From Proof of Value to Game-Changer: An Inside Look at AI in Credit Unions
“AI is not coming. It’s here,” says Kirk Drake. And credit unions, with their rich troves of structured data, are in a prime position to benefit.
Yet many are still in wait-and-see mode. This episode aims to change that. Drake and Jeter share how TruStone moved from strategy to execution in under four months and why other credit unions need to act now.
TruStone’s AI Success Story
Gary Jeter, CTO at TruStone, launched an AI initiative focused on a familiar pain point: outdated, scattered policies and procedures. Partnering with Senseo AI, his team created “True Assist,” a generative AI-powered tool that:
- Replaced the outdated intranet system
- Centralized 600+ policies into one searchable interface
- Improved staff productivity and compliance tracking
What started as a proof of value with 70 documents and 10 users grew into a full rollout across all 19 Minnesota branches in just four months. Response quality jumped from 23% to over 85%.
Strategy, Not Tech, Drives Results
Jeter credits early leadership alignment and strong business sponsorship as key to success. “This wasn’t an IT project—it was a strategy initiative,” he said.
He also emphasized the importance of rapid iteration: “We didn’t wait for perfection. The AI helped us identify broken processes faster than we ever could manually.”
Expanding AI’s Reach
TruStone isn’t stopping with internal tools. The team is:
- Deploying predictive models for member attrition and next-best product targeting
- Developing LLM-powered chatbots to replace static, rule-based systems
- Exploring agentic AI to handle full workflows without human input
Meanwhile, Drake’s Credit Union 2.0 is helping credit unions analyze core provider data, automate loan participation strategies, and benchmark FinTech performance using AI.
Where Credit Unions Should Start
Both leaders agree: don’t wait for the perfect solution. Start small, focus on real problems, and learn as you go. Their advice:
- Choose a clear use case with measurable value (like policies or attrition modeling)
- Educate your executive team to drive buy-in
- Assign a business sponsor to lead implementation, not just IT
And above all, create space for experimentation. Kirk recommends starting a regular “AI jam session” where teams explore quick wins together.
Stream this Episode to Learn More
Highlights from the episode:
- What it takes to launch a successful AI pilot in just 4 months
- How to measure AI effectiveness—and improve it fast
- Why AI adoption needs culture change as much as technology
Kirk Drake and Gary Jeter are not affiliated with or endorsed by ACT Advisors, LLC.
Episode Links
Audio Transcription (pulled from the podcast)
Doug English: [00:00:00]
Kirk Drake, welcome back to CU on the show. Uh, Kirk from Credit Union 2.0 and Gary Jeter, welcome from True Stone Financial. I’m glad to have you both join me today. I’m looking forward to our conversation on ai.
So, uh, if you would both, uh, tell me how you got started working in the credit Union movement, and then what kind of work you’re doing with credit unions today.
Kirk Drake: Sure. Uh, so I started a, a high school bank back in 1993. Um, and when I got to college. Suddenly got a girlfriend and needed to get a job. I applied for a bank, bunch of bank jobs, nobody called me back.
Kind of panicked a bit and realized to put the wrong phone number on my resume. Um, at which point I applied for a bunch of credit union jobs, uh, and agriculture Credit Union hired me as a, as a teller. I got fired from being a teller about three days in, and they put me in the IT department,
like I cashed some check that I wasn’t supposed to have. [00:01:00] Uh, and uh, and that was sort of the, the rest is history. I’ve spent the last 30 years since then, uh, in, on the credit union side of things. Um, and, uh, I’m pretty focused today, I would say on anybody who knows me, knows I’m not a very focused person in the first place.
But the, the three deepest areas right now that I’m intentionally working on, uh, one is and have been for the last five, six years on how AI impacts. Um, credit unions, uh, in every aspect, staffing, technology, et cetera. One is, um, the balance sheet management side of, of it, specifically with AI and how to buy, sell, trade, hedge, swap loan participations in an automated way that rebalances risk within the industry, um, and eliminates brokers, um, kind of in that process.
And then one is, um, highly focused in that same vertical, one layer deeper, deeper of the use of, um, shorter term sale [00:02:00] leaseback transactions for credit unions to unlock net hidden net worth within their, uh, real estate assets. So there you go. There’s my focus right now.
Doug English: I see. So some light, fluffy subjects, Kirk.
Excellent. Yeah. Good, good. Always, uh, always good to count on you for those light fluffy subjects. Awesome. Well, welcome back. Glad to have you. Over to you, Gary.
Gary Jeter: Yeah. Hey, uh, uh, so that’s fascinating, Kirk, uh, where a career progression gets fired from a teller and moving into it. That’s the, that’s the first I’ve heard.
Kirk Drake: They didn’t call it firing, they just said I’d be better suited over here.
Gary Jeter: Yeah, it’s, it’s a way into it for sure. And so that’s, that’s me. So I’m, uh, so, um, yeah. I’m Gary Jeter. I’m the Chief Technology Officer here at, uh, true Stone. And, uh, how I actually gotten in credit unions was about five years ago. So primarily I spent most of my time in, uh, corporate America working at companies, very large Fortune 500 companies.
Uh, essentially I was living at, in Boston at the time and, [00:03:00] um, uh, Tru at Firefly at the time, um, which was a credit union that was getting ready for a very big merger with Tru Stone. And so, uh, younger CEO reached out, uh, just a, a great guy across the board, uh, jumped on board, helped with the merger. And so from there it’s really been focus on learning about credit unions.
And now that I’m in the space, I just absolutely love it. And what I love about it is it’s uh, it’s not about shareholder value, it’s about member value. And I’ve worked in a lot of companies and they may say customer focus, but it’s really about shareholder value. And um, you know, I serve some time in the military as well.
