Tom Pierce, Corpay’s Chief Enterprise Architect, explains how AI is being used in the payments space, its future potential, and pitfalls.
In this episode, Brennan Robison interviews Tom Pierce to discuss the impact of artificial intelligence (AI) on the financial services industry. Tom begins by sharing his extensive background in technology and his role as chief enterprise architect at Corpay, where he leads enterprise architecture and shared applications. The conversation then shifts to the public's understanding of AI, with Tom likening our current grasp of AI to the first inning of a baseball game. He explains that while AI, including generative AI tools like chatbots, is becoming more visible and integrated into daily life, many people are just beginning to realize its potential and long-term presence.
The discussion moves to the practical applications of AI in the financial services sector, particularly in fraud detection, customer screening, and onboarding. Tom highlights the significant role AI plays in identifying and preventing fraudulent activities, emphasizing the cat-and-mouse dynamic between defenders and perpetrators who also utilize AI. He notes the efficiencies AI brings to customer onboarding by automating data validation and error correction processes. The challenge of transparency in AI-driven decisions, such as credit approvals, is also addressed, with Tom suggesting that while AI can assist in these decisions, human judgment remains crucial to ensure accountability and understanding of AI's recommendations.
Lastly, the conversation touches on how Corpay is fostering an AI-friendly culture, emphasizing the importance of experimentation and cross-functional collaboration. Tom shares examples of AI applications at Corpay, including a chatbot named Carol used in Brazil to enhance customer service via WhatsApp, and AI tools integrated into developer environments to boost productivity. He stresses the importance of building an AI culture within companies to stay competitive and adapt to the rapidly evolving technological landscape, concluding with insights into the future role of AI in the payment space, particularly in improving customer service interactions.
Brennan Robison
This is Smarter Payments by Corpay… I’m your host, Brennan Robison, Director of Corporate Communications for Corpay. In this episode, Tom Pierce, Corpay’s Chief Enterprise Architect, explains how A-I is being used in the payments space… its future potential and pitfalls. Here’s our conversation.
Brennan Robison
Well, hey Tom, welcome to the show.
Tom Pierce
Hey, thank you, bruh.
Brennan Robison
So we've had you on to help unpack the intrigue and mystery around AI. But let's start a little with a little bit about you first. Give us a brief history of your career, how you came to Corpay and your specific role here.
Tom Pierce
Sure, I've been in technology for over 30 years now, and I've held various positions, mostly technical, certainly a lot of architectural positions. I came to Corpay via a company called Comdata that was acquired about 10 years ago. And since I've been at Corpay, I've had various IT leadership positions. And currently I am the chief architect for Corpay, where I lead enterprise architecture in a group called Shared Applications.
Brennan Robison
Very good. So, we'll jump into talking about AI and we'll start with broad strokes. I mean, everywhere you turn, there seems to be another application of AI. It's so ubiquitous that even pointing out how ubiquitous it is has itself become a bit cliché, but I'll do it anyway. I was listening to an ad on the radio and it boasted an AI product that is hallucination free and LLM agnostic. Now these are terms that wouldn't have made a lot of sense to most people a few years ago, but now they're becoming commonplace. So, my first question is, in terms of a public understanding of what AI is and what it can do, where would you say we are in this journey? Like in baseball terms, what inning are we in terms of getting a grasp around AI, its potential, both good and bad?
Tom Pierce
Definitely first inning. And I would say we're still filling out the pitcher. I don't know how far I can go with this analogy, but AI is something that people are experiencing largely for the first time in a very personal way, but it's been around them for a long time, whether they know it or not. So, the current crop of AI tools that are built around generative AI that are creating chat bots like ChatGPT, those tools are raising awareness with people, even more than tools like Siri or Amazon Echo or Google Home did with their ability to talk to devices. Now the devices are much more intelligent thanks to recent advances. So I think people are starting to realize that this is something that's gonna be with us for a while, that we're all going to experience over time. And I think that's reinforcing the interest that people are feeling in these technologies.
