Inside Grasshopper Bank’s Claude AI Integration

Jim Marous (00:12):

Welcome to Banking Transformed, the top podcast in retail banking. I'm your host, Jim Marous.

Jim Marous (00:17):

What if your business banking clients could ask a question in plain English and get an instant intelligent answer? No spreadsheets, no manual reports, just a client, their data, and AI working together.

Jim Marous (00:32):

That may sound like science fiction, but it’s happening right now. Grasshopper Bank has become the first US bank to enable customers to connect their business banking data directly to Anthropics Claude.ai assistant, and the implications are truly staggering.

Jim Marous (00:49):

Today, we're joined by the visionaries who made this happen: Chris Tremont, the Chief Digital Officer, and Pete Chapman, the CTO of Grasshopper Bank. They're going to review how they built a secure bridge between banking and artificial intelligence, why they race to be first, and what this means for every business owner in America, and actually every finance institution.

Jim Marous (01:12):

This is a conversation where the future of banking actually gets written live. The companies that understand this shift will thrive while those who don't will certainly be left behind.

Jim Marous (01:25):

So, Chris and Pete, before we start, can you introduce yourselves to our audience a little bit?

Pete Chapman (01:30):

Sure. Hey, I'll go first. Jim, thanks for having us on. I think first and foremost, excited to have the conversation. You used the term “visionary” kind of in your introduction there. That's a very strong term when you think about guys like Chris and I, but we'll do our best to kind of live up to that and looking forward to the conversation.

Pete Chapman (01:49):

So, I'm the CTO here at Grasshopper Bank. I've been in financial services for the entirety of my career, which is coming up 25 years or so at this point. And I've spent my career kind of on both sides of the fence working for fintechs, working for banks as well, and so I've gotten to see the space kind of from both sides which has been interesting for me.

Pete Chapman (02:12):

So, here as the CTO at Grasshopper, I work closely with my colleague Chris and others ultimately to deliver the technology, architecture and platforms to enable the business, and there's a lot that goes into that. I've got a great team that I work with that are way smarter than me on all of this stuff, but it's been fun, it's been a good ride, and I've been here for about four and a half years or so. Chris, over to you.

Chris Tremont (02:37):

Thanks, Pete and Jim, thanks for having us. I'm Chris Tremont, Chief Digital Officer for Grasshopper Bank. Much like Pete, I've spent all 20 years or so of my career in financial services working for a larger regional bank, and then 12 or 13 years with a smaller bank up in Boston.

Chris Tremont (02:55):

Pete and I worked together with a company called Radius that we kind of digitized, and eventually, sold to LendingClub in 2021, and then a number of us kind of we “got the band back together” and joined Grasshopper in 2021, and small company.

Chris Tremont (03:13):

So, like Pete, we wear a few hats, both him and I. My primary remit is sort of on the business side, leading our direct business banking efforts to small and mid-size companies, our indirect fintech sponsor banking or banking as a service efforts, and then some affinity banking that we do through a partnership with the Auto Club Trust in AAA, and then product and marketing report up to me as well. So, we're looking forward to the conversation.

Jim Marous (03:44):

So, this is kind of interesting. So, we really have a group that has banking background, but also fintech background, and is taking innovation to new level like a fund sandbox for both of you. So, can you walk us through what led Grasshopper Bank to become, as I said, the first US bank to integrate Claude.ai or a LLM for customer banking, data analysis, and what was the aha moment that spurred that initiative and actually had you do it?

Jim Marous (04:12):

Because we have a lot of people talking about it in the industry, but man, you can count on less than one hand how many people are actually doing anything that is customer-facing. So, what guides you to this point?

Pete Chapman (04:26):

Maybe I'll take a stab at that question here. And I think to answer that question, you got to kind of go back a little bit and think about how we've jumped into AI over the last year and a half or two years.

Pete Chapman (04:37)

And so, we've been at it for the better part of two years really, and trying to understand how could a bank like us leverage AI, I think first and foremost to drive probably what we would call internal back office efficiencies, and then how could we eventually get to know this technology and over time, develop some sort of externally facing roadmap that leveraged this to create a better digital banking experience for lack of a better term there.

Pete Chapman (05:02):

And so, we've been doing that for two years, and the focus, like I just said, has been mostly on back-office efficiency, and well, we found a couple of things. There have been a lot of lessons learned kind of as we've deployed that.

Pete Chapman (05:18):

And I think as we stepped into it, first and foremost, the thing we said to ourselves is there's a lot of concern out there, is AI going to take my job? You hear it everywhere, everyone's talking about that, and that's not what we're trying to do with AI.