And that, that service actually, um, really ties well into credit union. ’cause you really care about the member, you’re really serving them and you’re driving forward. So, absolutely love the credit union space. Been here five years and I hope to be here a lot longer. As far as what I’m working on, I guess, is essentially, um, we’re really focused on true stone, on [00:04:00] looking at our, um, you know, technology modernization, our architecture, um, you know, those type of things.
Then also from a digital transformation, a lot of our processes on bringing in, um, a scaled agile framework and, uh, cultural transformation. And then the last thing is about, uh, fintechs and bringing in a lot of modern technology and artificial intelligence. It’s, it can’t be more exciting time to be involved with financial industries, fintechs and ai.
It’s just, uh, just really, really cool and that’s why I’m focused on it. How do we leverage those, bring member value and bring a team member value and credit union value to, uh, leveraging some of these technologies.
Doug English: Awesome. Well, thank you both. So what I, what I would love, uh, and I’m gonna kind of, you know, knowing Kirk, I’m gonna kind of put it in your court and then I’ll see if I have anything to offer at all.
Uh, two things that I really wanna cover, wanna know what, what Gary’s up to, uh, and what, uh, sort of, uh, in place uses of AI are already there. And let’s make it nice and [00:05:00] tangible and talk about who the various providers are so our listeners can, uh, actual, you know, actually take something and, and put it into action from that.
And then back up if you would, and talk bigger about AI and strategy and, uh, and, and like suggested first steps for credit unions that aren’t yet there yet. ’cause we know that’s most of ’em, right? Most of ’em are not there yet. And the outcome I really wanna achieve is I wanna help them to get involved and to lean forward into ai.
Kirk Drake: Yeah, lemme set it up slightly, which is, um, there is no industry that has more structured data than ours. Um, everything is a number at the end of the day, um, in a, in a credit union and how it works. And AI thrives on structured data. Now, it’ll also work, you know, on unstructured data these days. But at its core, to me, the, the reason why it is when I look for use cases of how it’s going to be disruptive, um, [00:06:00] I find 90% of them are in the financial services industry.
And when I, I, I went to an AI ed tech conference this week with my daughter ’cause I decided I would teach the faculty some stuff about AI so that they don’t just pretend like calculators aren’t always gonna be here. Um, and uh, and I went to that and it’s, it’s, there are two use cases that a hundred people are trying to solve versus the credit union industry where you’ll see a hundred use cases that a thousand people are trying to solve.
So set it up that way. Gary, I, I think you were one of the first people that started working on this and kind of jumped in early with Senso and some of what you’re doing. So I’d love to have you kind of explain what you guys been doing with that and some of the success in there. Yeah, hang on just Gary, just gimme one second.
I wanna just back up one little bit. I’d really love to know the strategic source behind that. Like was there a board strategy? Was the CEO calling out a strategy? Like what got what you’re doing started, where’d it come from?
Yeah. So, uh, like I mentioned, our, [00:07:00] our CEO, he is very forward-leaning when it comes to technology. And so when I jumped on board once we got through the merger, it was really about, you know, how do we prep the, uh, from a strategy perspective, how do we prep the credit union as we move in this, this new age?
Right? And so a lot of that’s technology architecture to be able to plug and play different things into it on that journey and discovering different fintechs and discovering what type of capabilities exist out there. Um, one of the groups I went got involved in was a CU 2.0, uh, with Kirk, um, went out to his VIP, um, um.
Uh, session. It was like a mini conference, if you will. Um, and, uh, that’s an opportunity to meet with a lot of fintechs with not a lot of pressure on sales, just more of a learning type thing. And one of the speakers, uh, was, uh, from Senseo ai and he shared what generative ai. So he gave a generative [00:08:00] AI 1 0 1 class and, you know, we had all heard about chat, GPT.
This was, uh, what was it like February, I believe, uh, chat. GPT was released to the world in November. So it wasn’t long after that. And it just blew my mind. I recognized right away that this was going to completely change the world. We were at a, a really big tipping point. Um, you know, just like when the web was launched and this was something, uh, because of the way technology’s moving and where it’s, we’re not at the bottom of the exponential curve.
We’re moving up. Right? And unless you jump on it right now. You’re not gonna be able to get on it. And so brought that back, uh, talked with, uh, you know, my fellow executives EVPs, uh, and um, you know, basically through an education type process, uh, got people to say, yep, you know what, we’ll do a proof of value with it.
And then it kind of moved from there. So, um, yeah, so strategic imperative, maybe [00:09:00] that was a long way to answer what you’re asking. Uh, but it was really driven from a strategy and then, um, yeah, a learning perspective. So, I mean, my recommendation to anyone out there is go to these FinTech meetups, get involved in these conferences.
I just came back from one. Um, and you just, you learn so much. He and you. It’s, it was fantastic. And so I actually thank Kurt for, uh, getting me involved with this one.
Doug English: So, so go into what you’re doing, Gary, I wanna hear what, what you Oh yeah. Absolute doing today. Absolutely.
Gary Jeter: So, yeah, so, uh, what I’ll tell you, uh, what, what we’ve, uh, what we’ve done with generative ai, I can talk to a little bit of what we’ve done with some, uh, predictive models with machine learning, and I can talk about where we, where we hope to go.
Right? So, uh, I talked about linking up, um, you know, meeting oop from sense of ai, who’s the CEO and founder. Um, and so came back and essentially, um, uh, had a, had a session [00:10:00] where we said, okay, let’s start with a proof of value. So I explained, gave a, uh, you know, this very similar class to my peers and CEO as far as what generative AI is.
