For decades, digital banking has been built around clicks, menus, and workflows. But AI is changing that, creating banking experiences built around conversation, context, and intent.
In this episode of Banking on Community, we sit down with Chris Cox, General Manager of Digital Engagement Solutions at CSI, and Daniel Haisley, CSI’s Chief Data & AI Officer, to discuss how conversational banking is reshaping the digital experience for consumers and businesses. We explore how AI is helping banks move from reactive service to proactive guidance, deliver more personalized experiences, and strengthen customer relationships.
We also discuss the practical realities of AI adoption, including data readiness, governance, risk management, and the rise of AI agents. If your institution is thinking about what’s next in digital banking, this conversation is a good place to start.
Transcript
Saxon Prater (SP): Welcome to “Banking on Community,” the podcast for community bankers. I’m Saxon Prater, and today we’re talking about how digital banking is being transformed in the era of AI. And to do that, I’ve got two digital banking experts, Chris Cox, the General Manager of our Digital Engagement Solutions here at CSI, and Daniel Haisley, our Chief Data and AI Officer. Did I get those titles right?
Chris Cox (CC): You got it. They just roll off the tongue, don’t they?
SP: Yeah. For those of you at home, I struggled with that just a little bit in the beginning, but hey, here we are.
CC: Thank you, Saxon. Glad to be here. I was thinking, this is the first time since Apiture became part of the CSI family that we’ve done this podcast, so I admired it from afar before. And we’re just really excited to be part of the CSI family, so thank you for hosting this.
SP: Fantastic. Well, welcome. And I’m really excited to talk to you both because we’ve talked on this podcast a lot about AI, and the industry as a whole has been talking about AI ad nauseam, right? I’m a little less concerned with AI just as a general concept. I’m more interested in how that’s affecting how account holders interact with their banks. And oftentimes, the way that they’re interacting with their banks is through digital banking. So maybe let’s start there. Where in the modern digital banking space, let’s say, is there still friction? Are things too complicated, or where are there opportunities for AI to help streamline the experience for people?
Daniel Haisley (DH): If we think of what the experience has been for community banks and their customers over the last, call it 30 years, digital banking, by and large, has stayed roughly the same. And by the way, that’s an indictment on me having been in this space for a very long time and trying to push forward and trying to make meaningful changes. You could argue that the last big innovation in this space is the ability to take a picture of a check to deposit it. Community banking itself hasn’t really evolved. You still trade at a premium on trust and on relationships. So finding that, how do you take those things that community banks do really well and be able to integrate that more effectively into that channel, which is now overwhelmingly important. The digital experience is where you have the overwhelming majority of interactions with the financial institution.
It’s fun to see over the last 18 months, call it two years, as AI, and really for the purposes of this discussion, we’ll say AI, most people are going to be thinking about large language models, ChatGPT- Of course and Claude and so forth, that it is genuinely a moment of inflection, where the things that we’ve wanted to do as an industry, the things that allowed community banks to differentiate themselves.
So they differentiated themselves by location, and we still have some of that. But also, having that sort of a relationship where you know the customers uniquely, you know what’s happening with them, to be able to give sage advice, to be able to do the things that help customers maximize their financial potential. Now it’s awesome in the community banking space that those tools are available through the form of AI, kind of AI-influenced things, to be able to now be proactive instead of reactive. Right. To provide the kind of financial literacy, to provide the same breadth of tools that some of the bigger institutions have, that just haven’t been available to community banks up until this time, is a really interesting time to be in the space.
CC: Daniel and I, together, three or four years ago, we did a conference presentation where we talked about the sort of vision for what we were calling at the time conversational banking. And that it started with the idea that our industry has spent 30 years building technology to replace what used to be only an in-branch, in-person interaction, to make it possible over digital channels. A whole bunch of fundamentally important and really difficult technical evolution that’s happened. But then you get to sort of what we think about as the traditional digital banking interface, which is the business that we’re in. And we’re proud of the user experience that we offer, but it’s still very sort of rigid and form-guided, and people have to navigate it people have to navigate. The entire bank is only 13 clicks away. So we started thinking about first, what if you could replace clicking around to do things in digital banking with something that’s just more prompt-based. So just tell me what you want to do. But then that’s just the beginning. Then that evolves into how can you use technology so that as a financial institution, you can build really meaningful and deep relationships with your customers through technology like you used to do when you saw them every week in the branch. And I think at the time when we were sort of vision casting, none of the AI-centric technology that exists today was around. There weren’t MCP servers, and so there was sort of this notion we knew ChatGPT was coming. But now fast-forward to really only two or three years later, and the vision is now very much a reality, and I think probably very much a mandate in terms of how we evolve technology to support the customers of our clients.