Pete Chapman (05:33):

We have the luxury of being a relatively young, quickly growing and scaling company with a really sharp team of people. So, we're not trying to necessarily go in and replace 25% of our employees like a lot of big companies are, we're more trying to make our crack team far more capable so as this bank continues to grow, maybe we don't have to kind of leverage the more hiring the way a traditional bank would.

Pete Chapman (06:04):

So, there's that component and we deliver that message kind of like front and center as we talked with our staff about, "Hey, we want to start learning AI." So, we wanted to make sure we disarm people right away from thinking, “I'm going to lose my job.” That's not what we're trying to do. We're trying to make you more efficient, we're trying to have you live up here, we're trying to get out of the repetitive tasks that are down here, and ultimately, we think this will be a good thing for our business as this bank grows.

Pete Chapman (06:31):

And so, we've been on that, that kind of like journey for a year and a half, two years or so. We kind of set the expectation front and center culturally. That's how we're going to go about it. We're big Google shops, so we've gone kind of deep down the rabbit hole with Google, Gemini, and we've got that kind of embedded deeply kind of across our entire infrastructure and ecosystem.

Pete Chapman (06:52):

And we said to ourselves this year especially, "Hey, what if we could give the typical knowledge worker one to two business days back by the end of the year and free them up to be able to once again operate up here a little bit more instead of doing the repetitive tasks."

Pete Chapman (07:06):

And we've made a lot of progress in that, a lot of progress. In fact, we've blown by a lot of our expectations that we had at the beginning of this year and certainly, in the last year as well. And so, this has allowed us I think, to really understand artificial intelligence as a technology, how to engage with it, what works, what does not work, et cetera.

Pete Chapman (07:25):

And all along the way, we kind of said to ourselves, okay, once again, once we're good at this stuff or once we get better at this stuff and better understand it, let's find a way to make sure we can deploy this thing out to our customers.

Pete Chapman (07:35):

So, there've been a bunch of aha moments as we've thought about the internal deployment of AI. But stepping back and thinking more specifically about the MCP server itself, there are a number of us that we call ourselves maybe AI native, whatever that means right now, but personally, in my life and I have my consumer banking account with a separate bank because Grasshopper doesn't offer direct consumer accounts, but I've got my finances elsewhere.

Pete Chapman (08:02):

I have been a PFM nerd for the better part of 20 years, and I've always said, "Hey, look, there's something here about, hey, we've got all this data, we've got to find a way as a financial institution to be able to kind of advice our clients a little bit better in a really proactive way."

Pete Chapman (08:18):

So, I've tried every PFM under the sun, but I have taken personally my banking data, my PFM data, and for a year, year and a half now, I've been dumping it into my ChatGPT, that's what I use personally in my life.

Pete Chapman (08:34):

And that thing has really been able to engage with me and help me better understand my finances, better understand my cash flow, better plan my investments, et cetera, in a way that like the old school PFM tools just have never been able to do. And I'm doing all that in a super and efficient way where I take a CSV file and I anonymize it to some extent, and I dump it into the LLM.

Pete Chapman (08:58):

And so, as I'm doing that, and there are others at the bank that have been doing that exact same thing as well, we just said to ourselves, guys, there's got to be some way over time to simply standardize and secure our data and plug this thing into some of the LLM providers. And Jim, if you think about that right there, that's the thing that we as banks have been doing forever with QuickBooks, with all of these different third-party integrations that we've enabled.

Pete Chapman (09:25):

So, it's very similar to that but we just said to ourselves, as soon as there's some LLMs out there that start to standardize around some sort of format, let's find a way to safely securely, build around that format and start to enable kind of these third-party connections.

Pete Chapman (09:40):

And so, MCP has thankfully come out and like a couple of the big shops have said we're going to use this thing, and so it was the right time and the right place for us to go about something like that. So, anyways, I'm not sure if I got to your question specifically, but that's a little bit about the background there and what's kind of led us to this point.

Jim Marous (10:02):

God, there's so much to unravel there. Number one, we have done a great job of, as you said, the ability to give knowledge to customers has been there, but we do it almost like a report. There's not a so what to that.

Jim Marous (10:21):

And basically, your history, I'm familiar with it, that has been my challenge to a lot of these organizations that do a great job putting data into nice formats, is go further because most financial institutions have a hard time taking that to the next level and going, “I'm going to provide you the so what.” Either I'm going to provide you how I'll market you based on what I learned or I'm going to give you access to the knowledge, but give you the tools to find out what you should do with that.

Jim Marous (10:52):

And I think the big leap here, especially with LLMs, is the ability to go, “I can feed you all the data, I can tell you what your cash flow's going to be, I can find out recommendations based on what's happened in the past, but deeper than that, I can build that so what through time and through the question answer part of it.”

