Cornell Keynotes

Disrupting Traditional Banking: The AI-Powered Fintech Revolution

Episode Summary

This podcast discusses how Varo Bank is using AI and machine learning to serve underserved populations by reducing costs and better assessing risk through real-time cashflow data rather than traditional credit scores. The conversation demonstrates how AI-powered fraud detection, personalized lending, and automated customer service are democratizing financial services for the 180 million Americans living paycheck to paycheck. Don't miss this opportunity to learn how AI is shaping the future of financial services.

Episode Notes

AI is transforming industries worldwide, with banking at the forefront of this revolution. The fintech sector offers unprecedented opportunities to modernize financial services and make them more accessible to everyone.

Join us for an insightful discussion with two industry leaders: Lutz Finger from Cornell University and Colin Walsh, Founder and Board Director of Varo Bank — the first fintech company to receive a U.S. national banking license.

Our panel will explore how AI is revolutionizing banking, ways to expand financial access through technology, opportunities for disruption in fintech, and why successful implementation matters more than the technology itself.

Don't miss this opportunity to learn how AI is shaping the future of financial services.

Episode Transcription

K052225 - Disrupting Traditional Banking

Chris Wofford: [00:00:00] On today's episode of Cornell Keynotes, we are diving into AI powered FinTech with two pioneering voices. We are joined by Cornell University, professor Lutz Finger, who also acts as today's guest host, and he is joined by Colin Walsh, who is the visionary founder and board director of Varo Bank, which is the first FinTech company to secure a national banking charter in the United States.

The two discuss how AI is reshaping banking and democratizing financial services. Which creates new possibilities for inclusive banking. One big takeaway from today's pod is that success in AI driven FinTech. Isn't just about having cutting edge technology, but it's about masterful execution and actually demonstrating real leadership.

Much of what Lutz and Colin discuss connects with topics and lessons that are covered in the Designing and building AI solutions certificate program, which is brought to you in partnership by the Johnson School of Cornell University [00:01:00] and eCornell. So if this piques your interest, please check out the episode notes.

Now let's join our discussion on the future of banking with Lutz Finger and Colin Walsh. 

Lutz Finger: can you explain a little bit more about today's situation for the underserved population?

Because I think many people in this audience. They don't even know that part of this country in which we are living here, doesn't have access to banking. 

Colin Walsh: Yeah. So I would say that the US is largely banked. I mean, there's probably about 7% of the population that are not, that don't have some form of a bank account.

So compared to other parts of the world, I would say the US. population is largely banked and there's also

Lutz Finger: 7%, like I I think it's amazing, but it's 7%, right? These are still,

Colin Walsh: yeah, people that , don't actually formally operate inside the banking system. They might be using cash, they're using prepaid cards.

Other things that they, you know, they may be transferring money through money transfer apps, but for the large, but [00:02:00] 93% of people have some access to financial services. Also, the US is a very concentrated system from a supply perspective. You've got banks and financial institutions competing at the local level, the regional level, the national level, online digital.

But the problem is that unless you have a very low cost, very efficient, sophisticated platform to, help people, you tend to offer not very good services. Like I like to say that people are underserved and overcharged in the US and so, so as I mentioned earlier, many people who don't have a lot of wealth or income are experiencing an abundance of fees and charges.

Many of them are not terribly transparent. they have difficult time sort of building credit. If they're new to credit or they've damaged their credit, they have a very difficult time getting access to transparent, affordable credit. they don't necessarily have access to tools to help them build savings and greater financial resiliency.

And so really that was what Borrow was all about, is how do you bring [00:03:00] all of these things elements together in a digital platform to be able to help the, and this is the shocking statistic is that there's 180 million people in the United States that are just living paycheck to paycheck.

Like they're just trying to get by. And, and with all of the economic stress and economic uncertainty and political uncertainty that exists in this country, that is a real burden for many people because they're feeling this acute sort of financial stress and anxiety. So being able to alleviate that.

It also, and I know , Lutz, from your background, you know, you also have experience in healthcare and, and health tech and, you know, the financial pressures lead to health issues as well. And so Varo's mission is really around how to be the bank for all of us. How to be able to provide free banking services, quality banking services, but also being able to help put money back in our customer's 

pockets.

Colin Walsh: And, and we'll talk about how we do that, um, 

Lutz Finger: actually, And let's stay for the moment at the actual [00:04:00] problem at this situation because one of the things, so as I met you, I don't know, like, how many years is this? 