DH: It’s interesting. At that point, Bank of America was a number of years into their Erica product, which was their kind of chatbot that is on steroids. And still, the biggest thing they could do was store hours, and what’s the routing number? And that’s Bank of America. To see the progress that’s been made in the last 24 months is astounding.
SP: Yeah. One thing that’s really exciting to me is, obviously community banks, they’re built on relationships. And for the longest time, part of the discussion was, how do you maintain those relationships in a digital-first world? And there were plenty of products that were rolled out, like video conferencing-type solutions. But the cool thing here is exactly what you were saying, where it’s like, gone are the days of navigation. Now it’s maybe somebody just types in, “What is my forecasted budget for next month?” Or something like that that’s actually useful.
DH: I think what’s really interesting, there was a stat, and I think it’s from 2022, where roughly 44% of Americans would not qualify as being financially literate. And if you’re a community bank, that is devastating. It should be for all of us. The notion that you are relying on customers then to know the job to be done, and you’re just optimizing your experience to be able to minimize the number of clicks that are required for them to go and do the thing, all of that is, it’s backwards. What’s fantastic, the ace in the sleeve for community banks is they’re the nexus for some of the most, we’ll say sensitive, but we’ll say important data that consumers and small business owners have.
It’s where do they spend their money. So being able to see transactions, being able to see accounts, being able to see when accounts get opened, account origination data and loan origination data, analytics data. When people are clicking through, what are they clicking? If they get stuck someplace, what happened, why? Banks are in the kind of nexus of all of this data. So if they’re not walking into the branch and having the conversations that they would have historically of, “Hey, daughter’s about to graduate high school. She’s going to be going off to school.” The signals from that still show up in the data. So now the financial institutions that we’re seeing be most successful have a couple of things. They are clear on what they want to be. Who do they want to be the kind of optimize for or provide the most value for? So they’re really good at the thing that they do, and they’re really good at understanding, reading into the data, and doing something about it. So even if I’m not having the conversation across from the teller line or the desk that we used to be able to, I still know as much or more than I did, and now I have the tools to be able to proactively go and engage with customers in a conversational way.
SP: Right. So even what I was saying before, where somebody is manually entering in, “Help me with my budget, help me with” whatever it is that they are asking about. Sounds to me like that stage of them asking for help, you could potentially sidestep that altogether and be more proactive.
DH: You can preempt it, for sure.
DH: So we’re in a good fortune. We have millions of customers that are using our solutions today. You’re solving for an entire distribution. So folks are going to be hyper-sophisticated, and they are going to be financially savvy, and they are going to have really complex needs all the way to folks who are just getting started. One of the things that’s fantastic about the kind of models and AI and from a single interface, how it’s able to flex, is you can aptly support that entire distribution really effectively.
CC: I think what’s really exciting about this next revolution of technology, and in my lifetime, I’m a little older than Daniel, but we went through- a lot older. We went through fundamental transformations, the internet, and then cloud computing, and now AI. So AI has a potential to really democratize what’s possible with technology problem solving for even the smallest financial institution.
So if you think about, if you go all the way back to the beginning of banking, right? Banks exist because of trust. So I trust you to manage the most valuable things to me and convenience, and I can access those things when I need to. And over the years, we’ve applied different layers of technology onto that to make enhanced trustworthiness and convenience. But really, if you think about also the sort of community banking especially has become really commoditized, right? Most people have a DDA account.
Their paycheck goes into it. The money goes out of it in a number of different ways, and it’s just sort of a transactional utility. What if we can use this technology to make the financial institution, the community financial institution, the center of the customer’s life again? And I think that comes back to answering a really fundamental question. So customers of banks don’t care about necessarily the features in a mobile app or the payment rails they can use. Really what they care about is, “Help me maximize my wealth.” If you as a financial institution can help me maximize my wealth, which might mean higher interest rate investments or lower cost loans, or a number of different things, then I want to build that relationship with you.