Jim Marous (11:13):

So, Chris, you've mentioned in the dialogue, and for those who may not be familiar with model context protocol, can you explain a little bit about what that server is and how it acts as a secure link, but how it really makes it so that this integration of data with an AI tool outward to the customer can be done better that way, where it would not have been able to be done before.

Chris Tremont (11:40):

Yeah. Maybe I'll take a shot at that. So, the MCP server model context protocol, simply a standardized format that I think was initially introduced maybe by Anthropic, those Claude and a couple of the others have slowly kind of said, “Fine, we'll start to support this format as well,” no one wants to support … they're all in competition with each other. But thankfully, they're all starting to kind of standardize around this a little bit.

Chris Tremont (12:08):

And so, I think at this point, you got Open AI who's done some standardization around it, and in turn, Microsoft who started to do some standardization around it, and I think Google's eventually getting there, and we'll see kind of like what xAI does but that's hitting on most of the big ones right there.

Chris Tremont (12:26):

And so, this thing, it may sound complex, but all this thing is, the way we look at this thing is the same way we look at honestly at API banking which is something we've been doing for a long time. MCP is simply a format and you got to still kind of approach MCP with the same exact kind of InfoSec approach as you would any API.

Chris Tremont (12:50):

So, you got to look at this thing, you got to harden it appropriately, you got to kind of think about least possible permissions, you got to kind of go up and down with it and make sure you do the proper pen test, vulnerability assessments, et cetera.

Chris Tremont (13:04):

One of the interesting things Jim, is when we came out with this thing six weeks ago, or seven weeks ago or whenever that was, we heard a bunch of reaction from the market, and one of the common questions that came from the industry was, is this thing secure?

Chris Tremont (13:18):

And I would say there is nothing that is inherently insecure or secure about MCP. It's all about standard kind of like DevOps practices that you've got to surround this thing with and deploy it appropriately, and that's been our approach on this thing.

Chris Tremont (13:34):

And so, that's why we would say we were cautious as we kind of initially kind of stepped this thing out, but alongside that, this is a brave new world. Like these are uncharted territories here. And so, we intentionally have gone out to market with this thing as a private beta initially. So, just a handful of clients who are clients that are honestly all in the AI space are leveraging it right now, so they know, they understand these things.

Chris Tremont (14:01):

But the other thing most importantly is this is read only access for now. So, we're not enabling any sort of right access and once you start talking about enabling this thing from a right perspective, boy, that opens up a lot of different things we'll probably do that at some point, we can talk about that a little bit later.

Chris Tremont (14:17):

But anyways, high level Jim, MCP, I think that the summary there is nothing inherently secure or insecure about this protocol by itself. You've just got to think about deploying this thing in your larger infrastructure, the exact same way you would deploy a new API endpoint or anything along those lines.

Jim Marous (14:36):

So, Chris, obviously, you took it from being an internal tool to being something that your customers can actually use. So, while it provides the ability for your employees to build a better profile around your business customers, this allows your business customers to leverage a tool based on information at their bank account to better access data in a way that they could have done on their own, but it would've taken a lot more work as was initially mentioned by Pete around what can be done.

Jim Marous (15:12):

What have you seen your customers do with the tool you've given them in the last six weeks?

Chris Tremont (15:21):

Well, Jim, we're pretty early on, and I think I'd like to layer into what Pete was talking about to your earlier question around how we got here. And for us, it always starts with a core value of being client first and aligning to our mission around serving the business and innovation economy.

Chris Tremont (15:40):

And so, the development of AI over time is probably not where Pete and I started in 2021 when we were rebuilding the platform, but it was like how do we better serve our customers? And this tool has kind of come along to your point that we can now put in front of our clients to help them better manage their cash flow, their operations, things like that.

Chris Tremont (16:01):

And if you look at an SMB, you'd probably say small business, there's three or four things that they need in their day-to-day operations is they got to get money in, they got to get money out, probably need access to capital from time to time, and when the money's in the bank help me better manage my cash or my treasury management tools.

Chris Tremont (16:22):

And so, I think this is where Pete and I would say this is where from an AI perspective in this MCP tool really has started to step in and help people build, I don't know, a cashflow statement a little bit. Like give me all my transactions and categorize it in such a way, or a faster way to sort through transactions, build maybe a cash flow statement, do some more analysis, maybe a little bit of a cash burn, forecast, things like that.

Chris Tremont (16:54):

So, we're very early on though, in terms of being able to provide a lot of examples. As Pete said, we look at it kind of akin to API banking, which he and I were probably looking at six, seven years ago, and you find some learnings along the way of how do you provide a better connection and an interface.