Colin Walsh: A long time ago. 

Lutz Finger: Long time ago, right? you had started the journey.

Um, one of the things which I found as a storyline very fascinating, which I think you should tell here, is if people, those hun over a hundred million people who will live paycheck to paycheck they get a paycheck. If they don't have a bank account, they go to the bank and let's just say the paycheck is a thousand dollars.

Then the bank charges them. A tremendous fee to actually, 

Colin Walsh: oh yeah, a minimum balance fees or if they're running short before the next paycheck arrives, they're paying punitive overdraft fees, you know, they're getting zero interest. so they're able to put some money aside for savings, if they are able to get access to a loan, it's probably at an extraordinarily high rate, and if they miss a payment, they're charged fees on those missed payments.

So, on average our customers, I mean, we save. Them hundreds of dollars of fees a year, which for people who [00:05:00] really are trying to make every dollar work for them, that's very meaningful. And then you add in the other sort of services that we're providing to help build much more control and resiliency.

Lutz Finger: Now, before you dig into Varo why is it that the banks. Before Varo came to the scene, weren't supplying. I mean, that's a, it's a huge market, right? Why didn't banks give that market, uh, a service or charge them so much that this was actually. Bad, right? I mean, 

Colin Walsh: well, , there's sort of, there's 2 2 reasons.

And the one I'll focus on is more the economic incentives. that it was really difficult for the banks to make money off of consumers that transact very frequently, but more low dollar transactions, so they're not necessarily high value transactions. they're perceived as higher credit risk.

Because the banks are using backward looking systems that are trying to look at credit performance as opposed to actually [00:06:00] monitoring real time cash flow and activities, with the banking system. And again, it, it requires much more sophisticated tooling and modeling capabilities to do that well.

And we'll talk about that today. and also vulnerable populations are more susceptible to fraud. Scams and other things. And so there's high cost of fraud and, dispute processing and other things. And so, so the economic incentives made it very difficult for, uh, and to this day make it difficult for incumbent institutions to effectively serve this large population of consumers.

And so this is where technology comes in now. There's other issues that play too, but that's, yeah. 

Lutz Finger: Now this is, this is so cool because whenever we talk About AI and technology, then the typical product viewer from a background, right. I'm a product manager, like I come out from the Google School.

Then it's like, okay, what's, what's the user flow? What's the user's problem? And we have that here. People are overcharged, they have [00:07:00] a little money, and that little money, a high percentage is actually going away for, for fees. And that happens because their risk is too high and the cost of doing business is too high.

So this is a perfect situation where technology can play a role, and this is the moment as you entered with a vision, and I listened to you and I was like, gee, this guy is amazing. So tell us a little bit, like, how did you pitch initially Varo Bank in terms of the technology?

Colin Walsh: Yeah. it first started with kind of what are the business problems that we need to solve as an institution, you know, so as a digital bank, thinking about fraud detection, thinking about, you know, how do you speed up money movement. Now, one of the issues with

making funds available quickly and moving money quickly is that it also creates vectors for fraud. So you have to have very strong tooling around that. because we're a regulated institution and we're moving money, we also have to look at things like Bank Secrecy Act and anti-money laundering and [00:08:00] terrorist financing, and making sure that none of those activities are happening and having systems to be able to detect anomalies and to understand how our customers are using the platform.

again, we've touched on a couple of times, lending. And being able to provide access to consumers who maybe the banks would not underwrite or feel are a good credit risk. So using advanced machine learning capabilities to be able to make loans available to folks. And then also thinking about personalization.

How do you get to know your customer and be able to serve up? Things that are relevant to their lives and making sure you're kind of there, meeting them where they are, and whether it's through how you acquire customers and making sure you're finding the right people, how you prove them, how you then manage those relationships over time.

And that also speaks to how you service these customers. Because one of the things we learned early on is we were providing voice support for everybody. But it was getting very expensive. And so then we moved more towards a [00:09:00] chat, sort of a live person chat, but then we started to build an AI chat bot that is now containing over 55% of our inbound inquiries, and it's allowed us to reduce our operational costs by about 60%.

But getting, but continuing to develop that. And, and as you, you mentioned. Lutz you know, kind of thinking about that journey. So first it starts with the use cases, but then it's really thinking about the platform itself and how do you think about your organization and the infrastructure investments you're making that are gonna help.