So this technology now allows the community financial institution to be sort of the center of an ecosystem of different financial providers that service the customer’s entire financial life. And that’s a bunch of Sometimes I feel like these are like conceptual things. But so what does that really mean in practice? And we’ve got Daniel and his team have got some really good ideas around like, what does that really mean for a financial institution to become the center of its customer’s life?
DH: And we think a lot, and by default, we talk a lot about the consumers and consumer experience because we can all empathize with it. We’re all individual consumers. But businesses, so for community banks, which are getting pressure from every direction- the ability to support businesses effectively and be able to support larger and larger businesses is increasingly important for their profitability, for their ability to have the privilege to continue to serve their communities.
And so when a small business owner or, I mean a mid-size business, when the thing that is most important for them is just to know what is their real-time cash position, that we’re in a much better place to truly be able to answer that now. And then subsequent to that is the appropriate deployment of capital, which is ultimately a really important function that banks provide into the communities and nationally and down to the community level. And community banks know best how to deploy capital securely in a way to grow their communities. So using AI to be able to help inform that so that we’re optimizing how capital’s deployed, how we’re growing these communities, is really important. I think that we’re in an influential time and space to be able to do that.
SP: That’s fantastic. So yeah, so I hadn’t even thought about it from the business side of things. I was thinking from primarily the retail consumer, the account holder, but yeah, that makes absolute sense. These tools could be applied in both directions to make sure that these institutions are central. What capabilities or where do you think Two questions. How are banks doing in this? We’ve talked at a very high level. We’re excited about the things that are coming, but how is it actually going on the front lines, and what capabilities should they prioritize?
DH: I think that we have a responsibility as a service provider to give them the tools to succeed. And I think that there’s, and this is not just specific to, I love CSI, it’s not just specific to CSI, but as an industry, how well we’re able to serve businesses and serve consumers. I think we’re really well-served to have community banks that are growing and thriving, and I think that there’s opportunity for improvement on all of those fronts. I said the banks that I’ve seen be most successful are those banks who can articulate a reason why someone should bank with them. And that may be that they’ve focused on a vertical, that they’ve focused on a specific segment, that they have figured out that they’re really good at the university system. They’re really good with whatever. Pick your particular industry, your particular niche, and they lean heavily into that. So having the tools to be able to continue to lean, things like integrating faster. So I want to be able to connect my ERP systems. Wildly important to be able to do. I want to be able to connect wealth systems, be able to see my data. Wildly important to be able to do. I need to be able to move money, be it small sums, be it large sums, so have the different rails available. They don’t care about the rails, but they care about the ability, to Chris’s point, to be able to move that money becomes really important, so we can arm them, arm the banks to be able to facilitate these needs.
CC: It’s interesting to think about. We’re really only two and a half years into this sort of AI revolution, and it’s already fundamentally changing the ways that all companies, including financial institutions, do business. And Daniel makes a really good point in terms of helping financial institutions understand how they can use the technology to grow their businesses, to solve the specific business problems they have. In some ways, it’s our responsibility to help show the way, and we’re all kind of learning together. In 2024, I did a conference presentation, and I thought I was so clever. It was right when ChatGPT kind of hit the mainstream, and I used ChatGPT to create the slides for my presentation about what’s coming with AI. At the end of the presentation, it’s like, “Oh, and the big reveal is I used ChatGPT to create these slides.” I thought I was so smart.
Things like in just- in just two years, we are so far beyond that. And some of the mindset shift is if you’re still, as a banker, if you’re still thinking about AI as just an enhanced Google search, you’re kind of entirely missing the point. So the mindset shift is you’re not using these tools to help you find information. You’re using these tools to help you build things that help you run your business. And you don’t anymore have to have a deep technology background to be able to build things using these tools to help you run your business. You have to have some creativity, and you have to have some willingness to experiment with the tools. And I always tell people, “The first thing you should do is just go try something. Get the $9.99 Claude subscription and just see what you can do.” And you’ll be amazed. It’s pretty amazing what’s coming.
SP: Yeah. You can knock that presentation out on any given Wednesday.
CC: That’s right, exactly.
SP: Wednesday morning.
CC: That’s right.