And let you know, define a use case around policies and procedures, um, with, uh, with Senseo to say, Hey, uh. Let’s use this to be able to help, uh, solve a pain point that we have within the credit union, which is maintaining our policies and procedures. They were all over the, uh, our intranet. Uh, many of them were outta date, probably most of them were outta date.
There was always a challenge between what our training organization was, was training versus policies and procedures, uh, that were on the intranet. So managers were always being asked by new employees as far as what they, uh, um, you know, how to do, how to do things, um, and also looking up, you know, looking up things while members are there.
Bankers are looking up, okay, if they haven’t done something in a long time, like open up an IRA, um, just, just [00:11:00] really painful, bad member experience. Uh, and, um. Poor productivity and just even from a compliance perspective, wasn’t happening. So, uh, um, got buy-in from the EVPs to do a proof of value. So we did a very small, uh, proof of value with probably 70 policies, uh, 10 folks from operations, retail, um, and, uh.
You know, we created, it’s, uh, agent Fetch is what Senseo calls it. We, we branded it True Assist, but essentially it’s just like chat. GPT uh, pulls off policies and procedures, which were all PDFs and was able to answer questions as far as what are the steps I need to do to open our a Um, when we first did that, um, you know, it was a, uh, the model was a little bit immature.
They’re using the, um, um, uh, chat, G-P-T-A-A-P-I, but as far as reading the data, the data quality or the response quality was about 23%. Uh, however the value was there, there was a lot of opportunity to be able to focus it, so we [00:12:00] decided to move to, uh, what we taught said was a proof of concept, so we expanded it.
From, from, uh, 73 policies, we put 600 policies in there. We pulled in about, uh, I wanna say 50, uh, team members within retail. And the way the model gets better is a thumbs up or thumbs down. Uh, so essentially that proof of concept went through, uh, response quality and proves significantly we got to about 75% response quality.
And the other thing we discovered was we didn’t wait till our policies were updated till, uh, to reflect accuracy. We just put them in there and what we found was. We were able to identify through the response quality, what policies were out of date, what policies were inaccurate. So it wasn’t the large language model or uh, true assist that was necessarily giving bad responses.
It was our policies that were giving bad responses. And so that created a whole, an operations department that said, you know what? We [00:13:00] didn’t have this centralized before. We’re gonna pull it all, all under one, uh, uh, team member that essentially will be the person that serves as the master orchestrator of our policies.
So as soon as he got a thumb down response, he would reach out to the process, uh, owner, the manager, update the policy, drop it in there, replace the old one, and within near real time, five minutes, all of a sudden the right procedure was in there. Proof of concept was just immensely. Uh. Um, successful. And so we, we moved to what we, we called a pilot.
So in the pilot we expanded it to all Minnesota branches. We’re talking 19. Uh, we had, uh, at that point we added some more policies. Like, uh, we discovered like a lot of the questions that were coming in where essentially HR policies, you know, like what’s the maternity leave policy? So we expanded pretty much everything, uh, within our, within our organization that you would find on the internet.
And, uh, um, again, response quality stepped up. We created training videos, [00:14:00] hype videos, uh, you know, integrated into our training material. Um, really refined that operations as far as how quickly can we identify a thumbs down or a, uh, I cannot, you know, I can’t respond to you like, uh, and you know. Smooth that out as far as being able to quickly update policies.
And, uh, within a month we then did a full production rollout. So everyone, uh, everyone now uses it. And we don’t no longer have any policy and procedures on our, um, intranet that’s all gone. Um, ’cause you can’t keep ’em up to date. They’re only with in Truist and it’s used every single day. And it’s just the, the value it’s given, it’s been a real game changer.
It’s, it’s been huge. Yeah. And I think, go ahead Kirk.
Kirk Drake: I was gonna say that, that that continuous lifecycle piece of it, I think is the magic, right? And then you, when you start thinking about what sense is doing all the places that that continuous improvement process can be made to work way more [00:15:00] efficiently.
Um. The, the website, you know, most credit union’s websites are actually pretty disjointed between what members are asking and searching for versus what they’re finding. And I’m not talking like chatbots. I’m talking like, Google search. How do I do this thing? How do I, you know, most people want to figure it out before they call the call center, right?
And so being able to optimize that and make sure that they’re seeing the real information in there is a, is a, a big game changer. Um, same thing on, uh, financial analysis, uh, end of month exception reporting and, and accounting, right? Like all of these sorts of things really thrive on the speed of business.
And to me that’s a, a perfect example where senso, the more and more use cases it can tackle to solve those speed of business type problems. Um, a, the more value it is, the more d differentiated is and the more that credit unit is staying on the. The, the edge of the super competitive, you know, [00:16:00] reacting in the marketplace, you know, side of things.
Doug English: Yeah. You went from 23% of what I interpreted it as accuracy to 75% accuracy. Right. And that’s the, that was the what, what I would, what I would term, uh, that you created a single source of knowledge for true stone’s, processes and procedures. Uh, and then you fed it more and more data, and as you fed it more data, it got increasingly accurate.
Uh, and now you have it as a fully, uh, scale, uh, uh, model. Are you, are you finding the accuracy now to be near a hundred percent?
Gary Jeter: Yeah. Well, uh, I would say mostly, and where I’d say the, uh, so it’s, it’s response quality, uh, versus accuracy. Um, and that’s really measured through a thumbs up, thumbs down, right?
So, uh, from a, um. So it was seven. The response quality before we moved into the pilot was about 73%. Now we’re [00:17:00] sitting probably 85, 90%. Um, and it’s, it’s, it’s because the, the data that the model is pulling might be pulling from the wrong source and giving, um, giving information that either can’t respond or, or doesn’t really answer the question.
Um, so I would put it along those lines versus quality, uh, or, or accuracy. Um, so it’s, yeah, for the most part it’s really, really solid.