Make smarter decisions faster that are gonna try to identify causal links, you know, the accuracy of the decision making. also when as a regulated institution, it's very important to have explainability of your models and how you incorporate explainability into the development pipelines. and that also requires a lot of work thinking about instrumentation of the platform.

So if I, if I start trying to right. 

Lutz Finger: because we [00:10:00] covered a lot of ground, so let me just bring this together. You had two problems, right? The problem, number one, the operational cost for traditional banks were too high to serve. 

Colin Walsh: That's right.

Lutz Finger: Underserved. And the risk to do it was too high, which added to the cost.

Right. And I think you just talked through. So many good things that I wanna actually, 

Colin Walsh: yeah, we could do a class on each one of these. Exactly. 

Lutz Finger: Like a, like each of them. 

Colin Walsh: And I'll add a third actually is personalization too. 

Lutz Finger: Oh, personalization. Yes. 

Colin Walsh: So you don't have a terribly personalized experience.

So you have high cost, you know, what's perceived as high risk and then a lack of a real kind of, personal experience for the consumer. 

Lutz Finger: And all of this is actually a technology or machine learning play. 

Colin Walsh: Yeah. 

Lutz Finger: You are using ai. To manage risk, to look at fraud, to reduce the risk for lending. you talked about the personalization and the operational part, and all of [00:11:00] those are actually features in. Something which makes you, a very, very successful FinTech case, but it's not that there is one AI that solves it all.

You build an AI every time 

Colin Walsh: And, and I was just about to kind of geek out on the platform when you, uh, stop me because I can get, you know, get into it. You know. No, that's fine. But, um, like there's a lot in terms of. What you have to build.

you know, in terms of inferencing, inference and model and, sorry, the, the monitoring side, I mean, it's really important that as you build out these tools, and how you think about the model development pipelines and the feature platforms, and as you were just talking about, you know, there's many features that go into being able to actually provide these outcomes through the, the model development effort. And so, going back to your earlier question, so as we were approaching this, really thinking about how to design a platform from the onset that would meet some of these objectives [00:12:00] of better, faster decision making, explainability so that you can actually take your models to the regulators and have them understand them. I mean, these are all critical components that ultimately get us to the outcome of efficiency, lowering cost, making better credit decisions, having a more personalized experience for our consumers. 

Lutz Finger: So how, and this is like some of our audience, is doing fundraising.

They like Cornell has a very strong startup community, as you know, right. We just said there as a problem. The problem is nice described. we said there is technology, which can solve it, and you went on into the details as you said, you geeked up and kinda like you can say, oh, I can solve this. 

Colin Walsh: You're just getting me going a lot.

Lutz Finger: Yes, I know, I know. 

Colin Walsh: I got more, more, more to say, 

Lutz Finger: which, which is absolutely, like, which makes you an extremely good product leader and the leader for Varo in this case. How did you communicate this to, uh, you got strong backing as you initially [00:13:00] started. This was not a cheap journey to start with.

What was your. Pitch at that time.

Colin Walsh: I think the, the key pitch to our investors and, and we have real kind of blue chip private equity investors , and, and we have VCs, we have a whole kind of gamut of, of investors that it, it starts with, um, 

Lutz Finger: and you had me 

Colin Walsh: and, and you and, and you are one of our very early investors.

So, but it, it, it starts with, understanding how deep. Rooted. The problem is that, and, and the fact that there's a massive audience of consumers that could ultimately benefit by getting this right. And then you get to the conversation around, okay, well how are you gonna do this differently?

Like, what are gonna be the kind of unique, you know, sort of differentiators, whether it's in terms of the consumer facing proposition that you're putting into the market, or are there business model advantages? Like one of the real features of Varo is the fact that we are. A nationally chartered bank, and I think we're still the only FinTech in the United States to get a full [00:14:00] national consumer charter.

Um, and, and it's, 

Lutz Finger: can you explain this a little bit? Like, I know that this is important, but, I don't think the audience 

Colin Walsh: we're, we're veering off of AI a little bit, but your point around how did, how did I pitch? Varo was that I didn't want to just be operating with a sponsor bank, and we started that way because that was the only way to get to market quickly, do proof of concept, scale the business, really understand if it was gonna work.

