Welcome to the American Banker Podcast. I'm Penny Crosman. Adoption of advanced AI, including large language models, is happening in U.S. banks faster than many of us thought it would. Banks have given generative AI models to call-center agents, to software developers, to researchers, and to all their employees. Theo Lau, founder of Unconventional Ventures, has done a lot of research in this area, and she's published a book called "Banking on Artificial Intelligence, Navigating the Realities of AI in Financial Services." Welcome, Theo.
Theo Lau (00:36):
Thank you so much, Penny, for having me.
Penny Crossman (00:39):
Thanks for coming. So, as you researched this book, what were some of the most advanced examples of AI that you came across?
Theo Lau (00:47):
Ooh, that's a good question. Truth be told, AI has been employed and deployed in financial services for quite a while now. Most of the instances being used in the backend automating processes, in some cases navigating through fraud and trying to detect fraud in advance. So that has been used and if we want to talk about from a quote unquote, where more the advanced use cases are, I like to look to the East as I often say, where we find innovation, where we find new ideas, and this was, gosh, I would say back in 2017 or even maybe a little bit before that was when we had the likes of Alipay, where Alibaba was doing a lot of things, advancing financial inclusion in China as well as the rest of Asia. They were using a lot of data to help them extend microloans to consumers.
(01:50):
And the one thing I can think back that was fascinating to learn was their 3 1 0 model and that's how they use the data. They know the customer so well that they can tell people, "Hey, within three minutes when you apply for the loan, I would dispense the money to you within one second and zero human interaction." Now thinking back, this was back in 2016, 2017 when Superb was deployed, and fast forward to now 2025, we still haven't seen much of that similar advancement, if you will, in the U.S. or for the better part, the West.
Penny Crossman (02:37):
You think the U.S. is kind of behind overall, or was that just one interesting example?
Theo Lau (02:45):
That is an interesting question that requires a lot of context and nuance. I think our focus is a little different, and a lot of that is because of need. So we think back, looking at the two different ecosystems, at what we have in the U.S., which is more-established industry, versus in China and for the most part Southeast Asia, right? They didn't have the infrastructure, they did not have the legacy that we have. There were a lot of people that are unbanked. If you look at the number of people, the demographics that's there is massive. You're talking about a country with a billion people, you're talking about Southeast Asia with 40-something countries, all different religions, all different cultures. Their need is very different, and they had the advantage of being able to leapfrog precisely because they didn't have the legacy that we had.
(03:37):
So when they were looking at what technology is out there, what toolset is out there, that was what they were able to lean into, was at that time having a lot of data, having everyone on the platform, having the technology that they had and the need to find something to serve people who didn't have anything else. You have the small-business owners, a lot of micro entrepreneurs that need access to loans, that need ways to save money, that need ways to invest money. As a matter of fact, that was how Grab and Gojek started way back, back around the same time or a little bit after Alipay and WeChat Pay. They started as companies that had people that were driving similar to Uber. So they have drivers going around, they have delivery people going around, and then they realized, wait a minute, these people, these micro entrepreneurs, solo entrepreneurs, they didn't have a means to save the money that they earned. They didn't have ways to invest the money that they collected. And so those services started extending two different parts of a consumer life and that became the version of the super apps that they have.
Penny Crossman (04:49):
Is there any place you would like to see banks here deploy AI to kind of make consumers' lives easier the way you're describing quick loans or easier savings or features like that?
Theo Lau (05:07):
Absolutely. Because if you think about AI beyond the shiny new toy, beyond all the things that people get so jazzed about, fundamentally what it is really good at is being able to take a massive amount of data, find relationships between the dots and get insights out from it. And using that insight to help us get to things that we couldn't do before. Just like how we had, back then, AAA. I'm old enough to remember when we used to have to take a car journey, you had to go to the AAA office. You have to find out exactly where you're going, get the paper maps, draw out the routes, and be very careful and precise in where you're going from A to B. If you get lost, make a wrong turn, you pull over, try to figure out where you are and get back on the road.
(05:55):
Now we don't have that anymore because now we have GPS, we have all these software and apps that can detect where we are, that can help us get back on track to go from A to B. Similarly, I would love to get that in financial services, and I used this example quite extensively back, gosh, when you and I started talking almost a decade ago now, and I still use that example because I'm still looking for that solution. I am 52 this year. My kids are teenagers now. Oh, I feel old. In just a few years, they're going to be in college. My dad is turning 80 this year. And so if you think about where I am right now, I have my own business. In about 10 years, I will be 62. I really need to think about what I'm going to do in my retirement, secure my future.