Chris Tremont (17:16):

And I think we're in a better spot here in 2025 than we were in 2018 or '19 when this stuff was first starting to roll out, at least from our perspective. And it'll be interesting, maybe we'll come back on in a few months and have a few more learnings for you around how our clients as we move out of the beta phase of this platform are really using it. And I do think the next unlock will get past just reading the data to writing as well, transactional based in a safe and sound manner.

Jim Marous (17:45):

And it's interesting because we're even seeing the AI tools evolve to the point where I've talked about it on a couple podcasts where ChatGPT now ask follow-up questions around what you're asking it to do, which provides us a whole wealth of opportunity.

Jim Marous (18:03):

Especially, I'm a small business customer, and I did not know what I was getting into when I started my small business at all. I mean, there's so many different war stories, but the more we can provide tools to the small business customer around, here's data that you have, here's what you can do with the data based on what normal businesses do, and get them over that whole learning curve.

Jim Marous (18:27):

I mean, some organizations obviously are taking it as Pete did, taking the data and doing it themselves anyway, but you provide that bridge and you provide ways to use the tools that are out there and continually evolve with them, and knowing how as more and more customers start to use the data in your beta site, you're going to go, "Oh man, we never thought about this opportunity that was there." And it not only helps the customers but it helps the people that are assigned to these customers from the banking organization.

Jim Marous (19:01):

I'm thinking about how much more powerful AI can make the calling officer in meeting with customers and helping them do their business better in the way that shows some empathy but also shows that we're actually partnering with our business clients as opposed to simply seeing them as a revenue source.

Jim Marous (19:22):

So, Pete, how did the partnership with Narmi originate and why was Narmi important to this whole capability?

Pete Chapman (19:33):

Let me answer that question but just real quick, let me go back to something you said there because I think it was just fundamentally important to our approach to AI, and to how this whole thing is working as well.

Pete Chapman (19:41):

And ultimately, that conversation we just had there was, okay, I think I might know what I would use this MCP server for, Chris knows what he would use it for. What are our clients going to use it for? And that's the power of this whole thing, is you've got these LLMs that are open, that are extensible, that are so powerful that can do really a lot of different things.

Pete Chapman (20:08):

And so, the use cases that … so in a traditional product development kind of lifecycle, we've got kind of tight use cases we're looking at and we're going and we're building for. This one's more about opening up and allowing for millions of different use cases. And so, what we're going to hear from our clients here or we have started hearing, but we're going to continue to hear from our clients, use cases we never would've thought about.

Pete Chapman (20:33):

And we see the same thing and when we deployed internal AI as well, the first thing we did was – well, not the first thing, but one of the first things we did is we said, let's get these tools in the hands of people that are closest to the work, and kind of enable them and build some sort of feedback loop and ask them how they're using it.

Pete Chapman (20:53):

That's going to be way better for us as we think about leveraging AI than me sitting up here being like, “Well, I want it to do this or I want it to do that.” So, I think that's just broadly, yeah.

Jim Marous (21:04):

It's crowdsourced innovation. I mean, it's something we've all wished we could find a way to do, but everything that's being done, you have access to see how people are using it, you're able to get these people involved. Small businesses by their own nature, innovative by their skill sets and what they're trying to accomplish, this just becomes very interesting as a way to innovate beyond the organization, which we've all talked about, but very few of us do.

Jim Marous (21:35):

We listen, but only deploy so much and again, that gets back to the beginning of my conversation, is that it's great to be able to provide new ways of looking at data to a customer, be it a business customer, a consumer, a corporate customer. It is much more important to provide them tools to deploy in such a way that the learning curve goes both directions, that you're learning at the same time that they're learning to use it, and something that they may not have dipped their toe in initially.

Jim Marous (22:05):

So, I'm getting back to the Narmi question a little bit about how the whole orchestration happened. I know we're jumping ahead and behind in the sequence of things, but I think it's important for organizations because there's no one out there today that I know that's doing what you're doing on any level.

Jim Marous (22:21):

We're all talking about it, and you're deploying it in a context that makes it so it's manageable, but it's also exciting because you kind of know already where some of those landing spots should be, it's just a matter of seeing, does the consumer or the small business agree?

Pete Chapman (22:39):

Yeah. Yeah, absolutely. All well said Jim. Now, back to your question around Narmi, let's talk about kind of how that came about and what role do they play. And so, for those of you who don't know, Narmi is a digital banking provider based out in New York. We’ve known these guys for a long time.

Pete Chapman (22:56):

Chris Tremont in a past life, built out a partnership with them to build out I think their first digital banking experience seven, eight years ago or something along those lines. So, these guys have been good partners of ours. We call them design partners and we really believe in what Nikhil and Chris have built over there, and kind of where they're going. And so, they aren’t a vendor for us, they’re a strategic partner for us.