But all along I said to my investors that we will be a much more sustainable business if we actually are operating with a bank charter, because that gives us direct access to the payment systems. We were a direct member of the FDIC. We don't have any of this, what I, I like to call, sort of stroke of the pen risk of working with a small sponsor bank that suddenly gets into trouble and they shut you down because you're affiliated with that sponsor bank and it may be something that you didn't even do.

It could be another program that's screwed up. And so really kind of controlling our own regulatory destiny. 

Lutz Finger: Um, I, I would like to put a bow on it because many companies currently [00:15:00] we see are putting just a little bit of lipstick. We, we call it wrapping. and you see this in large language models, right?

Like they see Chat GPT can do amazing things. They're get an API from Chat GPT, and then they put lipstick around it and kind of saying, look at how good this is. 

Colin Walsh: Yeah. And then they're really not, someone else is actually providing the, the value models. 

Lutz Finger: And we have this in FinTech a lot. We have a lot of so-called neo banks. That kind of take an existing bank layer. An existing bank provider, and then they just make a very, very nice interface into it and they, they can offer personalization, but what they cannot offer is the risk reduction and the cost reduction, and therefore they don't go into the actual vision that you have pointed out? 

Colin Walsh: Well, and I think this is what really makes Varo unique, is that we've created this vertical integration of our stack from front end customer facing all the way through operating as a fully functional [00:16:00] national bank, but able to do it. With very sophisticated technology, and this gets back to some of the early decisions around how we wanted to build the platform, how we wanted to think about using machine learning and AI and embed it into almost every aspect of our business.

I had a conversation a couple weeks ago with another digital bank, CEO, different part of the world. Someone from Asia and we were talking and, and say, look, well, do you get asked questions about ai? And the response was like, do I breathe? I mean, like, yeah. Of, I mean, like, it's not like we have to go sell it as something special.

It's just part of how we run our business. And, but that's all choices that get made at the early design stage in terms of how you. Build out your technology platform and you instrument it in a way that can allow you to get these benefits around cost reduction, around fraud risk management, around, being able to lend to consumers that traditional institutions won't lend to.

But it's all kind of starts at the, at the [00:17:00] design stage. 

Lutz Finger: Yeah. Now, . Let's walk through one of those use cases, because like, I, I think for all of those different areas, from personalization, over chatbots, over in, in better lending or cost reduction, there are so many things which we could actually talk, but let's be very specific.

How about we talk about risk? 

Colin Walsh: Let me maybe give a little more background and context.

And so, uh, so the customer that we serve is financially strapped, places an enormous amount of value around getting real time access to credit. It's, it's largely kind of liquidity solution. And so the product that I'm gonna showcase here today, and we've got a couple other credit products, but the one that is, is what we call the Varo advance.

And so it allows customers to get up to $500 and immediately real time, and it's available in the app, but it's based on very sophisticated tools that allow us to determine [00:18:00] who can access and how much they can access at any given point in time. But it really creates a critical lifeline for consumers, and it's a bridge between paychecks.

So if you're. You know, if you've had to reduce your hours because you're dealing with childcare issues or you have a bill, an unexpected bill that comes in and then suddenly you're, you just find yourself short before the next paycheck and it can be used for groceries and, and filling the tank with gas or whatever essential things you need to do.

People rely on this product. And so we've been able to, we're actually now on our third generation machine learning model. Based on a series of, it's using cashflow data, it's using the direct deposit data that we collect from our customers. It's looking at their spending habits, their account balance and it creates a series of features that allow us to use this sophisticated cashflow underwriting to determine how much credit we can give customers. And I can just say [00:19:00] just from the initial deployment of the model to the latest version that we're using, we've actually been able to double the number of consumers that can actually access the full $500 amount.

Um, and as I said, this is something that is incredibly meaningful for people who have just fallen short before that next paycheck. And the models are displaying very strong risk sloping properties. So this allows us to separate goods from bads, and being able to determine how to extend more credit to people that actually have the capacity and the ability to repay it. and then controlling how much credit available we make available to people that might have a problem paying that back. And it also fits very much. Within our mission of being able to provide access to customers for these sort of critical services. unlike traditional credit risk models that tend to use backward looking data, we're using reinforcement learning.

So there's a tool. Called XG Boost, which is an open source machine [00:20:00] learning algorithm. And it's been used in a lot of industries. I don't think as much in banking and financial services, that allows us to constantly monitor and learn what we can do and how much credit we can extend. The other thing that's also unique from Varo's perspective is that.