Kirk Drake: And I, and I think it’s important to think about like, traditionally, how would we do this, you know, and, and. And the idea that this is a set in stone time period, right?
Because the reality is the business rules, the regulation, the technology is constantly changing everything. Every, if, if you’re core, your mobile banking, you’re, you know, whatever it is, it’s all in an agile mode. Then how about you? I, it’s funny, I, I log into, I pull up my phone every morning, I hit update and there’s always 18 new things, right?
Like every day there’s 18 new updates to the apps, right? Like, and so you just think about that process going on. [00:18:00] And the old way we would do it would be, we’re gonna send it to our training department, who’s right to read the procedures, document it, then we’re gonna run some classes and we’re gonna educate people, and then we’re gonna, you know, upgrade the software and it will be slightly different than what we thought and, and kind of we’re in this slow, methodical, you know, kind of approach.
And that was like best practices. And I think in the modern world of the, of the tech world, I. Every big tech company I’ve, I’ve worked at, like, they don’t, there’s no training, like there’s no training department. It’s expected that you figure this out kind of as you go. And it’s a clunky and it’s painful, but that’s how it works.
Um, and in this world, you know, if you start to rethink how we do it, which is okay, now the new procedure, my subject matter expert, that software updates, they just record a video of them figuring it out, right? Like, not even the right way to do it. Just them, the, the, and they narrate the process of them figuring out how to do the new thing.
And then that gets ingested into the ai that then the first person who needs to do that [00:19:00] thing watches that video, and then they give it thumbs up, thumbs down, and it gets a little refinement within about three to four cycles that that learning process gets optimized way faster. And it’s not like the fifth person who, who asks the question ever sees the first four bad answers, right?
It’s, it’s, and so. When you think about the most common problems that occurred, get the fastest revisions and get to the most number of people quickly, which then keeps the whole system operating way faster, way more agile than whatever you would’ve done historically. Right.
Gary Jeter: Which is really, really interesting.
’cause one of the things that we got feedback initially, especially from some of the senior management EBPs, is like, how can you launch this when it’s 80% response quality? Well, the, the answer back or the question back was, well. What’s the following? The process quality today. Right? Because the process [00:20:00] quality today is either people can’t find the pro the procedure on the internet.
They have to go ask their manager who, you know, if that procedure changed, they don’t know about it because they didn’t read the email that went out last week. Uh, the training department may be going off it. So if you look at what the quality of, of how we were consistently following processes don’t have data, but I can guarantee you it’s a lot less than non, you know, 80%, 90%.
Kirk Drake: Yeah. Yeah. And, and you don’t even really know how to evaluate if Teller A does it a way that’s, that’s five minutes faster than teller B, like how long does it take that data to get back to the training department to optimize it? And then how do you convince the other a hundred tellers to do it as person A, right?
Like, and so we spend all this time trying to teach our brains to memorize these stuff. And I, and in some ways. I feel like the challenge of, of, of Gen X and, and Gen Z is, is switching to a, a mode of [00:21:00] accepting that we don’t actually have to memorize how to do anything. Right? Yeah. What we have to do is know how to find the best way to do it right now, and it’s gonna be different tomorrow, so there’s really no point in even memorizing it.
Right?
Gary Jeter: Yeah. And the, it literally, it’s five minutes updated and everyone’s using the same information. Right. Once you find that best practice, um, and, uh, it’s, it is, uh, it is so cool. Yeah.
Kirk Drake: So I think that’s a, a great example of like how to get started. One other way that I would suggest that we’ve had success in building this in other credit unions and in CU two itself was just sitting down with our team and saying, look, here’s what I know.
I know if you aren’t embracing ai, you’re probably gonna be irrelevant long term. And I, and I hate to say that, but that is, it’s like pretending like the internet’s not gonna happen or calculators aren’t gonna happen. Like eventually there will become a point in time where your skills will not be relevant to the problems that need to be solved in today’s [00:22:00] world.
And maybe you’re three years away from retirement and you’re willing to say like, don’t care, don’t need to learn this. Um, but for the most of us, we’re not that close to retirement. We’re gonna need to have some of these skills. So if that’s true, then let’s start a, a weekly or biweekly session. And I and I coach credit unions and their teams through this all the time.
And it’s not a negative session. It’s simply come to this meeting with all the things that are, take less than 15 minutes that you’re doing right now, quick inventory. And then as a group, we’re gonna spend an hour, hour and a half figuring out how to use AI and if a, figuring out a can we use AI to make this thing faster, better, cheaper?
B, um, where are the limitations and problems in it? And c, if it works, how do we begin to roll this out to the whole organization, right? Um, and teach everybody else. And it’s an opt-in meeting. There’s, if you don’t want to come, fine, don’t come. But if you, if you want to be part of this, here’s how we’re gonna do it.
And we guys get on this call, [00:23:00] we celebrate the wins that they figured out from the prior. Time period, we innovate and try new things and within about three or four meetings, it blows people’s mind what they’re able to do and all the different use cases, whether it’s writing job descriptions, whether it’s analyzing.
Um, I had one where I uploaded five different bank statements and had it figure out my recurring transactions across all of that as opposed to just my one credit unit account. And then be able to connect the dots between interesting ways of how money was moving in the, in those different, um, paradigms.
I’ve had it, you know, all all sorts of really interesting ways to use it to analyze problems that. You just wouldn’t have had the time to tackle before or, um, you know, wouldn’t have, have seen it. We, uh, in, in various ways. And so that’s another way to get started that doesn’t require really buying anything more than o open AI and chat GPT.
