Is a Team of AI Agents Coming for Your Treasury? A Conversation with CSG’s Kyle Fisher

Speaker A: Foreign.

Speaker B: Welcome to another episode of FinTech Corner. We're here at AFP 25. I'm joined by Kyle Fisher. Uh, thank you for joining us, uh, with CSG International. Um, and I thought this was going to be a fun interview, uh, because when I read through the brief, it said disruptor. And I wanted to dig in just kind of right out of the gate. What about your career? Kind of jumping into organizations, seeing where they are from a, a technology, process, etc, perspective has kind of given you this almost Persona of a disruptor.

Speaker A: Um, well, I mean, I think I've had a somewhat storied career. Uh, you know, I've already been in maybe five publicly traded companies. Um, so I've jumped around a little bit. Um, but I think in each one, it's when you get in and kind of understand the current processes and procedures, we're coming from different, multiple different companies. You can get a lay of the land of like, hey, who's doing what and how they do the same. Treasury functions differently. And so that's where I, I think being the new kid on the block, you can quickly see like, oh, uh, we're doing maybe this process because we've always done that process, but it's not the best process. So there's always that like, why do you do it this way? Because we always have. And especially in companies that, you know, are more tenured, those processes are usually out of date. So that's where you can kind of get in there and as the new guy, be like, hey, let's leverage the new technologies to, you know, just create more automation, you know, reduce manual efforts, manual risks and just free up time

Speaker B: what have been almost, if you wanted to make like a, uh, list, maybe the top three innovators plan playbook that you've had in your back pocket. Every time you kind of approach a new situation. Have there been consistencies?

Speaker A: I mean, in treasury, right. I think, you know, your biggest ones are, you know, your bank, you know, your bank connections, right? Yeah. So do you have a tms? Yes or no? If you don't have a tms, how are you getting all your bank feeds? If it's you're pulling it manually, then you knew you needed to be doing something else. Uh, and then it's, you know, like your collections and your payments, right? Do you have visibility into your ar, your ap, how are those feeds into your bank? Do you have like automated payments files, right? Or are you manually like extracting a Nacha file and then manually upload? So I think those are usually where I first jump in and start because that's a big time saver.

Speaker B: Yeah.

Speaker A: Um, and then it usually audit is very happy with that as well because then there's less manual touch.

Speaker B: Right.

Speaker A: Especially those sensitive areas. And you know, even if it sits in like an, you know, accounting team or whatever, ultimately. Right. It benefits you in treasury because you get better cash visibility, quicker visibility. So that's always where I kind of, when I started new companies, understanding their processes and how it leads up to, okay, how can I better cash forecast.

Speaker B: So we obviously are in a different, you know, a lot has happened since COVID Um, I feel like everybody was pushed to adopt not just technology but the cloud, new SaaS tools, now AI, data platforms, etc. All these buzzwords are floating around as you've navigated that space. What have you seen as kind of the, the thread that you can consistently pull when implementing some of these systems? Have there been consistencies on that front?

Speaker A: I mean, I think, right. You guys will know, I think the cloud players are moving a lot faster. Your margins are better. So I think anybody who is doing on prem, I mean, erps, right. I mean you see Oracle in the news, right? Everybody who's like in cloud native solutions now are just moving faster and better.

Speaker B: Right.

Speaker A: And then especially, you know, the younger ones moving in that space like uh, the Travadas, right. Like you don't have all that legacy code so you can make some of those changes quicker and faster. And so yeah, whenever it's looking for these maybe third party solutions, the ones that are cloud native and potentially some of the younger ones sometimes are kind of doing it better than some of these older legacy systems.

Speaker B: Are there any trends that are taking you by surprise? Just kind of how fast they've been evolving and pressures that they've been putting on you just as a uh, program lead.

Speaker A: I mean the biggest. Right. Everybody's talking AI. So that's where, you know, I'm here at this conference to learn more about AI and talk to peers and figure out, okay, what should I be thinking about? How should I be thinking about it to. Because that's kind of, you know, that's our next move, right. Not in treasury, but across the globe. Right. Every aspect is going to be changed by AI and it's going to change fast. So you know, in one of the sessions they're talking like, you know, if you're, you're gonna, anybody who's not using AI, you're gonna get beat out by the person who is using AI.