Pete Chapman (23:24):

And so, we talk with them about strategic initiatives, they talk with us about strategic initiatives. And typically when they're doing something new, we like to think they come to us and they shoot us a text first saying, "Hey, are you guys interested in doing something like this?"

Pete Chapman (23:39):

And so, this idea had been on our kind of wall for a little bit, thinking about how do we get this thing deployed eventually. Narmi runs our digital banking ecosystem through online mobile in the entirety of the digital, we do some APIs through them as well. So, looked perfectly logical for them to be the partner that kind of takes this thing, turns it on, and starts to kind of enable it for us.

Pete Chapman (24:08):

And so, I think they were thinking about at the same time we were thinking about it, then we had a conversation around it, and then suddenly about four or five weeks later, we're out there kind of like deploy phase and we moved very, very quickly on. So, that's kind of how it goes with these guys a little bit.

Pete Chapman (24:25):

Once again, strategic partners and we talk with them more about strategy, less about typical things he would talk with a vendor about, and it just kind of crossed and intersected right at the perfect time. And Chris, you can probably talk a little bit more about the long-term partnership with Narmi, and you've known these guys for a long, long time, and kind of built that thing out 8, 9, 10 years ago.

Chris Tremont (24:51):

Well, that's a great point, Pete. Like they were sort of the catalyst for making this happen or the enablement Jim, that the technology for us to be able to bring this to our clients. And I mentioned it earlier, but it's like going past, like you said earlier, just some canned reports or tools that are built into a platform and now moving into the, we used the art of the possible or there's a little bit of learning here around what you're able to do with this MCP server.

Chris Tremont (25:23):

I go back to six, seven years ago and it was actually with a different technology partner, but we were trying to extend APIs out of our corporate banking system, and we thought it was going to be a home run, so many people are going to want to use this.

Chris Tremont (25:38):

And then you get into it and you realize, well, you need a dev shop to be able to connect this to certain ERP systems or a platform to make it useful, and a customer needs to be of a certain size and the complexity and we never really found product market fit.

Chris Tremont (25:55):

And we were like this is kind of interesting, and now, you fast forward to today where I think this technology really makes it much more accessible to smaller businesses, middle market companies in a place where they're already interacting, whether it be through Claude or ChatGPT eventually, like Pete, will talk a little bit about maybe where our roadmap is going, but sort of this next iteration of it.

Chris Tremont (26:19):

And even the connectivity between our commercial banking platform powered by Narmi into Claude and making that happen, we are working on part of why we're still in a beta phase is like how do you make that easier for the small business to connect their account, their Grasshopper account into Claude? But I just think it makes it so much more powerful than some canned reports.

Chris Tremont (26:47):

And this was just part of the roadmap with Narmi, when Pete and I joined, and we're just two of the folks talking today, there's a whole team at Grasshopper that's helped put this together, but we reconnected with them in 2021, and it's just been part of a roadmap where we're constantly trying to say, we're trying to provide a really great banking experience for our customers.

Chris Tremont (27:08):

We've got a pretty decent digital acquisition engine but now the pendulum swings a little bit, and you say to yourself, you built a fairly large portfolio, how do you drive a better value proposition which in turn drives more profitability for the bank? And this is just another tool or an engagement piece for us that we're excited about, but it's really early days obviously too.

Jim Marous (27:31):

Let's take a short break here and recognize the sponsor of this podcast.

[Music Playing]

Jim Marous (27:38):

So, Chris, you intentionally launched this as a read-only feature where Claude cannot initiate transactions like an agent. Can you discuss that decision, but also whether transaction capabilities might be added or what may be added from the standpoint of dialogue?

Jim Marous (27:57):

As I mentioned with ChatGPT, it asks you follow-up questions, is that capability part of this future view of how the customer is going to be able to actually access their data in a new way through LLMs?

Chris Tremont (28:15):

Pete, I'll take the start of this and then let you layer in some of your thoughts as it relates to the technology and the security. But Jim, at Grasshopper we're big fans that have taken the approach of sort of crawl, walk, jog, run. And I'd say like, as we were getting this out the door-

Jim Marous (28:32):

But it's a short span between those things too, I mean.

Chris Tremont (28:36):

We like to move fast, we like to-

Jim Marous (28:39):

We all have that pace, but it's do I run in the fifth year or the fourth year? That doesn't get to the run until the pace is a little quicker, so that helps.

Chris Tremont (28:51):

So, I mean, maybe our pace can be a little bit quicker than others, but we like to get out the door when we feel comfortable with something, but not bet the bank on it so to speak. And so, we wanted to get it out the door, I think everybody felt comfortable with this being a read only technology, let's learn from how our clients use it, the voice of customer is going to be really powerful over the next couple months or couple quarters.