We're using certain constraints to be able to allow the model to be easily interpreted for regulatory purposes. And this is very important as a regulated institution. 'cause you do not wanna embed bias. You don't wanna have things that are too much of a black box that your regulator can't understand.

and so we're using tools and techniques to make it much more explainable. and then this is really creating what I like to call, and the team calls a really a bridge between traditional banking and much more sophisticated tech-based lending. because we're not relying just on that traditional credit history.

And I think the credit history can be incredibly punitive and, and prevent access to [00:21:00] quality credit solutions for a lot of customers that, that we serve today. 

Lutz Finger: Essentially you using in off the tool machine learning program. Um, In order to more accurately 

Colin Walsh: That's right. 

Lutz Finger: Calculate risk and what are now do I think 

Colin Walsh: consistently as well. There's, but there's another thing about this too. And, and it took several generations of the model to get this right because, you know, people have an affinity towards these, um, maybe cash checker, uh, check cashers.

In their local community, but they're charging them exorbitant fees, but they, they feel like, okay, well I know if I go there, I'm gonna get the money I need, I'm gonna get it when I need it. So we've had to sort of recreate that in a digital context to be able to help customers understand that if they continue to bank with us on a consistent basis, and, and we, and we use advanced personalization to also help people understand the steps they need to take to be able to access more credit.

And that's actually in the app for every [00:22:00] person, every customer. It sort of says, here are some of the behaviors that if you do this, it's gonna increase your likelihood to be able to get higher limits. And now we have a line of credit product on top of our advance that can actually help them access even more credit.

But to the question, you know, it is like how do you, build that trust in a digital environment? And it is around creating both the tooling to, to be able to get the reliability, the speed, the accuracy, the consistency, and serving it up in a delightful, easy to use and intuitive platform. 

Lutz Finger: And I think that like, here come two, two amazing technologies together, right?

Because you use a traditional, um, I, I call this now a traditional learning tool like X Boost, which is open source and you can deploy. You have the data, you can figure out a risk. Now you can offer a product like Advance. Which the. The cash, um, the check cashier, cannot do so easily.

But you can do it because you have the volume of [00:23:00] information and you actually can saying, okay, they, they are. Cashing the check always. So I know my tool tells me and I can do an advance. Now the other thing you can do, and this is more generative ai. You can start to replicate this trust level by offering in a nice interface and chat. And so on 

Colin Walsh: And, and there's also an element of building digital literacy as well. And this is something that I'm exploring more globally right now and in other parts of the world that maybe aren't as sophisticated from a digital perspective as the United States, but even in the US you have people who've moved to the us they're, or they just haven't had a lot of exposure to digital banking solutions. So how do you make it so intuitive to help make people feel safe and comfortable using these types of solutions? 

Lutz Finger: Totally. So So now you have your risk model.

Can you tell us a little bit, how do you use now this technology towards the regulator because they wanna obviously avoid [00:24:00] fraud. They obviously want to see something. And, just give us a little bit the idea of what, how do you avoid fraud using essentially they're base technology now. 

Colin Walsh: Yeah. So, well again, a lot of similar principles around, you know, anomaly detection, understanding, and the, and the, from a, I think I heard a couple questions there. One was around regulators, which I'll talk about in a minute, but also just from a fraud use case of that.

There's, there's several aspects to this. One is. Just even at the front door, like account decisioning and looking at, you know, identity information, being able to look at kind of where they're coming from, the type of IP device, like. There's a whole series of things and tooling that you can use right at the front door to make better account decisions around who you even let in because you want to try to keep the bad actors out.

And there's also, you know, there's bots now that, that are not even human. And so trying to detect, you know, whether somebody is actually human or are they just a machine that's trying to open up an account for [00:25:00] fraudulent purposes. And so we've gotten a lot more sophisticated.

In terms of how we use our tools to, to identify those bad actors at the front door. But then as I mentioned earlier, when you're dealing with more financially vulnerable populations, people are more susceptible to scams. And so if somebody sees on their Instagram, oh, you know, Varo gonna give you a thousand dollars.

All you have to do is click here and give us your credentials. People are like, oh, a thousand dollars. That sounds pretty good. You know? And so, so you find that you do get in instances of account compromise. But then again, going back to the customer. The customer's, one of the things they value extraordinarily is access to their money.