SP: Very cool. Yeah. One thing that I hear is just a common concern is questions of security and compliance. So what are some considerations there that banks should think about if they’re not already? Which I know we’re preaching to the choir here. You probably are already thinking about all of these.
DH: Yeah. I think the entire world is trying to figure out what are the ramifications, what are the risks that exist, what are the risks that we can absorb, what are the risks that we need to manage away and avoid altogether. It’s not going to come as a surprise, but regulators aren’t exactly on the front lines of this yet. It’s not been codified. But there are guidelines that are out there that we can lean on and do lean on. Some of this goes back to, as financial institutions, understanding and having, not even just financial institutions, but everyone, having a culture that’s willing to try things and learn, and if it’s not successful that first time, that’s okay because you still learn from it. You get 3, 5, 10% better, and you pick up and try again. What AI has helped is to turn that feedback loop from something that would take months to happen to something that you can do in hours, hours or days.
That just wasn’t available before. But pursuant to that, you have to have a culture that’s willing to say, “Not everything is going to be a home run. We’re going to try, we’re going to learn, we’re going to do a little bit.” But having a risk posture which is well established, “I’m willing to take these risks. I’m not willing to take those risks,” and not compromising on that. I think it’s just mission-critical.
CC: Yeah, and if you are going to roll out AI tools across your enterprise, which you should be, by the way, there are a couple of things that are just top of mind, right? So you need to have people at your company who understand how customer data and proprietary company data is being protected. And there are ways, with enterprise models and things, with these tools, that you can make sure that you’re protecting those things. And then, questions about how do I know what my employees are doing with these tools? And so back to Daniel’s point about understanding your risk posture and putting risk policies around how you do and how you don’t want employees to use these tools is step one.
There’s a question, I think, about, so as a financial institution, your service providers, companies like CSI, are for sure going to be offering you solutions that are either exposing LLMs or AIs to you directly or using AI in the background, and you’re going to want to understand how those tools work and how they’re using your data and your customer’s data, which the vendor should be able to explain to you very easily. I think there’s a question about how much automation am I going to allow, and if I’m going to use these tools to let my customers do banking things or to let my team do things to protect my customers, how much do I need to be involved versus how much do the tools just do themselves? I think you think about a few things.
So one, for a given transaction or scenario, what happens if something goes wrong? And if the something that goes wrong is catastrophic or damaging to the business, that shouldn’t be automated. The second would be, can whatever was done be reversed? I mean, that’s some input into when you think about what should be automated. So, if a wire gets sent automatically in error by some AI tool, that can’t be reversed, so maybe you want to think about not doing that. So those are a couple of things I think you want to think about when you define your risk posture.
DH: Yeah. I think what’s interesting, you kind of have to skate to where the puck is going as well. And this is increasing the pace of play, meaning the models for what was available in January, February of this year versus what’s available now are night and day. So when people first got into ChatGPT and it was early 2025 or late 2024, and they said, “Ah, it hallucinates. It comes up with things.” Hallucinations still happen, but it is wildly better than it was then. So if we think if we continue forward with that kind of path of progression, so today there are discussions around, “Well, I would never let AI make a credit decision. I would never let AI handle deployments. I would never let AI” All of those things are going to happen. It’s a function of time.
So today, you lean very heavily on, to Chris’s point, humans in the loop. So for those really important decisions, you need a human there. You need to be able to make sure and to validate, if for no other reason, if something were to go wrong, you need to be able to stand up in front of your customer, in front of the regulator, in front of whomever, and provide a clear and cogent answer on what are the controls, how could this have been avoided?You can’t just hand over the keys. But know that this is only going to continue to increase. So if I’m running a community bank right now, I’m not thinking about what is the state of the state today. I’m thinking of what am I doing 18 months from now when people can just go to Claude and say: “Optimize my cash for me.” Why are they going to keep their cash at my financial institution? And I’m working feverishly on solving that problem- right now. Right.
SP: Right.
CC: Hey, Daniel, can I ask Daniel a question, go for it Saxon?
SP: Go for it. I like it. Turning the tables.
CC: The idea of if and when AI is used to make a credit decision, as an example. So the regulators and the bank, at some level, are going to need to explain how that decision was made. So talk about from a technology perspective, where do you insert the parameters so that you can explain a credit decision, as an example?
