two different kinds of compliance.
I mean, there's the type of compliance which is, you know, let's analyze a particular model for a particular outcome or a particular set of inputs.
And then there's the compliance of, you know, I just sent out a marketing mailer to this comply with all of my obligations under, you know, various rules that I have for marketing this product.
I just received a complaint from a customer.
Did the, you know, three pages of like personal life story and like my dog just died and all of that.
Did any of that trigger a UDAP concern? Right?
And again, can you do it with traditional machine learning?
Yes, absolutely.
It would be incredibly time consuming and cost inefficient.
What generative AI has presented us is a set of models where you don't have to start from zero.
So if you think about what GPT stands for, it stands for generative pre-trained transformer models, right?
Everyone focuses on the G, which is generative.
What you should be focused on is the P, which is this pre-trained.
It comes out of the box with a vast corpus of human knowledge.
And you can now use that as a springboard to do way more advanced use cases than you could if you just started with the transformer model from zero.
Yeah. As a former Google, I think the transformer part's very important because of my stock.
Let's just make sure that gets said.
I wanna make sure we hear — I saw lots of Yes.
That latter example is what I align with.
So tell us more about that. Absolutely.
Can you introduce yourself real quickly?
Not a problem. So I'm Doug Milow, senior Director of Credit and Fraud Risk Strategy.
I'm also the fair lending officer for Best Egg.
We're a consumer lending platform.
Here's where I respectfully disagree and agree with John.
It's complaints, right?
We've got so much data we collect to see how we're doing.
We've got online surveys, we've got surveys from our funnel, we've got customers talking to agents.
You've got that agent shorthand that they love to write in to describe what happens.
We've got the tracking of escalations to the management layer, we've got the regulatory complaints coming in.
We've got all kinds of signals and it's just completely unstructured.
And that's where Gen AI can really shine.
We're we're working with the Amazon, you know, platform to kind of figure out where to go with that.
So wasn't talking about, I should have clarified my point.
Got it. I was more about serving the customer, not the complaint side of it, but Yeah, some of, some of those. Good point. Yeah, totally agree with you on that part.
You know, it's, Hey, if your APR is above X and your state says it can only go to Y, you've got a problem.
That's a rule. We agree.
Hey, Wije, can you expand a little bit?
'cause you touched on it, at, Moov, we had Matt Harris from Bain Capital, if you know Matt, he's been doing this forever.
The OG when it comes to this.
And he talked about AI and agents and some other components and how it's just gonna wreck havoc with NIM.
I mean, you mentioned that before on the net, you know, can you expand a little bit on what you were saying On the NIM? Yeah.
Look, the problem, the problem is in order to generate the name you or NII you, you need to be able to lend, most organizations are struggling to figure out how to lend one of the spaces is in the Best Egg space.
We should be going after all of the Best Egg customers because they're taking your customers by the way.
And the reason is they're really good at predicting losses, pricing, and getting to the customer.
And the banking system isn't, and because of that, you've left a lot of room available to the fintechs to thrive.
SoFi who, you know, I was on the board of for five years, they did $11.5 billion of unsecured personal loan originations last year.
Okay? I don't know how much you guys did, but you've done billions and like total of 30 billion, if I'm not mistaken in your lifetime.
That's business that belongs to the banking system.
I mean, just bank because you already have the customers by the way, they had to go get their customers.
And that's, I mean, that's money, right?
So that's one aspect of, the other aspect of it is in, Because CAC is a big deal.
It's a big deal. I mean, that's huge.
The other aspect of it is your losses.
You haven't invested in what companies like Best Egg have invested in.
So your losses are higher.
Then the other part of it is your production costs.
I mean, 27 screens, and going to the branch to close a loan And on legacy tech that was built pre-internet for your course not to go down that path too much. But we know that, I don't mean, I'm being, matter of fact here, this is not a criticism of anybody.
It's just lay of the land and we're experiencing something right now in terms of AI that is totally transformational.
That's why I think it needs a blank sheet of paper approach.
First principle thinking, but that's hard to do.
So can it be transformational without the blank slate, right?
If the regulatory structure doesn't change, what happens?
I think this is a Google / Microsoft type situation where the big guys are gonna win.
I was gonna ask you guys this question.
How much do you think Chase and B of A are spending on tech this year?
10 billion, right? Who's at 10 billion? Come on.