Um, you know, you can, in theory you could [00:24:00] use copilot. I find it to be significantly inferior compared to the, the current models, um, and the cutting edge side of, of it. But it’s the safe buy. No one’s gonna get fired for buying Microsoft, even though it, it doesn’t really, um, I haven’t found that it creates the unique IP that would be useful, um, within the organization.
Um, and so I’ve, I’ve done that now in probably 20 different organizations and had really good success in about six to 12 weeks of getting 10, 20, 30 people really embracing this and seeing the light. And then once they’re off. They’re off, they start connecting the dots every place.
Doug English: What level is that? Who, what are, are they, uh, in mid-level managers or SVPs?
What are what level?
Kirk Drake: Uh, I mean, in, in C two’s case, we went from frontline admin staff all the way to advanced sales technology. We’re using one use case we’re using now. We, we’ve, um, taken every FinTech that we’ve ever worked with their client list, uploaded it and mapped it to key, [00:25:00] um, key things. The FinTech impacts in their 5,300 data.
So like in the case of senso, we’ve ma we’ve said, Hey, they’ve got these 20 clients. Let’s look at how senso impacts opex over time, right? And then create a peer group of all the credit unions using Senseo compared to everybody else and says, we should be able to see that this group of credit unions operating expenses should be slightly lower than peer, um, if this thing is producing this.
So, so we did that technical mapping, and then we take all that data thrown into chat, GPT. And it writes a FinTech analyst analysis report. It tells us, um, we have one right now running that. Does, does Simar or Fiserv produce a better result financially over time? Hmm. Does Alchemy or Q2? No one’s gonna like when we publish this.
Doug English: Yeah, I bet someone’s not gonna like it.
Kirk Drake: I can tell you right now, early sneak peak, Fiserv, credit unions generate two times the members of Jack Henry Credit [00:26:00] Unions. Don’t know why, why you heard it here, right? Using deep research in chat GPT, we just dump all the data and say, tell me what the insights are out of it.
Right now we spend a lot more time customizing the prompts and then writing the reports and generating all that. But that’s a great example where we just never would’ve had the time to start to get to that level without chat, GPT and ai. And where it’s, it’s helps you too, is uh, every person in the organization went from bottleneck and overwhelmed to a couple hours a day of free time.
Um, as we’ve built these tools and components into the whole system, oh’s huge, hugely impactful with, um, I’ve got another company that we’ve done this with. We’ve plugged it into their ticketing system and analyzed the ticket, suggests a response. To the customer, to the end Techno technologist, um, combines all the policies and procedures and it’s being used by the frontline staff and the, and the frontline staff are part of this biweekly [00:27:00] call.
And the CEO is it, and sometimes I’ll do coaching one-on-one with the CEO separately from that, but, uh, all levels in the organization can get there and can get the value, um, in it.
Doug English: Hmm. Really, really, really interesting. And, and when credit, when you’re doing this with credit unions, uh, and, and, uh, what levels of folks are in the meeting that you described with credit unions?
Or, or does it vary?
Kirk Drake: It varies. It’s whatever they want, but, but I’ve done it with board members too. So I’ve, I’ve got some credit unions, um, where, uh, a large top five credit union, I had me train their whole board and their board uses it to analyze their board package. They each have their own custom prompts that they look at the board package with each month, and then it, it suggests questions for them in the board meeting and I mean, really leveled up the entire discourse about what they’re working on.
Doug English: Wow. AI interpreting your board packet for you, right? Helping you understand and build smart questions. That is brilliant. Wow.
Kirk Drake: My, my [00:28:00] all time favorite use case, which has nothing to do with credit unions, is my, um, surprise date night generator. Um, so I took a whole bunch of information about my wife. What she likes, what music she likes, what food she likes, what things we’ve done memory wise, like in our 20 year marriage that she’s really enjoyed, put ’em all in there.
And then I have chat, GPT suggest 20 surprise date ideas. And then I pick one and then it’s got a set set of prompts that plan out the date as if I spent a hundred hours planning the date, takes about four minutes, and then I execute the date based on that. And she’s like, oh my God, this date was amazing.
You put a lot of thought into it. It’s like, no, I really didn’t, but I did once when I prompted, I did once dbt. Yeah, exactly. That’s so it suggested a custom playlist. I mean, it, it like, it, it like was really over the top in kind of, uh, in those kind of things. So to me that’s where those group, um, ’cause, ’cause at the end of the day, if your employees are happy and productive, whether it’s specifically related to credit union stuff or not, all of that stuff ends [00:29:00] up producing a much better workforce, much better culture, much speedier time to market, you know, in all of these different ways.
Doug English: Your members are happier ’cause they’re getting accurate, they’re filling out the right forms, they’re doing the right processes. They’re not having to do things twice, they’re getting the right information. I mean, that, that makes, that makes a lot of sense. I think it’s interesting that the process that you chose to start with was a, an internal process that, uh, you know, still goes through the employees of True Stone to, uh, impact the members.
Uh, are, have you, are there other things that you’re, uh, already involved with, with AI or what’s next if you’re not,
Gary Jeter: I. No, we’re, uh, we are right now, so today, right? And most people have chatbots, right? But the, their chatbots are, uh, more of a rule based. And our, our chat bot too, for our omni omnichannel solution is, uh, rule-based, right?
So you get intense, I think, I believe there is a little bit of machine learning involved, but it’s primarily rule-based, right? [00:30:00] So, um, which isn’t a large language model, right? It can’t understand context, it can’t understand what, what questions were asked previously. So we’re in the process, uh, working with a partner to say, okay, let’s, let’s use a large language model to essentially, uh, bring this member facing, uh, with our chatbots.