Like if they feel like they're gonna get blocked. So if you detect something that looks. Anomalous, like somebody's logging in from a different IP or a different device that maybe the account's not bound to or, you know, can you in real time sort of block those logins? Can you force a password reset, but still allow the customer to have access [00:26:00] to their, um, their funds?

Like, you know, there's been a series of tests and learning and over time, but you know, the fraudsters are also adopting many of these tools, so they're becoming quite sophisticated as well. So you always have to stay sort of two, three steps ahead of, to be able to make sure you can keep the platform safe.

You can continue to deliver an experience that these customers want. Because something I said earlier, you know, there's an abundance of supplies, so if you get it wrong, customers are gonna go somewhere else. And so, so, so again, using these sort of platform choices the way we've designed our system. Lots of experimentation.

I think that's the other thing is just being able to continuously experiment with the types of things that you can deploy and then have the ability to just keep building more sophisticated modeling capabilities over time. 

Lutz Finger: What I think is, it's so amazing because it is really one-on-one in the playbook for how to use AI in a business sense.

Just to wrap our conversation like that, just the flow [00:27:00] where, where we went, you identified a problem. There's an underserved community. It is because they're missing risk. Uh, they, they have higher perceived risk. You are using their data. You are better than the FICO score. You not rely like everybody relies on the FIO score, you said, no, we don't.

We use our data and therefore you could offer better services, more personalized services and that got appreciated and therefore. You again, get money 

Colin Walsh: and, and keep the platform safe. I mean, that, it's so foundational, particularly when you're dealing with banking services. I mean, you're dealing with people's money.

So, you know, I say this a lot, that the, the currency we trade in is trust and, and that customers have to be able to trust that the platform's gonna be safe, that their money is gonna be safe. Let me quickly go back to the, the regulatory point. So what regulators care most about are customer treatment and, you know, making sure you're not violating laws, you're not ripping people off, you're not doing things that are unethical and, you know, so really making sure that you're [00:28:00] compliant with the laws and regs.

Now, obviously, I founded this company, you know, with a mission and a belief that we can make the world a better place. So, I don't, I don't have a lot of concerns around ethical treatment, but, you know, you need to also adhere to the letter of the law, so, so ensuring that the customer treatment is right.

The other thing that regulators care a lot about is safety and soundness. And so making sure that you are not going to be exposed to large risks that you can't control, whether they're operational risks, whether they're fraud risk, cost of fraud, or it's, uh, credit risk. And so, so building tools that you can explain to regulators that say that we understand the risks that we have, this is how we're managing them, and how we're able to run an institution in a safe and sound manner is incredibly important.

And, and to your earlier point. there's a lot of companies out there that sort of delegate that to their bank because they don't want to take that on directly. They just want to design a nice interface and, and try to capture customers upfront. 

Lutz Finger: Very cool. [00:29:00] So now in.

Do you actually even use the FICO score? 

Colin Walsh: Uh, we do not on our advance product, but we do look at credit scores as sort of a one feature in the higher line of credit, but we use much more our cashflow modeling system to be able to extend credit even for the the higher loan amounts that we make.

But it's, so we really try, have tried to distance ourselves from some of these more traditional systems that, that again, can be very punitive and very exclusionary for people that don't have a credit history or maybe have, you know, damaged their credit and, and they lack context. And so by using all of this real time data that we have on our customers, it allows us to make better decisions and very effectively manage the credit risk.

Lutz Finger: And I, I think this is core to the understanding of how AI and data is changing every industry. And I mean, this is like the, the bigger topic which I'm trying to, to bring here to the audience [00:30:00] is. You guys don't even use a FICO score anymore anymore, right? Because you have built, a better way to detect a better way to serve. And therefore that knowledge has helped. 

Colin Walsh: But let, let me tell you it's not, and this is, uh, I think just words of advice for, for folks that are. You know, thinking about creating systems like this, it requires a lot of investment but also a lot of ability to learn. And so like one, so we built, you know, our first generation model, we started to collect data.

We started to extend loans, you know, it was performing well. But we felt it could perform better. And so then we introduced a second generation model, which was doing a much better job at discriminating the goods and the bads and, and the sort of risk sloping. But what was happening is it was very volatile from a customer perspective so that it would see that, you know, one of the things I think I mentioned was like, if your earnings were reduced for, you know, maybe you had two jobs and you had an hourly job, and so your [00:31:00] direct deposit came down for a period of time, it would immediately lower your advance limit.