Chris Tremont (29:19):

And to your question, what Pete and I sitting here right now is that we'd say yes, definitely is on the roadmap to get to potentially more functionality like a right functionality. And what we're talking about here is maybe you're making instructions to send a wire or ACH, moving money from this platform where you're not technically logged into an online or mobile banking app, you're doing it where you live.

Chris Tremont (29:44):

So, we do have that on the roadmap in terms of when we're going to roll that out, sort of to be determined. But there's been a lot of thought internally and Pete and his team lead this around, how do we iterate and how fast do we move? So, Pete, I'll kick it over to you maybe to layer in some additional comments.

Pete Chapman (30:02):

So, I think the focus for right now is to get this in more clients' hands to get feedback. And so, up to this point, the enrollment experience with this thing has been not simple. To be fully frank, you have to know a little bit of Python. You got to go edit a clawed, config file, that's obviously not something you can deploy massively, but it's helped us get out kind of out there with a handful of different clients.

Pete Chapman (30:35):

So, we're focusing on ease of enrollment first and foremost. We've got an OAuth enrollment flow that is going to be going live here shortly. In fact, we just put it into our production environment this morning, and I was live testing it right before this call.

Pete Chapman (30:52):

So, that will help. So, instead of having to go edit a config file or something along those lines, it does SSO. So, it kind of logs you out there, opens up your Grasshopper Bank browser, and you kind of just authenticate. So, that's helpful. So, first and foremost, ease of enrollment.

Pete Chapman (31:11):

The second thing is we've got to get this thing extended to more LLMs, it's Claude only right now. So, we went out and we know the market numbers, obviously you got ChatGPT, and then you got Gemini, and then maybe xAI and then some followers after that, but Claude's not the top one or two for sure.

Pete Chapman (31:30):

And as we've gone out and we've talked to our clients and run a couple surveys, what we've seen is ChatGPT is number one, Gemini's number two, Claude is kind of like a distant number three in usage amongst our small businesses.

Pete Chapman (31:46):

So, we've said, look, we got to deal with enrollment, make sure that's nice and simple, we've got to go extend this thing into other LLMs that are the LLMs that our clients are using. So, we should have ChatGPT available, probably 30 days from now, it'll be available. So, once we get those two things kind of behind us, we feel like we can probably open this thing up.

Pete Chapman (32:10):

So those two things, once again, ease of enrollment, and access to ChatGPT and then eventually Google, we can start to open this thing up beyond just a private beta, which is what it's in right now. And we'll probably open it up into something that's a little bit more public and we'll actually start going and engaging with our clients directly saying, "Hey, look, this thing is available." We haven't done that. We just haven't done that up to this point.

Pete Chapman (32:35):

So, once we get that behind us, then we're going to get a ton of feedback like either this thing's great or hey, you guys are a bunch of fintech geeks that don't know what you're talking about and no one uses this thing or something in between. We don't know, but we're going to try to stand up some sort of ecosystem for people to give feedback for us to kind of hear that feedback.

Pete Chapman (32:55):

Maybe kind of circulate some of that feedback more broadly amongst the community of clients that are using it, et cetera, just once again, the same way we did our internal AI technology approach. So, we're going to try to do all that in the next 30, 60 days, and then maybe eventually get the right access which is going to be a little different.

Jim Marous (33:11):

What's interesting, it reminds me of the early days of Erica with Bank of America. Everybody said, “Really is this really that big of a deal?” And as it evolved, they continually changed what the capabilities were, what was going to be able to be captured, what was able to be said, what was going to come out of the conversations, what the conversation meant.

Jim Marous (33:31):

Because the technology and the mindset of the consumer has changed so dramatically certainly since COVID, but even before that, people are starting to understand the power. And I think you're in an interesting space right now because you have moving targets as well as moving propositions from the different LLMs.

Jim Marous (33:50):

As I mentioned, ChatGPT, all of a sudden, starts to have follow-up questions and Claude.ai isn't there yet, but Claude.ai from my perspective, has some of the best writing skills, some of the best dialogue skills out there while ChatGPT and Perplexity has another version of what they're best at.

Jim Marous (34:08):

I'm sure you're going to find customers that are going to gravitate to one platform or another, the learning process, the listening capabilities, interestingly, that AI keep on saying that all these LLMs are great listeners and you've got to structure your questions the right way – well, you're going to be able to build that as you go. And it's one of those things that you don't really know where the landing zone is from just talking to of you two today, and that kind of makes it even more special in that the evolution, this is going to go very fast.

Jim Marous (34:43):

And once you get a couple moments where you go, “Oh man, they nailed something here that's very marketable today.” And again, from the very first conversations that both you guys had with me, you're looking at how this can be deployed as opposed to simply, it's just a new way to look at the exact same data, the exact same way, and what value transfer happens.