DH: So if you’re going to do something like extensions of credit, the regulations around that are going to be very different than a lot of the other applications that we’ve talked about. When we think of AI, there’s deterministic flows and there’s non-deterministic. So if it’s a, “I’m going to extend credit/I’m not going to extend credit,” I think you’re going to lean very heavily on the types of AI that have existed in the past, things like machine learning. It can’t be a black box that, “Well, I don’t know exactly how it came out with the decision, but the decisions are usually pretty good.” That’s just not going to fly. So you’re going to need to hold to account very clearly, very articulately how and why it got to this decision, and drive towards a deterministic model as much as possible. You’re just not going to be able to exist with, “Huh, well, that’s funny. It’s not what I thought what it would’ve said, but yeah, what are you going to do?”
CC: At some level, you could manage that with well-defined prompts.
DH: Yes. For sure. So there’s a lot that you can do with pre-processing and prompts. There’s a lot you can do with handling your own kind of embeddings and tuning models in order to make sure that it has access to the right information. It weighs that information the way that you want to, but there is still a It needs to be very deterministic. If X plus Y is greater than Z, then A. For things like credit extension, you have to get to that point.
It doesn’t mean that you can’t use LLMs around it and it still be conversational, all of the things. Yes, absolutely. But there are particular hot buttons out there that you just have to be crystal clear on.
SP: One term that we have evaded, or I don’t think anybody has said over this entire conversation, is agentic or agents. So let’s talk about that really quickly. Where does that fit into the conversation? And maybe for the audience also, let’s define what it actually means.
DH: We’ll use the software development life cycle as an example, and we’ll walk through how, I think, software development life cycle, which is just how we go about building products. Every company that builds products has what’s called an SDLC, software development life cycle. And today, that has these, whatever, 20 flows. And you have a design and a product organization that says, “This is what I think we should build.” And they go out and test it, and they design it, and then you work with engineering. Engineering goes out and builds it, quality tests it, and then commercialize it at the end, and you roll it out. All of that exists today.
The next phase of that is AI-assisted development. So when I’m coming up with Not me, but when the smart people come up with designs and the designers use designs, they’re using things like Figma Make to help improve the speed with which they were able to do that. The engineers are using things like GitHub Copilot. The product folks are using things like Claude or Microsoft Copilot to be able to help with the stories themselves. That’s AI assisted.
The next phase of that is creating skills. So we talked about pre-processing. It’s the things that you can put into a prompt beforehand to explain, just answer this as though you are the best product manager in the world, and you know da, da, da. You give it all this context of the things that would help inform if it were the best product manager in the world, the things that the data it would need to know in order to inform the best decision. Those become skills. As you’re able to then have those run themselves, they become agents.
So you’re going to see software development life cycles, and are seeing, that move from that kind of phase two into a phase three, where the agents are kind of chaining together. So it doesn’t mean that you don’t have humans in the loop. You still do. They’re still kind of managing checkpoints, but the foundational work of writing the code or writing the stories or writing the acceptance criteria or writing the implementation notes or the things that come out of it, instead are being able to evolve those at light speed.
CC: By the way, Saxon, you may know this, but I’m not a technology person per se, but I was- as I’ve been watching our team do what they do, so I hear prompt, right? And what I think of is just the little chat bar where you type in. What is agentic AI? When they talk about prompt, it’s similar, but they’re talking about what can be really pages and pages of setting context around what you want an AI agent to do and not do.
So remember, we’re moving from using AI to ask for information to using AI to do things. So a prompt can be And it’s written in English, right? That’s the beauty of it. You don’t have to be a coder to write a prompt, but it can be very detailed structured, “This is what I want this agent to do.” So it’s more than just asking something in a LLM chatbot. Right? Did I get that right?
DH: Yeah, no, that’s right. It’s been really fascinating because I think you used the word democratize earlier, that the roles of the true technologist, the roles of the person who knows the business really well, the role of the person who knows consumer psychology and design interaction methodology, those are kind of munging together now so that you don’t have to be an engineer for 15 years in order to produce an application. If you do have that architectural background, you do have that engineering background, it certainly helps you. There is value to that, no question, but the amount of information that’s available, I said before, we’re kind of moving into a world of jeopardy where the value isn’t in knowing the answer, the value is in knowing what questions to ask.