And so, so what would happen is a customer, and this goes back to the sort of, okay, well I can rely on the payday lender 'cause I'm always gonna get this amount of money from them. You know, if one day you had a $500 advance, but the next day you had a hundred dollars advance, like customers would complain and say, well wait a second.

I was relying on that. I thought I was gonna have that $500. So we then introduced. The next generation model, which started to address the smoothing factors. And it started to reduce that volatility that our customers were experiencing while also allowing more discrimination of the modeling to be able to allow us to give more higher limits to more people.

And that has been hugely effective. So not only have we seen reduction in credit losses, but a huge reduction in complaints as well. so it, and going back to this idea of trust and reliability. Being able to provide that solution and having customers rely on it and know that it's gonna be there for them is incredibly important.

But that took [00:32:00] multiple iterations over the course of several years. 

Lutz Finger: So I have a lot of questions of people who think about your future growth, and I will pick one here. How do you capture now shares from company like, um, SoFi, other banks?

Other people ask for how do you capture shares for like some groups talks about Canadians. The Canadians need have as well, undeserved, like, how are you going there? How do you think about. You, you have now data, you have now a tool. Yeah. How do you think about this expansion? 

Colin Walsh: There's a, there's a few different ways we think about it.

One is just kind of go to market strategies and how we continue to reach audiences through a variety of different channels, and a lot of that is testing and learning and making sure that we're attracting the right audiences with the right messaging. and so we've just over, even over the course of this last year, experimented with a number of different channels.

Like we're now back on tv. We, again, we also impose a lot of discipline on kind of our [00:33:00] marketing and how, you know, we look at return on ad spend and making sure that channels are bringing in customers that have good economics, that are engaging in the right way. That's certainly one way.

The other is just continuing to innovate and offer new solutions. So like right now we're testing an advance solution that will allow customers to get the first advance without actually having to do the payroll direct deposit. So they can actually link an account and we can now underwrite using the tools that we've built off of data from another bank account.

And so we're trying to learn like, okay, what is the, what is the performance? How do they engage? but you know, early signals are very positive that customers, you know, they come to us, they open up an account and the idea if they really need that bridge financing quicker, you know, we're not gonna give them the full amount, but we'll give them something if they have the right risk profile using that third party data.

And like that's, you know, again, there's just a series of innovations and then thinking about what the next set of problems that our customers need us to solve and making [00:34:00] sure that those are in our pipeline from a product development, from a technology development standpoint. So, so it's, it's go to market, it's marketing activities, it's lifecycle management.

It's also thinking about the product development pipeline as well. 

Lutz Finger: So let's, let's shift a little bit gears and for that, I would like to use a question from, uh, Dr. Uh, Paramita. who talks about how does AI help to minimize financial risk in a more global context. If I want to be a little bit edgy and I'm European, then I'm saying, ah, Colin, this is all awesome.

So you generated a new score, but you are spying on people, aren't you? Because you look at behavioral thing and now you create a score. Is that actually fair? And you saying, well, I use that data to offer a new service. If you take a global perspective and tell, how does it tie it back to the bigger vision you started off.

Colin Walsh: Yeah. I mean, I think if you look at the broader, if you [00:35:00] zoom out on a global level, I mean, there's so many opportunities. Like there's a, a billion people on this planet that don't actually have what would be considered a verifiable id. You know, and so being able to use tools to help people start to be able to access, uh, you know, whether it's even just essential services or it's being able to access payment systems, opening up bank accounts, opening up digital wallets, being able to access micro loan financing could, could change the trajectory of, you know, a, a huge number of people on this planet. and being able to help people, one of the things that I'm actually working on with some of the big multilateral agencies, it's like, how to help displaced people. You know, we have. One in 70 people on this planet that have been forcibly displaced from their homes.

But one of the biggest issues they have is proving identity and then being able to access some of these services, whether it's payments or lending solutions or access to medical care and education and employment if they're in a [00:36:00] host country. And so, so there's a number of things that these technologies are gonna only help us bring people out of poverty, help reduce human suffering and hardship when people are forced to migrate. So there's just a number of applications that this technology can have that are for good, that can really help preserve the quality of life of so many people around the planet. 

Lutz Finger: And I, I I think this this brings your overall vision was I wanna bring banking services because you, you are like, I, I, I don't want to disclose your age, but like, you have been in the financial industry for a while.

Colin Walsh: For a while. 