Brennan Robison
Now, a million dollar question in terms of business and the potential to make millions of dollars, especially in the payment space, where can AI truly provide value in the process of making and accepting financial payments quickly, accurately, safely?
Tom Pierce
Yeah, that's a fantastic question. So, to me, there's probably several places, but the three that always spring to mind for me are starting with fraud detection. So, fraud detection in the world of financial services is becoming increasingly important. There are much more attempts to defraud the financial services, companies, institutions. The growing fraud that we see, which would include hacking attempts from foreign groups or even sometimes state-sponsored groups to disrupt financial services industries, is very real and very much growing. There's a lot of law enforcement work to help stem the tide of fraud, but every financial institution needs to be doing much more around fraud detection. So, I expect AI will continue to be a dominant driver in fraud detection. We use it today. A lot of people use AI already, different AI techniques. But as these techniques get better, as the computing power increases, as our ability to store... a lot of data and exchange a lot of data improves, we'll see fraud detection get better and better. Another area I think we'll see, yeah, sure.
Brennan Robison
You know, actually before you move on, let me ask you a question to follow up on that. Well, presumably if we're looking at using AI on defense, the perpetrators of crime are obviously also using AI and investing in their own technology. Is it just kind of a cat and mouse game?
Tom Pierce
Yeah, definitely the bad guys are using AI. And I think one of the most powerful uses right now is in phishing campaigns. So phishing is where a bad actor will reach out to somebody and try to convince them to share information with them or get them to do something on behalf of the bad guy, right? And phishing traditionally has meant another human whatever communication mechanism, whether it's a phone call or email, interacting and trying to convince somebody to do something. Well, now with generative AI, it's much more convincing than it used to be to have a machine do the phishing instead of an actual human. And because generative AI is getting faster and faster in its responses and going back and forth, I think it's going to be very dangerous to have bad guys have the ability to use generative AI in these phishing attacks. I think that's definitely happening. And yes, it is a bit of a game of cat and mouse because it's going to get harder and harder to detect phishing. It's already getting harder. But as AI can generate more convincing content or more convincing conversations, it's going to get... pretty difficult.
Brennan Robison
So, beyond fraud protection, what are some positive efficiencies that can result from using AI in the payment space?
Tom Pierce
Customer screening and onboarding is going to be a big winner from AI technologies, from the use of AI technologies. Customer screening is very time consuming, sometimes labor intensive, because you are gathering data from multiple sources, and then a human looks at it and compares them and makes an assessment of whether we want to do business with a particular customer or not, or if there's something in the customer's background that is troublesome. So, if we can automate more of that, speed the process up, make it more accurate, computers don't get tired as an example. So, I think that that's gonna be really powerful. And then the more we can use computers and AI to do better data validation, better correction of data, going back and forth with customers to make sure that they correct any keying errors the customer had in an application, for example, that's going to be a big powerful thing for financial services.
Brennan Robison
So, when you're a lender, there are certain regulations around that. And if you decline to issue someone's credit, for example, there's a certain level of transparency that has to take place. If it's AI, will you always know the reason that someone was recommended not to receive credit?
Tom Pierce
That's a fantastic question. The AI technologies are a black box right now in terms of why they made a decision. It's sometimes difficult to figure out the precise reason that an AI model gave a particular answer. However, we will know an AI made the decision. So I think what will probably end up happening at least in the short term is that AIs will render more of an initial opinion will still leave the decision up to humans and will use the human's ability to have their own judgment to override the AI or to question why the AI came up with something to doubt that it was totally accurate. So I don't think we'll see like completely automated lending decisions or underwriting or credit approval or things like that. But I do think we'll see AIs assist humans in those decisions.
Brennan Robison
Certainly. So conceptually, there's always a lot of new ideas. Businesses are under pressure to use AI in some way or another. How important is it for them to have a real, tangible use case before going down that path?