Speaker B: Yeah.

Speaker A: So not as like a Competition, but like it's something to be thinking about. So that's where our company is very. Not only in our own solutions, are putting in AI into our products and solutions, but internally we have like every department is really tasked with. You need to start using AI.

Speaker B: Yeah.

Speaker A: Not like in whatever aspect. Whether you're just using you know, Copilot to like Google stuff.

Speaker B: Right.

Speaker A: Or if you're like really embedding it to, you know, start your vibe coding, that's where it's going. So that's where I'm trying to start. You know, especially I've got takeaways, I got a little notebook of AI stuff and like, okay, what could I start being doing? And I think that's, that's where a lot of focus is going to be because again like I'm leveraging these tools but there's other manual stuff I'm doing.

Speaker B: Right.

Speaker A: Can AI agents do that for me so I can free up more time for more analytics or you know, other projects?

Speaker B: So it's very interesting. In fact, just the previous conversation, uh, I was speaking to one, one of your peers, different company, um, and he was telling me about how they have adopted Copilot, just like you mentioned, they have um, chatgpt as well in their fleet. And the topic du jour is not whether or not they're going to use AI, they're using it. But now it's a question of where is it going to get its data from? How is it going to be integrated into every system? And so he kind of positioned the conversation back at me, which is not fun because I'm supposed to be doing the interviewing. And he said, what are you guys going to do about data portability? And what's interesting about that question is it's your data, right? So, mhm, it's always been your data. You can do whatever you want with it. Today we offer it back through APIs.

Speaker A: Right.

Speaker B: What's interesting about the world of AI is that might not be good enough. That might be slowing things down to a degree. I mean we're talking about AI stacks that have their own dedicated servers that feed them data exclusively. There's no other thing that is powering it. And I wonder, have you guys had any conversations internally about the future of your own data infrastructure to feed the AI programs that are in their infancy right now, but obviously going to be adopted really aggressively?

Speaker A: I, I wouldn't pretend to know the extent of that, but like our CIO organization are aggressively pursuing this and they are having those conversations, they have like internal like competitions with themselves on um, who can come up with like the coolest AI use cases and ah, like develop these things. So it's very much a top of mind. I, I don't know the ins and outs of all the conversations but again like our CIO organization is pushing extremely hard on that. So yeah, that is, yeah the intent, uh, like how are we storing our data, how are we managing the integrity of our data? Right. How do we make sure that it is being like clean data. Right. That's the whole thing. You can have data great, but it needs to be clean good data.

Speaker B: Right.

Speaker A: For then you know, AI to be used for. So it's, I mean I think it's a bit of a uh, revolving conversation.

Speaker B: Yeah.

Speaker A: Right. As, as these things keep progressing we're going to keep learning more and know how to do it better and faster. But yeah, that's our CIO organization. I think the single minded focus is AI. How do we deploy AI in like every aspect of our business internal and then in our products.

Speaker B: One thing that we've been seeing just you know we've, we've had AI in Truvada at this point for almost three years, uh, built on ChatGPT. We haven't hidden that at all. Like it, it is chatgpt that powers it. One thing though that has been curious about the rollout is that it started with effectively a search replacement.

Speaker A: Mhm.

Speaker B: People were using it instead of, you know what our hypothesis was at the beginning. You could use it to build reports and do all kinds of complicated things. Instead of people were using it instead of going to the search bar, just going in there and saying needle in the haystack, find this thing for me. Because I don't want to look for it myself or come up with the right filters to find it.

Speaker A: Right.

Speaker B: That was the first few years. Now I don't know if it's been influenced by GPT5 or even before at 4.1 but the questions are becoming paragraphs, they're kind of becoming more complex tasks. And where we're hitting a wall is we put a lot of restraint into our system. M and there seems to be this appetite of uh, no, just remove the restraint. I want it to have direct access to all my transactions balances and let it do whatever it wants in there. And I wanted to see what your take is on this. Do you feel like we're at a moment where we should be solving governance so that we can remove the restraints or do you think that we should still be cautious as we Move forward.