So that’s in, that’s what we’re working. We’re hoping to have that out, uh, um, this year. We purposely chose, started internally because, uh, of, you know, LLM hallucinations. Um, but you know, AI is a very broad scope, right? You have these chat bots, which are more rule-based. Then you have machine learning. So we just launched a machine learning model, uh, which essentially, um, about, um, member attrition, whose likelihood to a trip, um, used that model, trained it on our data.
So we worked with a, uh, a FinTech called Trellis, which I think most private people probably know, took their model, trained it on our data, and now that information is being, uh, fed to, um, marketing and [00:31:00] retail to identify those, you know, the top 10% of members that are, are TRI and how do we go in there and move, um, you know, retain those members.
Um, and the next model we’re, we’re working through, again, these are machine learning models. They’re not large language models, uh, but next best product. We’re almost getting ready to launch that in June, which will then turn around and tie very nicely with the attrition. Michael models. Mm-hmm. Go ahead.
Kirk Drake: I was gonna say, uh, I love those examples.
We’ve got two that I would share that we’re working on. Um, one is, um, it, it looks at the loans that you want to buy or sell right. From a participation perspective and figures out what credit unions in the marketplace would benefit or have loans to sell for you based on your specific balance sheet today.
Right. So, because sometimes, uh, the loans are worth more to, to, so if you’re selling a pool of auto loans, let’s say it’s a 7% yield. Those are worth more to a credit union in [00:32:00] North Carolina because they’re getting auto loans at five, 5.5% than they are to a credit union in Oregon because they’re booking loans at six point half, 7%.
Right. And so when you, when you have those arbitrage pieces where different markets are able to get different rates and, and assets from their members, you start having those opportunities. So it figures that out. And then the next one, um, we’re just starting this as we’re, we’ve uploaded about 10 different credit union balance sheets, and it comes back and suggests, um, strategies and matchmaking around buy, sell, hedge, um, and swaps and creates, um, uh, synthetic swaps between credit unions to offset their various risks, um, from a balance sheet perspective behind the scenes.
Um, and it can do that matchmaking and figure out where that exists within the industry.
Gary Jeter: You know, um, just going back, uh, you know, which Kurt was ta talking, you know, within, you know, how do you, how do you get started? How do you, how, how can you begin if you, if you haven’t even started yet, um, you know, during the, uh, for [00:33:00] generative ai, and one of the things we did was really started with a policy and that policy wasn’t so restrictive.
There wasn’t a fear base that says, oh my gosh, if I lose, you know, member information. It’s more of, okay, we’re gonna open up chat, GPT, we’re gonna open up, uh, Gemini to, uh, team members. We have a, a, a data loss prevention module, and we created a policy as far as what you’re responsible for. You’re responsible, make sure that the responses are accurate and said, Hey, go use this.
Right? Go use this in your day-to-day work, whether it’s email, uh, um, you know, writing, uh, you know, social media posts, but, you know, tie it into your day-to-day work. And that has been able to and through a lot of education. So I do a virtual co, uh, uh, A-A-C-T-O virtual coffee where I, I talk about, um, you know, artificial intelligence and data and things like that.
Topics vary. Um, um, so essentially with that, that’s, that’s bubbling up [00:34:00] use cases or like our compliance, you know, hey, this is some, something that’s really gonna be a force multiplier for us. So we’re finding use cases, uh, within the organization that then, then we’re going out with fintechs and we’re bringing those things forward to, to do a, uh, do a matching to bring in.
So. That’s, uh, uh, that’s a good place to get started, starting with that policy. Um, you know, and allowing people to use it, allowing people to be, uh, our team members and, and be safe.
Doug English: So the, the, the fin text or the AI companies that we’ve mentioned so far, uh, uh, Senseo AI is the one that you have used, uh, for the, uh, internal single source of knowledge that you talked about, Gary.
Mm-hmm. Uh, and then Kirk, uh, your company, credit Union 2.0 is, is is involved in, in a variety of different, uh, uh, AI applications that I don’t know, that I, uh, um, could do a good job of, of saying to credit unions, [00:35:00] here’s what Kirk does. ’cause I think you do a lot of stuff.
Kirk Drake: Yeah. I, it would be d difficult, I mean, I think at the end of the day, um, when I was researching the AI book, uh, in 19, in 2019, 2018 timeframe,
99% of the fintechs I talked to were using some form of machine learning or ai, um, in what they were doing, and no credit unions were at, at that time for the most part. Mm-hmm. Right. Mm-hmm. And so, uh, that was super eye-opening for me. Um, I would say, um, most of what we’re use working with today is using it in some form or fashion.
Um, if you want a comprehensive list, I can follow up with that, but it’s, it’s, it would be too many to state and we’d waste the rest of this time.
Doug English: Yeah. Well, my, my, my objective is to make it actionable for a lot listeners that are not yet there or just trying to, they’re, they’re gonna have their, uh, strategic planning session, uh, and they’re thinking about, okay, we need to get on board.
How do I get started? And my, my takeaway from [00:36:00] Gary is, uh, consider, uh, involvement in some of the, uh, the ai, uh, initiatives of the industry, uh, Kirk’s with Credit Union 2.0. I’m sure that, uh, the Cusso conference that’s, uh, very soon, or I guess by the time this airs may have been just recently, uh, is gonna have, uh, some more, uh, topics like that to help you get some ideas around.
What do you wanna go after? Uh, first. So true stone went after the single source of knowledge internally. I’ve heard, uh, I had, uh, Mitch, uh, Rutledge on the, on the call just the other day, and we talked about using it for marketing, uh, purposes and, and outreach and the next best product that you brought up, Gary, as your next step, uh, as what we talked about and, uh, and that, uh, podcast.