(06:45):
At the same time, my children will be in college, two of them in the U.S. We don't need a math genius to figure out exactly how much that's going to cost. And at the same time, my parents will be late 80s, 90, which since I'm the only child, there is a financial caregiving perspective I need to think about. So as consumers, as sandwich generations, as adult children, we need to figure out how we can plan for our own finances (and) at the same time plan for our children's education (and) at the same time plan for our parents and loved ones. This is a multitude of financial relationships we need to think about and navigate through. And that is the perfect way for us to use technology, look at where we are, not just myself as a person, but all the households and everything that's connected to my financial life. And think of it holistically, how do I help this cluster of households plan their financial future so that I can get to the goals that I need? Much like when you're driving.
Penny Crossman (07:54):
Yeah, that makes sense. That you could use an AI engine that's learned from thousands of other people in similar circumstances and the things that they did and what worked and what didn't work, and kind of plug in your numbers, I guess, and get some suggestions.
Theo Lau (08:13):
Well, and also navigate dynamically, just like if you run into traffic, the software can recommend a different route similar to now, especially now when we see all the uncertainties going on with the economy, when we have uncertainties with our jobs, when we have uncertainties with what is the cost of living and what is the best place for me to live. What will be the best place for me to live and what can I afford to do? There's a lot of what-if scenarios and a lot of questions. As humans, we might not even know what is the right question to ask. I would love to get help with that based on, like you say, data from people around me, based on data from people similar to me.
Penny Crossman (08:57):
Yeah, that makes sense. Well, speaking of uncertainty about jobs, you have a chapter on the future of work. Are you seeing AI replace people yet in this industry?
Theo Lau (09:10):
It depends on how much we believe in the press release.
(09:14):
If we look at what Klarna had said, for example, last year, that they say AI has replaced so many of their people to the extent that they think they can gain so much more efficiency, not just from last year but in the years beyond, then yes, it might beg the question, well, maybe we won't need as many people. Or with the likes of Amazon. Same thing with coders, and I've heard from CPAs who say that they don't need as many analysts, entry-level analysts to crunch their numbers, to read through things that they had before because now you can use technology to create summaries so that humans can actually review the final output. So I don't doubt that there is efficiency being gained from technology and there will continue to be, but is it to the extent that people would like for us to believe or is that a front for organizations to restructure their work groups?
(10:20):
I do have a big question mark with that, but I think when we think about the future of work, when we think about the role of technology with respect to what we do, the question that we ask needs to be repositioned in a way. I think is not so much "So, is it replacing humans?" It should be more "So, how can we help upskill and reskill people whose jobs might get displaced?" Because there's no lack of data and studies and surveys that said people's skills will be basically not needed in the next five years, or 30-some percent or 40-some percent of people need to learn new skills because what they have is not applicable. I think one of the latest came from the World Economic Forum that said 23% of jobs will change in the next few years. So when we see numbers like that, the question is, OK, then what do we do and how do we help people navigate and change? And that's the big question that I have not seen answers for. So gaining efficiencies and making your bottom line look good, that's great, that's wonderful, that's important as a business, as a bank or any other industry, but what do you do with the people? That is the question, are we going to train them for new positions? Are we going to create something else for them? And if so, whose responsibility is it? Is it the government's? Is it NGOs'? Is it the companies who are displacing employees? Those are the important questions that we have not answered.
Penny Crossman (12:09):
I think we've seen a few banks make an effort in this area, like Synchrony has done quite a bit to try to help reskill people, and Bank of America has a whole academy that they offer, and here and there we see an effort being made. But I don't know that it's universal by any means. So banks have a lot of decisions to make about AI, about which models to choose, about how much to spend, about who to put in charge, how to deal with the risks of errors, hallucinations and bias. There's an almost unlimited set of decisions and concerns to think about. What do you think they should be prioritizing as they navigate this migration to more and more AI and generative AI in their business?
Theo Lau (13:06):
I think it depends on the type of institution they are, because that will govern the model that they're going to use. Are they going to develop much of this in house? Do they have the ability to develop that in house? Do they have the talents to do that, or do they need to lean more on third-party partners? And if so, what does that look like? So those are some of the fundamental questions they need to answer, and it's based on, practically speaking, what they have and what they have at their disposal. But beyond that, I think the biggest question I often ask people is, "If you say you need AI, do you know what you want to solve for?" When I was in an airport, I was just walking through the terminals and I heard a gentleman screaming into a conference call. He was on speaker and he said, "Get me some of that Gen AI!" And I stopped and I stared because it reminded me of the "Jerry Maguire" movie. And I was thinking, in my mind, I'm like, "OK, this is not like a commodity you go buy and just throw it in."