This is interactive. Somebody stick their hand up.
10 billion. Good. 10 high end. Anybody more?
Okay, how much? Go for it.
5X that more? Not too much.
You, I mean, you went to the extreme.
It's 30 billion between the two of them. 17 and 12.
So 17 and 13 And 4 million of B of A is just on AI.
Yeah. Out of the 12 billion, 4 billion of that just for Bank of America is on AI And J.P. Morgan, I don't know what the exact number is, but I guarantee you it's a large number.
So let's assume they waste half of it, which they do.
Optimistic. It's still a large number.
And now you're competing against those guys who now have the chat bots, the products.
They're able to turn products around really quickly because this is how the AI works, is you start with the data, you create these dynamic products through the algorithm, and then you're feeding the product to the customer who's generating more data.
And it's a circle, right?
It's a virtuous, and they have, they're touching too many homes.
There's so many homes, they're getting so much data and they're constantly changing their products.
Now, I have a J.P. Morgan account, so I'm constantly getting, Hey, your car lease is up, or something like that.
And I don't even have a car lease with them.
How the hell did they find that out?
Well, there's insurance data, they connect to the dots, these dots around my head. That's what they're doing.
You know, it's a really interesting point.
You know, when, when I started Avant back in 2011, 2012, I think the feeling in the market was very much, Hey, like fintechs are gonna eat the banks' lunch, right?
Like, eventually we will own this product category.
And at the time there were a few major fintechs, right?
It was LendingClub, Prosper.
We were probably the third kind of major, you know, fintech player in the space.
And that didn't end up happening.
I think gradually we all realized that wasn't gonna happen.
So my thesis became sooner or later the banks will win because they own all of the customers.
They have access to all of the, you know, capital.
And that kind of didn't win either, because I mean, again, SoFi's been around for years and Yeah.
But so, a lot of companies have been around for years, right?
So I think the question is like, I'd love to hear your perspective, you know, given they've had at this point, you know, a decade and a half to react, why has the, you know, reaction been so lukewarm?
I, listen, I was an early believer in SoFi and a substantial investor.
I couldn't, I thought exactly what you were thinking.
I bought into Mike Cagney's, Kool-Aid.
I, you can have a great customer experience.
You can have some tech, but the funding is a big problem.
Absolutely. That constrains growth.
And then, you know, you just, it's a really difficult thing to build a bank.
Okay? You know, you need a license, you need to do all of that.
I know of only one company that has nailed it and they're coming to the U.S.
and that's a company called Newbank.
They're listed on the New York Stock Exchange.
Most people haven't heard of them, but it's a $60 billion market cap.
It's $5 billion less than PNCs. They're 12 years old.
They went from zero.
This is not a pitch for Newbank, but they went from zero customers to 105 million.
And this is the interesting stat.
Their revenue per customer per month is $11.
The cost to serve that customer per month is 90 cents.
Yeah.
The average cost to serve is for the banks in Brazil is compared to that.
Oh, it's outrageous. It's outrageous. It's not even close Because it's an oligopoly.
Yeah. It's four banks. Right.
And to be able to do that, I think it's scary that they're coming here at the time, Open Banking is happening.
Forget about J.P. Morgan.
I think J.P. Morgan and B of A and the likes are coming after your deposits.
Yeah. Right. 'cause J.P. Morgan has already said publicly.
They're going from 11% market share to 15.
And it's only a time, once it gets to 15, they're going to 20.
They're not coming. It's not, those deposits aren't coming from B of A, they're coming from everybody else.
Newbank.
I mean, listen, I don't know what their plans are for the U.S.
or anything like that, but what they've been able to do in Brazil, that is the role.
That's the model, that's the new AI bank.
And they went about it exactly the way in the opposite way to SoFi.
They went and bought a license.
They got a license, and they started from scratch. It can happen here And, in Mexico, by the way, they're really close.
They're in Mexico now And Columbia now. Yeah.
We have about 10 minutes left.
We definitely want to hear more, but I want to just pause and see if there are any questions or comments from the room.
Can you see him? Go ahead. Yeah. I was gonna say, the thing that strikes me the most about this, 'cause we've all been talking about it, how fast it's coming.
And I think most of us don't realize how fast it's coming.
And the story that I give back to that is, Yolande Piazza is the CEO of Google fintech, who, I went and followed her, or I'm sorry, yes — Citi FinTech. She left to go to Google now.