Tom Pierce
In my opinion, it's very important. These technologies are new and they're exciting to play with. People are going to want to play with them. The truth is, though, they're also time consuming to figure out how to use, to figure out how to put into a process. They're also expensive over time. So, even a small amount of cost per transaction with one of these AI models may end up being very expensive when you scale it up to all your customers or even all your employees. So I think it's very important to understand the benefit you're trying to derive from this technology and not just blindly deploy it as a means to make magic happen. I think that's probably the worst thing you could do.
I think even in the case of employee productivity, when you're talking about enabling knowledge workers to have an AI chat bot or an AI search assistant available to them, I think it's got to be important to you to do that. Certainly, as these technologies mature and compute continues to get cheaper, it'll be commonplace, it won't even be a question of whether you would enable personal productivity with an AI. But certainly these are early days. It can be quite expensive when you scale these costs up to a large enterprise or out to your customers.
Brennan Robison
So, until there are more definitive AI use cases, how can companies kind of dip their toe in the water and then test it out in a way that's productive but not cost prohibitive?
Tom Pierce
Yeah, you said it. I think the key is testing, what I would call experimentation. I would encourage anybody, any company to just do a lot of experimentation and stand up processes for intaking ideas, dispositioning those ideas, understanding the business value, and then doing very targeted experimentation based on those ideas. Try to drive out in the experimentation if the solution can actually provide real business value. Because sometimes it can or sometimes it's not cost effective to provide that business value. So I think experimentation is the key. And if you have an innovation process, great. One of the things we're doing at Corpay is we have begun to stand up a dedicated process around AI instead of a generic innovation process. Because the teams that we want working on it are going to be more familiar with AI. Since it's a new technology, it's not easy to assume a group of people or employees would have experience with these technologies. So we're trying to wrap the right expertise around that innovation process. And I would imagine that that... is going to help other companies as well if they make sure that they're not trying to make it a general kind of process because of the new and rapidly changing nature of this technology.
Brennan Robison
And when you say dedicated instead of generic, does that mean you're building it from scratch or you piece and get together from existing technology? How does that work?
Tom Pierce
Well, the process itself, you don't have to build from scratch. I mean, it's basically an innovation process, right? The technology, certainly, most people don't have these technologies in-house or very much of them in-house, or they've been put in pockets, you know, where there are data scientists, for example. There's probably pockets within every company where there's some AI modeling going on. But I think what we're going to have to do now is... make these enterprise wide, introduce the entire enterprise to these technologies and think more about how the enterprise consumes these technologies, which should be a part of your innovation process. I wouldn't ever recommend people to assume that their innovation process and their experimentation should result in something that's just ready to go to production.
I'm a big fan of throwing away experiments and then re-implementing in production when you have those considerations. But what I'm trying to say is that AI is something that doesn't or shouldn't happen in a small box in your enterprise anymore. It's time to pull it out into broader use, into broader processes and that includes how you approach experimenting with it. It should be cross-functional teams from around the company, not just small pockets of people.
Brennan Robison
If you want the ability to share this technology and innovate collectively, what risk does that pose in terms of say, company secrets escaping?
Tom Pierce
Yes, definitely, definitely concern for everybody. A lot of the vendors have been quick to come back and say, hey, you know, we're not going to leak your secrets, or if you sign up for this level of service, we'll guarantee a sandbox where you can put your data and not worry about it getting leaked. But I think that's a very big concern. And, you know, the thing about these technologies is, they're not necessarily different from other technologies. What's happening right now is that it's not very cost effective to run them in your own data centers. Most companies can't make the business case to invest in the amount of computation, the amount of compute, resources that you have to buy to run these models. So, a lot of us are outsourcing them to other, you know, hosted solutions like cloud vendors, like AWS or Microsoft, or even some of the specialty solutions themselves, you know, you can interact with their API directly, you know, to do your work. So, I think that all promote, you know, provides a bigger risk factor than some other technologies that we typically use in the enterprise. So, to mitigate that, to mitigate the... the risks of leaking data or doing something insecure, your company secrets being out there for everybody. I think it's very important that companies revisit their data policies and strategy. They need to make sure that they're managing the really secret data in a very secret way. And if you're going to have proprietary data, proprietary information acted on by, you know, AI that's hosted in an outside environment, you need to apply a level of rigor to that. And there are emerging thoughts in the security world about how, you know, frameworks for how to evaluate these things. And we'll see more of that over time as well. We'll see a lot more guidance. We'll see even legal policies and laws regarding what companies can do with processing data, I believe.