Jim Marous (35:07):

And I think that's really where we are today when we look at these LLMs and the capabilities that it really expands. But on the other hand, it puts a lot of pressure on both you and the rest of your teams to actually evolve as the potential of the LLMs evolve.

Jim Marous (35:25):

Because it's not like you're dealing with something that's static and they're not even like in the way they address issues. And it's not like I don't think you're getting a whole long lead time on what's going to happen next. You go, oh, look at it, some of this changed major with this platform and I didn't know it.

Pete Chapman (35:45):

I mean, Jim, that's all very well-said and we see that as the power of all of this. So, you've got these giant corporations (Google, Anthropic, Microsoft, Meta) that are battling it out right now and pouring billions and billions into R&D to iteratively improve their Gemini 2.0 to 2.5, and we get to be the benefactor of that.

Pete Chapman (36:15):

And so, we're focusing on safely and securely in the right context deploying out our structured data into those ecosystems so our clients can take advantage of the arms race that's happening right now.

Pete Chapman (36:30):

And you talked about like GPT-4, I don't think GPT-3, 5 actually had context, but 4 did, GPT-5 does, and like now my personal GPT-5 knows the last two years of my life and knows everything about me pretty much-

Jim Marous (36:45):

And it happens without an announcement, it just happens. And you go, anybody who knows it, their eyes light up and some of them, I'm sure it went a good couple weeks before I realized, wait, this is actually asking very intelligently … it's not just prompting me with what I said in that conversation, it is saying, “Do you want us to use your red and black branding for this?” You go like, “Holy crap, I don't remember actually saying that in any conversation, but they obviously picked up something here. Maybe they're watching my Amazon purchases.”

Jim Marous (37:20):

But as you look at this, your end game, it seems to me, and please correct me if I'm wrong or tell me that I'm right it really empowers your clients to do more with what's in the marketplace today. You're almost like their learning lab for doing things that they would probably have gotten to, but you're actually making that link faster Pete, than you were using before. I'm not making you pull down all these files from everywhere.

Pete Chapman (37:51):

That's right. And Jim, I would say and you know this and your listeners know this, that there's something happening in the industry right now where you've got some banks who are saying, “No, we're not going to enable connectivity out into the third party ecosystem into the fintech ecosystem; in fact, we're going to charge you or them to enable that.”

Pete Chapman (38:12):

And so, this has been something that's now been happening for a while, where we would say no. In fact, we're going to take the exact opposite approach. It's not our data, it's the client's data, and if the client says to us, "Hey, Pete, hey Chris, I want to securely take that data, my data and kind of pull it into this ecosystem," we're going to say, "Okay, let's try to find a way to make that happen."

Pete Chapman (38:35):

And so, we think like we're taking a completely separate approach, different approach than some of the big guys are taking when it comes to access to client data, and we think that we'll win over the long run. We're not naive enough to think that we can go build every mouse trap that our clients need to live a healthier financial life.

Pete Chapman (38:54):

We would much rather focus on the things we're great at which are banking, managing risk, listening to our clients, et cetera, and we would rather go engage or build pipes out into the third-party ecosystem – safe, secure pipes out the third-party ecosystem, allow our clients to do everything that they need to do. So, strategically that is our approach to this thing and we just see this as kind of like a new "battlefield" kind of on that front.

Jim Marous (39:20):

So, as a last question, and it's funny because we are so used to saying in the next three to five years, I'm going, no, I can't even go in that direction, I don't know if I'll be here in three to five years. So, in the evolution of this product, this move to what you've done here, how do you see this playing out? What's your thought on what can come from this?

Jim Marous (39:43):

Do you see your business bank and customers actually giving you access to data outside of your organization so that you can help them come up with even better answers that you couldn't get just with the limited data you have access to?

Jim Marous (39:59):

It sounds to me that this is almost a trust enhancement feature that says, “I'm going to give you your data back in a way you can use it with these new tools, how you use it is up to you,” but we want to enable you in a way that makes it so that we're not the keeper of the keys, we're simply the doorman, we're making it so that we can leave that door open and be there when you need us.

Pete Chapman (40:25):

I think I like what you said about enablement there Jim, I think that's how we look at this thing. And going back to where we started where I answered your first question around how did we get to this point and I talked about, hey, we're trying to leverage AI for internal efficiencies.

Pete Chapman (40:41):

I know our clients are going through that right now. They're going through that exact same journey, and their banking data is a critical component to that enablement. And so, I would've loved it if when we started going down this path, I had my core strategic vendors kind of open up access via an MCP to all of that data to me, it would've been way more powerful.