Speaker A: I mean there has to be caution, right? I know. Even when we are signing up for AI, the longest discussion isn't even the price point. It's the security. And like is our proprietary data safe?

Speaker B: Right.

Speaker A: And so I think that has to always be first and foremost. Um, because, right. Once it's out, it's out.

Speaker B: Right.

Speaker A: So, so it's gotta be, you know, yeah, maybe you're handcuffed, but we're all playing in this kind of new sandbox, so it's okay maybe if it takes a little bit longer for us to get in there. Because once we're in there, that's kind of it, right? We're all going to be playing an AI. It's not going away. So I think it's better to do it right the first time than to just kind of like go ham and then have to fix that later. But like, I agree it did very much. It's, you know, it replaces your like Google, right? Of like, hey, can you show me this transaction? Or like find this or uh, like what is that to then like could you like once you kind of build on that knowledge of like knowing and learning what AI can maybe do, that's when you start pushing those bounds where you might get from like uh, a mini sentence to like a full paragraph of like, can you do this?

Speaker B: A job description and you kind of

Speaker A: build up to it, right. And see like what, what are the bounds that it can do?

Speaker B: Yeah.

Speaker A: So I mean it's exploratory.

Speaker B: Let me ask you that million dollar question that everybody's trying to figure out at this conference. Will you trust Agentix systems with payments?

Speaker A: With payments and what? Like let them do the payments by themselves?

Speaker B: It's a good question. Let me frame it in a couple of different cases from least risky to most risky. The least and obvious ones right now are would you ever trust an Agentix system to effectively recommend the payments that should be made right now, with your approval, the medium risk is move funds strategically to balance, uh, your balance sheet and do it in a way that's taking risk into consideration. If you have FX, if you have a variety of different complex hedges, etc. Figure all that out, set up the money movement and then you approve it. And then finally the highest risk is, oh yeah, do all of that.

Speaker A: Just let it go. I mean, right. In this stage, no, because I think it's all trust but verify. Right, Right. So one, it would have to, you'd have to know where it's built. Like, you know, baby steps, right? You Got to get it ingrained into, like payables. You got to ingrained into receivables. You got to get it ingrained into your forecast. You got it ingrained into, you know, training it for, you know, what is, what are you aiming for? Right. Especially you have excess cash. What is your goal? Right, right. Are you doing share buybacks? Are you doing dividends? Are you paying down debt? Are, uh, you chasing interest income? Right? Like, what are your goals and your targets? And I mean, if you could do baby steps and get it up to that point where it's like it is constantly recommending what I want it to do every time. Well, I let it automatically do it, probably. No, but would I trust it to get to that point? And then it just says, hey, here's your list of things that we should do today. And then I can hit the button, say, sounds great, yeah, I'd love it. But, you know, it's that trust but verify. You got to get to that stage first. You got to build up to it. It's the same as the new employee, right? Like, you don't just give them the keys right out the gate. You got to train them, make sure, cool, you got this. Now let's move on to the next.

Speaker B: What's interesting about this technology is that the feedback cycle is so quick, right? With an employee, you might have to let that play out over a year and there might be a lot of errors that play out along the way. With these systems, you could have that cycle once per day and have, you know, the positive negative happen multiple times and you give it instructions that improve it, you know, materially in, in a single day. And it makes me wonder, and it's a framing for kind of the, uh, the last question that is kind of a futurist question. What do you think your job is going to look like over the next 10 years? You've already had such a transformative approach to Treasury, I'd say compared to your peers. One of the most interesting conversations that I had last year, uh, at the previous afp was with, uh, one of the treasurers at Block. That said something that really was confusing but made sense. After I thought about it for a little bit, they said that they felt like they're turning into more of a product manager for treasury than an actual treasurer. Almost as if it's their job to design systems, processes, and then make sure that those are working as they expect and less so doing the actual job of Treasury. And it was kind of like a really. That's playing out. But Then when you see kind of the operations of the apps that they're building and the degree of automation that they're shooting for, it made me wonder, is that the future of this job? But also maybe a hybrid of that is more leverage, Right. You're spending your time doing more strategic thinking than actually doing the day to day of Treasury. So I know that's a lot said, but the setup being futurist hat on, where do you think this job is?