Brennan Robison
And that leads me to my question about standards and regulations. Where do you see that evolving in terms of, you know, federal state, local regulations that will govern how we use AI?
Tom Pierce
There's definitely more regulation coming. So, governments are already putting out, you know, I would say directional policies. The US government certainly has done that. I think the biggest concern that I've seen out of the government, any government, is around fair use of AI, ethical use of AI, the impact of AI on the citizenry. And where that bleeds over into the corporate world is how we protect the rights of the individuals that we serve within the companies we serve. So I think yes, I think the legislation will definitely impact it. I think there's also going to be legislation that isn't as broad, that is probably focused on how certain industries process data with AI. So we've been talking about financial services. And I could see that there could be a time when there are questionable lending practices based on some AI decisions that were made that may get the attention of lawmakers, let's say, and that might limit how we can use AI to process those loans, like those loan applications, like we were talking about before. So... We have to be aware that there is going to be more legislation, more external forces, and even more compliance, even if it's not legislation, just more compliance from our customers, our vendors, people who want us to be very forthright with how we use AI and how we process their data.
Brennan Robison
So you've alluded to a couple of ways that Corpay is already using AI. What are a couple more examples that Corpay is currently using?
Tom Pierce
I can give you a couple of really specific examples that I like. We created a chat bot down in Brazil that we call Carol. And Carol interacts with customers in our Brazil market via WhatsApp. So it was important in Brazil to use WhatsApp because over 90% of the Brazilian population uses WhatsApp to communicate with one another. And it's... it's a growing or even large part of their communication fabric right now. So we wanted to take advantage of WhatsApp and the team created a chat bot using Microsoft technologies to help customers answer some questions about their account, payments that were processed, you know, those kind of things, and take some of the load off of the call center agents that we have, but also give people the ability to interact in the ways that they wanna interact. Because a lot of people, a growing amount of people I should say, don't wanna pick up the phone and call somebody. They'd rather just dash off a quick note. And it's been really successful. They've done really well with it. The way that they're measuring it is in terms of containment like you would call center metrics. So there are a lot of queries and interactions that are completely contained within Carol that never have to involve a human. And I think they're gonna continue to invest in that and expand it. So customer service is a really interesting area for all companies and particularly us in how we can better serve our customers.
Brennan Robison
Real quick, containment, meaning that you never had to get to a human. And of course that's good on the company side, but you're saying that the customers like it too, because they're getting satisfied or satisfaction with the interaction with this chat bot. Has the chat bot, is it learning over time? Is it getting better and better with each interaction?
Tom Pierce
No, we don't learn from those conversations and that's an intentional decision. We're not trying to train our own model. We're using a large language model to get technical. It's zero shot or one shot or few shot kind of queries to that model. So, we're not trying to train a model. We're not trying to learn from our customers data. In fact, we don't even store our customers data in the model itself. So, the way we make Carol better is by telling Carol we want her to react in certain ways or giving Carol more accurate data to work on. So that's how we improve Carol.
Brennan Robison
Got it, so you don't want it to get too smart and turn into HAL 9000.
Tom Pierce
No, no.
Brennan Robison
So what's another example of how AI is being used at Corpay?