She has a payments for PayPal.
When she was there, remember when Alexa came out and everybody told us we're gonna go to voice banking.
Do you remember that nonsense? Oh yeah, by the way, yeah.
There was people on stage here telling you you're gonna use Alexa for voice banking.
Citi panicked, they brought all of their executives and business line owners to meet with her.
And she ran the meeting in New York, flew them all in, sat down and for 15 minutes they all argued about how Alexa was gonna change all their business models with small business banking and everything else.
And after 10 minutes Yo stood up and said, how many of you own an Alexa?
Anybody want to guess how many hands went up? Zero. Yeah.
Zero. Not a one.
She stopped the meeting, sent everybody home, bought everybody an Alexa. They came back a month later.
Well, they set the meeting for a month later, she sent an email out and said, does anybody think we need a meeting?
And the answer was no. This is completely different.
This will transform.
I think five years is probably a long horizon.
It is going to be ridiculous.
I don't think you understand the tsunami that's headed this way for our industry.
It's, you know, the, I think the problem is that there is a lot lost in translation when it comes to AI.
That's true. And it's like the blind leading the blind.
We need to figure out where we want to go first.
Okay? So if you want to go and say, Hey, I want to have a 360 degree view of my customer at all times, then set it as that and then figure out how to get there.
Okay. And the source of truth for that is your transaction data.
'cause every transaction in the world begins and ends with a bank pretty much.
And you have a very detailed understanding of behavior, income, volatility, spending behavior. Risk can be managed because a good borrower is somebody who's a responsible spender.
You can figure all that out.
This is so, you know, the way I think about it is if you want to be in the 360 area, you gotta figure out how to gain their trust and get more data from them.
How do you gain their trust? You do more business with them.
So everybody talks about product penetration. Do it.
Don't just sit on a deposit account and say, that's good enough because you'll lose that deposit account to Chase.
'cause they will take the auto loan and when they take the auto loan, they will know what that deposit account is doing and they'll market the s**t out of it.
That's how it's gonna work.
And they can do it at scale at much lower cost because of the IP.
So go back to the Newbank example that I was telling you.
Every new customer per month, cost to serve per customer month.
That is the new metric. That's the new KPI.
No bank uses it.
And I think that's kind of where, you know, traditional machine learning, AI is kind of meeting during the AI is one of the unique capabilities.
Again, I kind of keep going back to its ability to parse unstructured human language data.
Well, one of the values of that capability is structuring unstructured data.
So now all of a sudden you can kind of take data from a variety of different sources and create structured, usable data out of it.
I think transaction data is probably one of the most exciting kind of areas in new data development.
I know back in, you know, 2016, again, when I built my last model in 2016, we were just starting to kind of tap the beginnings of kind of transaction data, as a way to kind of add value to the underwriting fraud, loan assignment, line assignment process.
And I'm sure kind of the space has evolved, you know, since then.
But one of the challenges was always how do you take this kind of pseudo structured, unstructured data set and make useful variables and useful attributes and, and kind of do variable creation from it?
And that's the type of thing that generative AI can help you do except to like the nth degree.
You can now take completely unstructured data points, like voice data from your calls with customers.
Maybe somebody in passing mentions something like, Hey, I just lost my job.
Well, it's really hard to get your customer service agents to create a structured data point out of that conversation.
AI can do that across your entire database in a matter of kind of, you know, minutes or hours I, on transaction data.
No way can generative AI do that job and do it accurately because it's not trained on 267 characters.
Okay. Generative AI is trained on sequential language.
So anybody who thinks that ChatGPT can do categorization, I think, well, you can do it, but I would argue accuracy is not gonna be much better than 80%.
I definitely wouldn't use a GPT model for that.
Right? I'd start with like some sort of like small language model and, you know, transformer model and train from there.
You would be, you were talking about transaction data, but the other part of it is these GPT models, there's an open source issue and some of the best ones aren't.
So what are you gonna do? Send them your data.
That's gonna be another thing to deal with.
And I don't think you can do that with bank data.
I mean, banks are paranoid about everything and I would be the most paranoid about data given all the cybersecurity issues and so on and running a bank.
That's my biggest risk.
Yeah. I mean, I will say one of the hardest things for regulators, big picture, and I think most people are basically asleep at the wheel on this, is we are gonna digitize all this data.