Pete Chapman (41:08):

So, we see this really as enablement, and who knows where it's going to go. But I would think that you've got the ability for a CFO to stop spending hours on end prepping for investor relations or board calls or whatever, and they've got some template in Claude or ChatGPT that simply sucks in this data and does it for them, and allows them to stop focusing on that and start focusing on something that's a little bit more important. But anyways, that's how I would look at it.

Jim Marous (41:35):

You definitely become an employee of the business and help them do their business better. Again, I got a whole lot of problems with the great looking tools that you can get on your mobile platform or your online platform or what this all means. It's a whole different play when you're starting to working with LLMs and the applicability of being able to build what you want as a business from scratch, but with your bank as a partner.

Jim Marous (42:01):

Chris, how do you see this playing out? What's your vision of what could truly happen here?

Chris Tremont (42:09):

And I think my thoughts are akin to what Pete was saying. As a 1.5 or $2 billion digital bank focused on serving SMBs in the middle market, certainly our approach is maybe a little bit different than the trillion-dollar banks thinking about open banking and data sharing, especially with their consumers.

Chris Tremont (42:30):

Our approach has been or our approach is sort of what Pete said is like we want to be a conduit or a tool to allowing small businesses to flourish and spend less time, I don't know, in their bank account trying to figure out what the hell's going on inside their operations, and more time running their company.

Chris Tremont (42:52):

And so, I think this is the start of that, of AI is changing the way all of us sort of interact both in our personal lives and the way we work. And so, this is us finding a way to participate in that, we are in the very early innings. Where it goes Jim, I don't know, there's like a bold prediction that you were asking us to make in October of 2025.

Chris Tremont (43:21):

But I mean, I do think it could be the start of I don't know, like less involvement with your bank maybe inside of the traditional commercial digital banking platform and more just access to your data using tools that they use today for managing other parts of their business.

Chris Tremont (43:42):

And it just starts to embed in their lives, sort of how we've built an embedded finance platform that's very complicated with some large fintechs. We didn't even talk about that here today, that's like a whole separate business.

Chris Tremont (43:57):

This is like bringing some of that concept or that applicability to smaller businesses that say, I'm not trying to stand up running a consumer checking account platform, I'm not trying to be a bank, but I'm trying to get at my data a little bit better to help me maybe make better business, acquisition decisions help the bank understand my needs for getting access to capital and grow.

Chris Tremont (44:18):

So, I think there's a lot of ways that as Pete was saying, we don't know yet, but as we start to listen to how our customers use it and even bring some more external data into the Grasshopper operation, it's going to help us get a lot smarter in serving our clients.

Jim Marous (44:34):

Well, and at the end of the day, I'm going to have Leah make sure that in the next couple days, she sets up a meeting with us six months from now. Because I think that the reality, the real excitement here is what your customers are going to be doing with the tools you're providing them and how you're going to make it so that you feed those experiences back to your customers.

Jim Marous (44:57):

So, it becomes a massive sandbox where your customers are doing a lot of the innovation on your behalf of the tools you provide them, and that you're going to feed this back to other customers saying, this person used this for payroll, this person used it for cashflow management, for loan operations, and that you integrate some of that within your organization saying, we're going to provide you … that's where the answers start coming.

Jim Marous (45:24):

Where, by the way, based on what you've been working on, it looks like you could use this and all of a sudden, you're well on the way to what we kind of envision agentic AI without the risk, where you're doing it based on what the customer's doing and having them provide the final answer as to what they want to do. But again, that's the ultimate benefit down the road, I think.

Jim Marous (45:48):

So, exciting to talk to you gentlemen, it took a while getting us together. You guys have been on all kinds of podcasts and meetings and write-ups on American Banker and everywhere else so I'm excited to see what happens.

[Music Playing]

Jim Marous (46:03):

As I looked at both your backgrounds, it kind of was the culmination of what you would think would come together where you're saying, you know what, I've been here, done this, done this, never been pleased with the outcome from the customer standpoint, you've kind of flipped the table. So, I really appreciate your time on the show.

Jim Marous (46:22):

Also, for anybody listening today, please download some of our other podcasts where we've talked about the power of generative AI. We have quite a few of them. It's obviously a hot topic, but this is something new that we had not seen before and wish you both a lot of luck.

Chris Tremont (46:38):

Thank you, Jim.

Pete Chapman (46:39):

Thanks, Jim.

Jim Marous (46:39):

Thanks for listening to Banking Transformed, the winner of three international awards for podcast excellence. If you enjoy what we're doing, we would really enjoy a positive review. Also, check out my recent articles in The Financial Brand, the research we're doing for the Digital Banking Report.

Jim Marous (46:59):

This has been a production of Evergreen Podcasts. A special thank you to our senior producer, Leah Haslage; audio engineer, Chris Fafalios, and video producer, Will Pritts.