Lutz Finger: Yeah. You serve a lot of people with money, with the banking, um, like tools, right? And your vision was, I wanna serve now the underserved population. 

Colin Walsh: It's about financial inclusion, helping people feel more included in the formal financial systems. Also helping to drive economic development and for individuals, for households and families, for [00:37:00] communities.

So when people have access to quality banking services, they have access to credit tools, they can start building savings habits and they can start having actually, even if it starts with small dollar saving savings, it helps people be more resilient when they have unexpected expenses. And so, so all of these things sort of ladder up towards, you know, how do we just create a fairer, more inclusive system?

And starting in the US and then thinking about how that can apply more globally. 

Lutz Finger: Yes, absolutely. But I actually think you bigger opportunity.

Lies in the regions of the world. They do have banks, but they have way more underserved people. 

Colin Walsh: A hundred percent, yes. And there's a ton of innovation happening across Africa right now, across Southeast Asia. When you think about the global south, that's probably where the biggest opportunities are from an inclusion perspective and how to use technologies to be able to help bring people into the formal system and access the things that they need to be able to get ahead.

Lutz Finger: [00:38:00] Now, we chatted a little bit in the prep about it. Can you say a few words about Grameen Bank? 

Colin Walsh: Sure. Well, I mean, Grameen was one of the pioneers in microfinance lending and they solved really early on that it made more sense to lend the women. So they, and the household, and they were gonna be more responsible credit risk, but they also pioneered a number of community based solutions to help people, uh, to help women in their communities sort of pool resources and be able to repay loans and, and start businesses.

And, and they've built a platform that has brought many people into a, a better place from a financial inclusion perspective and helped households and communities around the world. And so it's a great example of kind of the, the, it's the og microfinance lender. Now you have other companies out there.

Like I, a couple weeks ago I talked to the CEO of Talla, which is doing some great things in. Philippines and Africa and other parts of the world being able to provide microfinance solutions. there's lenders like Kiva and and Branch, and there's, [00:39:00] so there's a, a variety of them out there that are trying to provide these types of solutions on the back of what Grameen has done, and also using more advanced technologies, machine learning, AI type solutions.

Lutz Finger: So the whole point here is we are using data Grameen Bank did. You do, did. In order to calculate risk better 

Colin Walsh: and make better risk decisions every day. Exactly. 

Lutz Finger: Make better risk decision. And we are like, we are from the Johnson School, right.

We are in a business environment. And would we, in order to bring technology and a business together. If we lower the overall risk, then we can deploy investments better and make everything better off. 

Colin Walsh: and have better outcomes for consumers and, and businesses at the end of the day. But the other, we should just take a minute to talk about like, you know, we're at such early innings here too.

So like all the things that we've built and the tooling and the infrastructure, I think sets us up and, you know, kind of a culture of [00:40:00] experimentation and learning, but sets us up for this next generation. So now with. With generative ai, the large language models and importantly agentic ai. I think we're now entering into a whole new field of where, you know, we start to move towards more autonomous decision making on the parts of these models, which is, um, it's, it's exciting.

It's also a little frightening. But they're gonna be able to really drive that next level of transformation, and you move from a kind of reactive to more of a proactive AI where these tools are monitoring this information. They're serving up advice. They're being able to visualize data in a, in a much more robust, meaningful, personalized way.

And so you're gonna start to see, they can actually, if you give permission to some of these models, they can go do transactions on your behalf. So from a commerce perspective, you can say I wanna go to Hamburg. And then they'll say, okay, well what, here are the hotels and how much do you wanna spend?

And where do you want to eat? And they'll go execute it all for you. I mean, so, [00:41:00] so I think we're really just at a tipping point now where these tools, and now you saw obviously the announcement of Johnny Ives and, and Sam, although, you know, they're, they're thinking about this and like how to supercharge this next generation of models that are gonna be able to really change the way we live our lives. And that's certainly gonna have an impact on banking. Now, I feel like the early investments that we've made in terms of building a platform that can start to embrace some of this technology is gonna be a critical competitive advantage.

But I think we, there's so much we're still gonna learn in the in the kind of months and years ahead. 

Chris Wofford: Thank you for tuning into Cornell Keynotes. If you are interested in getting ahead with ai, be sure to check out the episode notes for details on the Designing and Building AI Solutions certificate program, which is brought to you by e Cornell.

So I want to thank you for listening, friends, and as always, please subscribe to stay in touch.

[00:42:00]