(14:28):
And it feels that way because it feels like you see a lot of press releases from financial institutions that said, "Oh, we are deploying a POC in X, Y, Z. We're doing this. We are doing that." A lot of proof of concepts, a lot of money and talent being thrown into pilots. But I don't see as much coming out from the other end that says, "Oh, we have successfully moved from POC to production," because it's hard. It's not an easy thing. So I don't think we're doing good service by just throwing all these things out. We need to be more thoughtful and thinking about what are the use cases we want to target. What are the real problems we want to solve for? Just in the very beginning when we started the conversation, you mentioned fraud. You mentioned customer service. You mentioned a lot of these use cases that banks are trying, and we need to think about are we trying to improve efficiency or are we trying to create new value? Because beyond the low-hanging fruit that we are solving for by putting a bot in the call center and helping employees go through the knowledge base and get information faster, there's much more we can do with the technology that can improve not just the bank operations, but also deliver new value for the members and the customers that they serve. So those are the questions they need to fundamentally ask themselves — what are you trying to do and why?
Penny Crossman (15:57):
Yeah, it seems like there's the sense that if we use generative AI, we're just going to save X hours of everybody's time per week or per day. And I still feel like, not to be a naysayer, but I still feel like when you use ChatGPT or Claude or a number of tools for something like summarizing a meeting, the results are still often pretty mixed.
Theo Lau (16:29):
They're not quite there yet. Right?
Penny Crossman (16:31):
Yeah. Do you think there's a danger in that kind of like "we're just going to become 20% more efficient by giving everybody this tool" and letting people lean on it? Do you think there could be some unintended consequences of that?
Theo Lau (16:48):
Oh, absolutely. And we have seen some of those. We have seen chatbots that hallucinated. In the very beginning, I think it was last year or two years ago, Air Canada had the chat bot incident where it invented a policy that didn't exist. Just recently, Cursor had the same thing. The chat bot said, "Oh, this is our policy, it's changed." As it turns out, three hours later, a human responded and said, "No, Sam (who is actually not Sam, it's a bot) was wrong." So there are definitely challenges with existing technology. And again, it goes back to being intentional, being thoughtful in what you're trying to solve for. Some things, it's OK if it's 80% correct and then you have humans to correct the rest, but some things you really cannot have it wrong. And we are in the industry where trust is important, where you need to make certain 100% of the time when you say something, you really mean that something, because you're dealing with people's money and money is emotional.
(17:49):
There are a lot of things that you need to be more careful and more thoughtful in what you do. And so what you tell customers is one of those. And I've seen credit unions and banks thinking through these problems and saying, "OK, we will deploy technology in the front in a way to understand what people are asking for. We bring the information back, but then we put many more guardrails in the back end before we produce an answer to present back to the customer," precisely just to work around the hallucination challenge. So there are ways to mitigate the risks, but we need to recognize and understand and acknowledge that there's risk. I think there was a report that was published recently that talked about when companies use generative AI, about 20 to 30% — I might be wrong, but I think it was less than 30% — of the instances where they actually check the output.
(18:51):
And that scares me because then the question becomes, well, what about the rest? Why are you not checking? Even when a human produces something, there's always a check and balance, there is always someone checking. So do we really trust the machines that much? And that is a huge question. There's a recent report from Edelman, talk about the
Penny Crossman (20:10):
For sure. So if you put your futurist hat on, where is this headed? What could the future of financial services look like when a lot of this is figured out and the use cases are figured out, the models are refined and the things that are now in pilot are in production, what are some of the things we might see that are different?
Theo Lau (20:42):
That's a very good question. I struggled with that a little bit in the last few months. The initial hypothesis that I had is that the larger organizations who have access to the tools, who have access to the talent and the capital and the data, all of that, they will be able to pull ahead much faster than the smaller organizations who need to figure out how they can innovate with less resources and how they can navigate through the whole partnership ecosystem. Because now they need to bring in different partners to build something. Similar to when we think about banking going digital, you see the smaller mid-tier organizations, the smaller community banks, a lot of them have more challenges in upgrading their system to provide the digital tools that consumers look for. So I do see that trend continuing. The question becomes how much of that gain that the bigger organizations will have will get offset by the legacy that they have, right?
(21:56):
Because when you are much larger, it just takes a little bit longer to get anything through. You might struggle more with silos in terms of data, in terms of systems, in terms of even departments of people. So there is a tug and pull, if you will. But I do see the BofAs and the JPMs and the larger organizations, the Capital One, who already have access to the technology. And think about it, using AI in ways such as trying to use it to detect fraud, building up different models and retuning the models and fine-tuning them and keep updating it and training it, that costs money. The cost of computing has gone down, but it's not zero. It needs a lot of data. It needs data scientists, and data scientists are hard to come by. So when you add all those factors together, I still think the larger organizations will have an upper-hand advantage, if you will, to be able to innovate more and find use cases that will help them up their game.
Penny Crossman (23:04):
Makes sense. Well, Theo, thanks so much for joining us and to all of you, thank you for listening to the American Banker Podcast. I produced this episode with audio production by Adnan Khan. Special thanks this week to Theo Lau at Unconventional Ventures. Rate us, review us and subscribe to our content at www.americanbanker.com/subscribe. For American Banker, I'm Penny Crosman, and thanks for listening.