So when you think about the industry as a whole, Kirk, what else, uh, comes to mind as far as, uh, what you’re seeing, maybe more credit unions adapting? Is there any kind of, what you’re seeing the most of [00:37:00] happening?
Kirk Drake: Uh, I think the most is, there’s two big things. One, it’s they deploy, um. Co-pilot with Microsoft.
Uh, and, and then they check the box and go, I’m doing AI stuff back in the sound. All set. Yeah. And, and like that’s just the, it’s the silliest thing. Um, but I see that probably nine out of 10 times, uh, out of, out of folks out there and then one out of 10 they’re intentionally picking some pilots and trying things kinda like Gary is, and, and diving deep on, um, usually around knowledge management, um, piece of it.
And, and I, while I think that is a phenomenal use case, I think it’s, it, the idea that you’re gonna have one AI is nonsense. Like you’re gonna have 50 or a hundred, you know, of these. Um, they’re gonna be specialty trained in all parts of your organization. And there are gonna be lots of agents for lots of people.
And there’s, you know, this is, this is, it’s like internet. It’s gonna change every single app and every single interaction with technology, even at the end of the day, like. You think I I, when we were [00:38:00] working on See Wallet 10, 12 years ago, no, 15 years ago, um, you know, we walked in and you, you thought about, you know, pulling up your app at a McDonald’s or a, a Subway and configuring your sandwich or your burger, right?
And I, and I even back then, I said, look, at the end of the day, me walking into a McDonald’s and saying, I want a quarter Pounder with cheese. No, no onions is faster than any app will ever be. No matter how well I know that user interface, the human language is the most efficient way to communicate, wants, desires, needs, you know, transactions.
And so now that generative AI is there, when we think about mobile banking, when we think about all those interactions. I don’t think the UI is gonna matter in, in five or 10 years. I think we’re gonna completely shifted to a, a different way of interacting that will get much more simplistic and much more responsive, and much easier for the average consumer than where we live in the world today.
And all the human norms of not being willing to ask a question that make [00:39:00] might make me look stupid, right? Go away when I’m asking, uh, A-A-G-P-T or ai because there’s no fear of looking stupid with the computer yet. Maybe we’ll find some way to inject that. Um, but you know, I, I think we’re gonna see some pretty significant shifts in there.
And at the end of the day, I just think trying these things and being open and being curious about what it can do today. Versus frankly, I’ll log on one week and try something and come back a week later and it works. Right? And, and if you’re not ready to take advantage of the works, right, you’re, you’re missing that opportunity and that speed and that learning and being able to see the fourth, fifth, sixth iteration of something, because it’s rarely, it’s like Gary started with the first proof of value, or, or, you know, was, was not all that much value.
Right. And it’s the third, fourth, fifth iteration that really starts to create the competitive [00:40:00] advantage. And when you’re the credit union that’s coming in, when Gary’s had this working for two and a half years, even if you plug it in on day one, you’re never catching it. Right?
Gary Jeter: Yeah. Let me, let me highlight one thing with that too.
The proof of value. We started at the end of October. We launched the full production at in February, the no next year. Wow. That’s so, so it’s, it is not a long cycle time to be able to, to, to deliver these things. It’s not like a one year CRM implementation. It is like bang, bang, bang. And I, I think Kurt’s directly on, uh, right, when you talk about it’s not one solution like Microsoft copilot and that, that, that covers my generative AI solutions.
What I think where it’s going is these, um, you know, which really makes me excited with the, is, uh, agentic ai where essentially these agents can take the information in, understand the context, and take the action [00:41:00] even without a human involved. And so, you know, I think a great example of that is, so a member calls up, right?
And says, Hey, you know, I see that company X, Y, ZI charge on my credit card, you know, essentially, uh, I don’t recognize it. It’s large charge. What is it about? Right? So you have natural language processing that can take that information in, right? And then essentially recognize the context behind it and can go back into the thing and, and, and say, okay, has she ever dealt with x, y, Z company before?
Has there ever been a charge around it? And then essentially take that and then feed it to different agents, right? So there’s gonna be an agent that looks at essentially things like, uh, um, you know, um, you know, what is the history along there, right? What is the, um, you know, the process essentially to be able to, um, you know, identify if there’s, um, you know.[00:42:00]
Again, kind of histories along those things. So you probably want to cut this part out. But, but essentially what I’m saying is that, that these things will be able to identify, identify like, okay, uh, you need to have an investigation because it is something that’s an outlier. This the, you know, based on her individual spending habit, it’s, uh, this charge has never occurred.
And kick off that, to be able to begin the, the process and provide a response to her that essentially will say things like, Hey, based on your purchase history, it is abnormal. Um, we’ve already started it and we’ve already kicked off that, you know, in a few days you’re going to be, uh, credit the money back onto your account and we’ll keep you updated as where, where, where it goes.
Again, there’s no human involved in that. That’s essentially, uh, the, the, the large language model or the genic AI that’s essentially doing that, um, for the number. So, um, that’s the type of things that I think, uh, will, will be really, really cool, [00:43:00] um, in the future. So it’s really a game changer.
Doug English: So how far away is that?
Like, uh, I mean, uh, agentic AI is something we, I’ve been reading about for months now. Uh, so, but uh, in credit unions, you know, it, it takes a while, right? So, so, you know, uh, we, I haven’t found any credit unions using an AG agentic AI form yet. Kirk have you, or, and then if you haven’t, would you like to make a prediction here for our listeners of when the first AG agentic AI use is going to be in a credit union?
Kirk Drake: I don’t think it’s gonna be long. I’m sure someone’s already probably playing along with it. Um, playing around with it. I, I, so, uh, whether they’re willing to speak about it openly, you know, that might be another year or two. But, uh, there, there’s lots of, I, the good news is, and, and I have a, a warped view of this or a biased view of this, which is the CU two clients tend to be on the cutting, leading edge in general, right?