Tom Pierce
Another example is employee productivity, particularly for our developers. A really common use case for generative AI solutions right now is developer productivity through add-ins to their editors and IDEs. So Microsoft GitHub Copilot is a very popular example of this that allows you to add AI directly into where you're editing code and get suggestions in real time about how to write the code, complete it. You can ask it to do things like document your code. You can ask it about errors in your code right there in the developer tool. So that is saving developers from having to do their own research in product documentation on the internet and trying to switch windows, use different sources. So really trying to make the developer more efficient in what they're doing. I think initially we've rolled it out to several teams, we're continuing to roll it out to more. And everywhere it goes in, there are a significant number of developers that do find a lot of benefit. We see improved velocity out of those developers, which is an agile term, which kind of means how fast you're moving, right?
Brennan Robison
How much code you're producing, how many deliverables you're rolling out the door.
Tom Pierce
Right. Yeah, that's right. It's a unit of measure of how fast the team's moving. Yeah. So that's all good news, right? And I think right now, it's early days for that kind of productivity as well. I think one, we'll see the productivity improvements for the developers get much better. Number two, we'll see that pattern used more. So Microsoft Copilot for Office 365 is an example of where they're gonna use the Copilot idea of having the computer along for the ride while you're editing something and making suggestions and correcting things for you. But I think we'll see that pattern in other products in other areas emerge more and more and get better and better. So I really think that idea of the computer and the human as a team has a lot of merit and a lot of runway left. And I think we'll see a lot of productivity improvements over time because of that.
Brennan Robison
So Microsoft has come a long way since Clippy.
Tom Pierce
Yes, this is way better than clipping. Way better than clipping.
Brennan Robison
So let's talk about company culture. How important is building a company culture that embraces AI in terms of training, exposure? It's not a magic bullet, but how is Corpay cultivating in AI culture?
Tom Pierce
Yes, so AI culture is important, again, because these technologies are very new, and we all need to learn more about them and figure out what to do with them. I think it's very important to do that in the corporate setting because AI is going to impact all of our roles in some way, and it's very important for us as individual employees to understand how they're going to impact our roles, and how we continue to contribute to the companies that we're a part of. So, learning those things is important and giving people the opportunity to learn those things, share ideas, share what they've been doing is all part of the culture that needs to merge around AI. Also this, like we've talked about before, this culture of experimentation. Again, it's early days. You know, there's a lot of ideas that need to be developed about how to use these things, you know. Something as easy as I automated something on my desktop in a very clever way, right? All the little things, but also, you know, the big things as well. So, I think having an AI culture is very important. And we've started trying to build an AI culture inside of Corpay. We're trying to bring groups of people together that are passionate about the technology. We've started looking at learning opportunities, you know, broad based learning opportunities for employees so that they can learn how AI can help them and how AI is evolving. And I think. You know, in some places we've even dedicated teams of people. We have dedicated full-time employees who are thinking about AI and how to use AI to provide much better business value out of our processes. You know, the thing about establishing a culture is it's a lot about people change. And people change takes time and it's very hard. So, I think we have to start now and we have to stay after it. You know, there's going to be people scared of AI technology. There's going to be, there's going to be people who think AI technology is over hyped and it can't help them. Right. So for every passionate person who's embracing it, I don't wanna overstate it, but there may be somebody that's sitting in the wings, pushing it away and missing out on opportunities. So I think we have to start working now.
Brennan Robison
What is the phrase that I heard? Uh, not every company will use AI in the futures, but the, but those who do will be beating those who don't. Am I getting that right?
Tom Pierce
Yeah, I'm yes, I would I would back that if that's how it was said, I would back. I mean, the principles, right, I think it's another tool in the toolbox, right? It is a very powerful tool. And if you choose not to use it, you are missing out on a tool in your toolbox. So, you know, if you go after the nails with a wrench, because you didn't get the hammer. You're missing out.
Brennan Robison
Missing out indeed. All right, Tom Pierce, Corpay Chief Enterprise Architect. Thanks, Tom.
Tom Pierce
Thank you.
Brennan Robison
That’s it for this episode of Smarter Payments. Thank you for listening. Be sure to follow the show wherever you get your podcasts so you don’t miss an episode. Smart Payments is a production of Corpay, Incorporated. Copyright 2024. I’m Brennan Robison. Until next time.