Speaker A: I mean, I think that's right though, right? Because you say like, huh, is that the role, I mean, isn't that the role of Treasury?

Speaker B: Right.

Speaker A: I don't know. That's. I feel like that's always been the role of Treasury. Right. Is like building those systems. Right. Because again, then it allows you to do your treasury job, which your treasury job is, you know, how do I manage this cash? How do I manage like all these things underneath me? And the best way to do that is to build those systems in the right way.

Speaker B: Right.

Speaker A: So you can get the data you need, you can get the answers you need and then you can just spend more time doing your analysis or a high level thought process. Right?

Speaker B: Yeah.

Speaker A: Especially then when you're like, hey, what are we doing with our cash? Are we doing M, M and A what? Like, so I think building the systems and infrastructure faster, better, stronger is part of your treasury gig. And yeah, the future of this is just maybe a lot more of those systems are going to be managing AI agents.

Speaker B: Right.

Speaker A: And having multiple AI agents that are, you know, this AI agent is doing this portion of, of a historical, maybe manual, treasury workbooks or whatever. Um, and you know, you're gonna have this AI agent that's doing, you know, maybe your tax portions and like your other, you know, AP and your AR and your M and a kind of decision making and you have another one layered over that's doing your scenarios or your investment portfolio. Right. And then you're managing like little AI bots as opposed to maybe doing so much infrastructure work hopefully because you've already laid the infrastructure or then, you know, maybe not have as big a robust treasury team, which, I mean, in my case it's usually a pretty lean, mean team anyway. So.

Speaker B: So it's almost like, correct me if this is too big of a stretch, but it's almost a transition from a risk management role to a risk architecture role.

Speaker A: Yeah.

Speaker B: Where I think it's fair not really actively managing it anymore because there's software managing it M. But it's more like you're designing, you're designing things that many pieces of software and orchestration are doing.

Speaker A: And I think it comes back to that original of like, do we release the guardrails?

Speaker B: Right.

Speaker A: Because. Right. If you get to that point where there is so much systems managing and AI managing so much stuff, you got to make sure that you can trust it.

Speaker B: All. Right?

Speaker A: Because then of course the fun, the most fun of a Treasury is then internal audit comes knocking and asks, hey, but is this all done correctly? Is it all done safe? Right, Right. And I think that's where audit is going to have a giant. You know, our IT team and our audit team is going to get probably closer together as they're going to probably be the ones tasked with really managing some of that.

Speaker B: Right. So I'm curious if this all plays out, let's say, because I don't really hear much about if I had the free time, what would I, what would I do with it? Right. And it's usually answers like, well, you know, I use it to analyze and you know, do forecasting, stuff like that. But I'm curious because you already run a very complex operation today. Would you make it any more complex if you had the leverage to take advantage of in the form of agents and things like that, would. Would you push the boundaries even further?

Speaker A: I mean, if you say, if I got some free time right now, I'm going to go start playing more with agents. Like that's where you know, my takeaways from some of these sessions, you know, yesterday and today is hey, here's some like agents that I've built. And I mean, I think even internally we have the tools to build our own agents to some degree.

Speaker B: Right.

Speaker A: You know, do we have the best and maybe the correct AI ones? Maybe not at the moment, but they're coming, um, you know, to the individual users. A CIO org has more. But that's where like my free time, if I do have free time, like I'm playing with AI.

Speaker B: Yeah.

Speaker A: And to that extent, if it's been that like initial of like, oh, just Googling like a one word thing to now, I, I'm, I'm querying paragraphs of like, what's the bounds that you can provide back. And I think playing that with agents is the next step for me.