Because all of the data is available. So when we talk about adding to context, meaning providing more data, more information to the particular agent, to the particular model, to help inform how it can answer, that’s kind of how that plays. But it’s very different. The world has changed meaningfully in the last two years, and I think the next two years are going to look wildly different than they do today.
SP: All right. Well, you guys have given us a lot to think about. In closing, what should financial institutions be doing today to prepare?
CC: That’s a great question. So I think the first thing that FIs need to do, and this is relevant to AI as technology or really any technology that’s out there, is clearly answer the question, what problem are you trying to solve? Because with AI, like every sort of technology innovation, you’re going to get into shiny object chasing, and, “I should be doing something with AI.” That’s backwards thinking. What problem are you trying to solve? How could I potentially use AI to solve that problem?
And I think right now in our industry, you’re seeing all kinds of press releases from companies launching AI frameworks and announcing things related to AI, which is probably pretty confusing to financial institutions if they don’t have the context around, well, what is the problem that I’m trying to solve, and how is what they’re describing in a press release relevant to helping me solve that problem? So that’s thing number one. Thing number two, I think would be getting yourself, if you’re an executive of a financial institution, and your team in the mindset of we need to embrace these tools. A change is coming that’s going to help us, and we need to figure out how to take advantage of that change.
DH: I think being comfortable with change generally, which I’ve spent one of the decade working in the community banking space, working for community banks, and I will say change generally wasn’t the thing that we were great at. So getting together as a leadership group and being an open champion for these types of change. There are things that are going to be You’re going to have a groundswell of people wanting access to these tools, and the tools are super cool.
But Chris’s point, staying laser-focused on what problems do we want to go and solve, not falling in love with the tools, but falling in love with the problems to be solved. And then I would encourage, particularly for the senior management teams, executive management teams of community banks, be hands-on. Be champions. Visible leadership matters a lot here. And as things change and evolve, find people that you trust. Everyone is kind of figuring this out. The world is figuring a lot of this stuff out as you go along. Be engaged. Be willing to try things. Be willing to fail. Fail fast. Try again. Tighten up the loops, and visible leadership up front.
CC: And then just to add in terms of what you need to do now, just very tactically get your data house in order, whether that’s you yourself, which means clean up your customer data, have one single view of your customer data, whether that’s you or working with your partners like CSI to do that. And then two, have somebody on your team who’s starting to think through an AI risk framework. So what we talked about earlier, what risks are you willing to tolerate? What risks are you not willing to tolerate, and what policies do you want to put in place around those? Those are two important tactical items.
DH: I will say, this is not a pitch, not a commercial, but it is really interesting for us thinking about get your data house in order. We were very fortunate. Coming from the Apiture side, the acquisition last year, we had been on this kind of data journey for a number of years, and we had, as part of that, put in a heavy focus on data. And one of the first projects we did is, we have to get our own data house in order and get data cleaned up. So as it comes in, standardize it, normalize it, all of these things, and we think, well, there’s value for us there.
What we found, and had no idea, that’s one of the most popular solutions now that we offer for financial institutions is their own data cleaned back up. Cleaned, normalized, so that they don’t have to have data scientists, you don’t have to have data analysts or data engineers. Have a partner that does, I’d love for it to be CSI, but have a partner that can help you with those sorts of things. Don’t think that you have to, “All right, well, there’s a $5 million allocation that we’re going to need to make.” That’s just not feasible for most community banks. And something that we just took for granted, “Eh, that’s something we have to do for ourselves as a service provider.” Surprised by how much value financial institutions found in us being able to solve that problem for them.
SP: I think it makes a ton of sense. It’s like you see it on a very small scale whenever you go into ChatGPT or Claude. Its output’s only as good as the data that you give it.
CC: Exactly. That’s right.
SP: So the data’s not cleaned up, that’s going to be a problem with the output, right?
CC: That’s right. You got it.
SP: All right. Well, this has been a fantastic conversation. I’ve really enjoyed talking with both of you and learning from you. Welcome again to the team, and thank you for joining Banking on Community.
CC: Thank you, Saxon.
DH: Thanks, Saxon.
SP: And thank you guys for listening. We’ll be back soon with our next episode.