It is going to deliver immense values to customers.
It is also probably true that in 15 years, most of it will have leaked somewhere or a lot of it will have leaked out, leaked somewhere.
It's really a hard thing to deal with, right?
I mean, and so, you know, we're buying this future for ourselves and I don't know that there's really any way to avert it, but it's also probably true that this is happening.
Yeah. I mean, wait till your supercomputers come in.
I mean Yeah, we didn't even talk about quantum computing supercomputers.
Right? And what is the, I mean, we're hearing noise about mid 2030, 2035, that's 10 years out.
But I have a lot of faith in humanity and the intelligence, and we are problem solvers.
You put a problem in front of us and we'll figure out how to solve it.
And I think every problem is solvable.
It takes time. Just somehow society is still running despite all the data leaks.
Right? Like, it's kind of a, despite us trying not to do our best.
So we have just a few minutes left.
Any questions in the room before I have one last question.
I'm gonna ask the panel.
I have a question for the audience. I would love that.
How many of you guys are running banks and working at banks?
Sorry, working what? No, running banks.
Because I mean, you, I wanna ask you how you guys are making money?
They're all out there Skiing. Yeah.
You see the most important question is how do you make money?
And this is an interesting discussion on how to make money.
Where is everybody by the way?
Spa. Spa. Enjoying the weather.
Petting that dog. D**n dog. Oh, there he is. D**n dog is here. Don't talk bad about the dog. So we need a lamer location next time.
Is that what I'm hearing? Oh God.
My question is, where do y'all, where does this group see AI playing in the decentralization environment and space?
In what way, Brett?
Well, as the consumer begins to, own their data. Own their data. Oh, okay.
I think my personal view on that is, you need to own your data.
And that data is yours. It's an asset. It's your asset.
I fundamentally believe in that.
I just, you know, don't know how it's actually gonna happen.
I mean, obviously it's gonna be on a block and things like that, but how we monetize that is gonna be really the key.
I actually think that in the future state, we're gonna have data hedge funds because it's an asset, a really valuable asset of yours.
I think it's, I think part of this is gonna come faster than you think and it's gonna come sideways.
So the example is, how many know about the United Healthcare hack that took place in February?
I love that. Most of the room hasn't heard about this.
So do you know how much money moves in healthcare payments in the U.S? We're talking payments.
This is what we do. How much moves a year?
3.4 trillion?
4.5 trillion. Do you know I’m ex-military?
Do you know how much we spend on our military?
It's 3X what we spend on our military moves.
Do you know how much money was moving in February?
Not a fricking dime. Zero. Nothing.
Remember when you went to the pharmacy and you couldn't get your stuff?
All right. You know, you're like, what's going on?
And they were faxing in all of this.
So I know you think banking and payments is bad.
Oh my God. You are gonna have that.
That's national infrastructure. That's national security.
By the way, Russian hack everybody just so you know, black cat and hardly got any attention.
And what if that happened in banking? Seriously.
Yeah, Seriously.
I mean, you see what I'm saying?
And that's just digitized again. Identity. Yeah.
Everybody's stolen everything about that.
And it basically shut down one of the most important industries in the U.S.
I think part of the problem is that 70% of the market is dominated by Epic.
And the DOJ should be investigating them for antitrust. Well, it should have been. And it yeah, it's a whole, Hey, how you doing? Well, We're digressing a little. It's the government guy. Well, no, I mean it's, it is true that one of the consequences of AI is introducing more single points of failure across the economy.
I mean, there are lots of forces making that happen, but traditional antitrust is all about harm to consumers, not about national security.
And so there's this, I've been in lots of discussions about how do we develop a new antitrust doctrine. Yep.
That actually accounts for national security, not just, you know, price harms to consumer. Yeah. Because the New War is digital. Yeah.
Thank You. All right, well, we are at time.
Thank you all so much. This has been fascinating.
I, we had lunch together and I said, a really great panel, the moderator doesn't have a job at all because they, do the job for you.
So I hope you all appreciated that discussion among this group.
Thank you for moderating. Thank you So much for your time. Yeah, thank you. And we have, happy hour this afternoon.
We have dinner tonight, so there'll be lots of opportunities for you to grab one of these gentlemen.
One-on-one and to ask questions. So Thank you all for coming and thank you to MX for hosting this.
Really appreciate it.
View Full Transcript