Speaker B: Absolutely.

Speaker A: And really seeing how far can I take this like internally myself with the current tools and then when I get access to the latest and greatest tools, can we take it further? I mean it's going to happen. And again, it's the uh, if you're not Using AI in the workplace, you're going to get left behind because just like if you weren't using the Internet in the 90s, like, you're going to start getting behind.

Speaker B: And with every technology, let's say, you know, truncation of time, it almost seems like the side effect is the leverage you get isn't just like speed and automation. It's, you can take more risk. Right. Because you have higher levels of precision. It's why, you know, I think of stories of Apple and other kind of big tech companies that operate on the bleeding edge of operations with regards to stock, um, of, you know, units that they distribute throughout the world, where in the past they might have to fill a warehouse to have stock for a market, and now they can have it within, you know, plus or minus one day of stock and trust that all of their infrastructure will be able to supply it. And all of the math that you had to do, you know, working backwards from I have a MacBook Pro that's on a shelf to did I have enough stock to manufacture it in the first place? Did we have enough money in operating accounts to pay for that stock? Uh, did we hedge the balance sheets appropriately so that we don't get hit every single time by FX fluctuations that are happening in, you know, wherever because of tariffs, et cetera? That complexity has been worked out to such a degree that they can operate on such a bleeding edge of one day of supply, basically.

Speaker A: Right.

Speaker B: And I'm curious, do you, do you feel like the same will play out because of agentic workflows?

Speaker A: I think so. I, I think it'll allow all of us to just tighten up. Right. Not have to operate with so much, you know, buffer and wiggle room or, you know, sandbagging my forecast. Right. And I, I think we're going to be able to shrink that degree of, like, you know, whatever you operate under, if it's a, uh, you know, a five degree of, uh, margin or, you know, to maybe a two.

Speaker B: Right.

Speaker A: And hopefully shorter and shorter. Right. Because you're going to get that and you'll be comfortable with it. So

Speaker B: it's a hard question because it doesn't exist yet.

Speaker A: No, but it's, it's. Where are we going? And I think no one denies that's where we're heading.

Speaker B: So, so for our viewers, a lot of the, the insight that they get is kind of the access to the brain of, uh, the person that we're interviewing. And I'm curious if you were to give one piece of advice to the audience that's watching this podcast right now and kind of evaluating their future. Whether it's purchasing a piece of software or creating a new system internally. What would it be?

Speaker A: Any piece of advice? Oh, man. Um, I think it's always better to do it right the first time and to be diligent and thoughtful with your execution. Even if it's slower, it's better in the long run. So whether that's right, like, if it takes longer to evaluate the software, it's fine. It's better to, uh, you know, not rush it and then have to unwind it.

Speaker B: Right.

Speaker A: Um, and. But like, ultimately you got to do it. So I think that's where it's like, what are your long term goals? And if it's getting a TMS system, it's getting into a new ERP, whether it's developing these AIs, it's like, start some form of it today. Yeah, start it moving. Don't feel rushed to like, you know, pull the trigger if you're not feeling comfortable. Like, make sure you get comfortable because it's the. You want to make sure you're, you're signing on for the right piece for you and your needs, and then once you have it, just go. So that's where, I mean, it's. I know it's hard to dedicate the time and the resources, but I feel like that's always where I get the most bang for my buck. Especially when I join a new company is like, yeah, hey, I know I have a lot to do, but like, this is where I got to spend my focus. So, like, for the first however long, this is going to be the only thing I'm really going to dedicate time

Speaker B: to and increase the space to get

Speaker A: to the value to do all the rest.

Speaker B: So that's awesome advice.

Speaker A: I mean, that's probably the same advice anybody would give, but, you know, it's the, like, don't, don't get in over your head and over your skis too early. Like, make sure you pick the right thing and go for it first.

Speaker B: Awesome. Well, Kyle, thank you so much for your time. We appreciate you coming by and hope you have a great rest of the conference.

Speaker A: Yeah, you as well. Appreciate it.

Speaker B: Sweet.