’cause they wanna look at the [00:44:00] early stuff. They’re not just putting their heads in the sand and they tend to be pretty innovative. So I tends to be, I tend to see a, a wider spectrum of this stuff much earlier. Um, that being said, uh, I would, you know, we were, if every FinTech was using ai, you know, four or five years ago, I think it is definitely on the awareness of, of credit unions at this point.
I’ve had a number of board planning sessions where we go through this and get, get things jump started really, really quick. And their boards love it and get on board. Again, self-selecting, right? The credit unions that are, that want to be relevant in that regard are, are, are the ones doing that. Um, but I do think, um, this will continue to be, uh, a significant change.
I think the, the real question to me is not so much when will we see these things? It’s when will credit unions go, okay, let’s pretend I built the credit union from the ground up using AI today. [00:45:00] And if I, if I were to do that, what would be different about how I approach business strategic planning, my, my data interactions, how my org structure looks, all that kind of stuff.
And once you start to see those patterns, then the question is, how fast do I need to close that gap, right? And how fast do I need to be investing in these other things, uh, to take advantage? ’cause we’ve already seen Rocket Mortgage going, Hey, just upload all your documents and we’ll fill out the forms for you.
Right? Yeah. Well, I don’t know about you. I would love to never have to fill through, you know, to do the database entry work for my credit union as a consumer. Again, how many credit unions are putting that in place? Not very many. Right? Um, you know, the, the, the, so, so we have examples, whether it’s Rocket Mortgage or, uh, a firm of big fintechs that are unicorns that are really dominant in their use of AI in interesting ways.
Um, and our, basically their business model is built. On [00:46:00] the success of this long term, and I haven’t seen any credit unions go, you know what, AI first, right?
Gary Jeter: Yeah. You know, I think it’s, uh, I, I, to, to go back in history as an analogy, right? So when you think of the automobile and Henry Ford came out with a, you know, the production line, he re redesigned the factory, right?
To be around that process, right? So if you think about ai, right? How do you redesign the credit union, the processes around that new capability, right? Yeah. And I think that’s, that’s the game changer and that’s the type of lens that, uh, you know, credit unions need to look at, uh, to be able to. Stay relevant, right?
Doug English: Yeah, yeah. I, I, uh, uh, and months ago in November, I, uh, published an interview with Nimbus, uh, N-Y-M-B-U-S, and they are a, uh, a digital, uh, brand for credit unions that, you know, you just plug into it and you’ve [00:47:00] got your fully, uh, operational, uh, digital, uh, digital system. And it makes me wonder if, uh, if the AI native, uh, system is another version of that, or maybe that’s what Nimbus becomes is where, uh, it it is your, you know, immediate to market, uh, AI centric organization that you plug into.
And then you, you, uh, slowly bring the mothership along at the pace that you can scale to,
Kirk Drake: well, especially ’cause you, we definitely are in the classic. Innovator’s dilemma where everybody goes, okay, everybody’s asking me about ai, so I’m gonna claim I’m doing ai. Even if I’m doing it half-assed and not very well, I’m gonna claim I’m doing it, versus things like that just got built from the ground up with it, right?
Like it is entirely built on the idea of generative ai. Um, they’re not having to change their language model. They’re not having to change anything else. They’re just building it and, and that any, there is no time in history from a company Innovator’s dilemma perspective, [00:48:00] where the startup from the ground, you know, show me one core that succeeded in online mobile banking that owns that space today.
All of them are independents that were built from the ground up, right? Mm-hmm. You know, every, even at the same time, show me an online banking provider that dominated with mobile. Nope. Right. Mobiles came up and added online banking after the fact, right? So, so each time, because it’s almost impossible to cannibalize yourself quickly enough there, you have these, um, challenges and, and to me, when credit union’s making their tech bet of going, is Fiserv gonna be really, really good at ai?
Well, look, they’re gonna have ai, don’t, don’t get me wrong, are they gonna be completely dominant and and dependent on that thing working from the ground up? Probably not. Right? They have enough legacy business and payments and everything else. They don’t need it to be, Oop, needs it to work. It. [00:49:00] His business model and his valuation and his ability to raise money can only work if, if he makes this work, right?
And, um, that needs-based approach to development. That, that’s to me, where I. As an industry, we have to recognize the disruptors that are coming for us are gonna be built on that. And we, we need to intentionally be willing to disrupt ourselves in that process and rethink of that world.
Gary Jeter: That.
Organizational change management is, is probably one of the biggest barriers, not the technology, but being able to, um, you know, culturally embrace that change, lean forward. Um, you know, general Zeki, uh, said that, uh, uh, if you don’t like change, you’re gonna like, uh, you’re not gonna like being relevant or irrelevant anymore, something like that.
But essentially, I think that’s a, that’s a great example where, uh, you gotta embrace change. Gotta have a culture of playing forward and a culture of trying and experiment. Experimenting.
Doug English: Yeah. [00:50:00] So, so I wanna wrap this up by Gary Gidding. I wanna get you to, I. Go back and think through the, the lifecycle of the way you brought AI into the credit union and what you did, in what order you did.
It would, if you could go back and do it again, assuming that someone is listening, uh, and they’re gonna start from scratch, how would you do it differently? What would you do sooner? What would you change? Uh, what, what would you do different than what you did it this first time? Or what would you do the same if you would say, yeah, this worked really great.
I would suggest your listeners do the same kind of thing.
Gary Jeter: Yeah, I would say I brought it in from, uh, an approach of learning from other less successful, uh, initiatives as far as bringing it in. So one, I, I think the key to success was a couple things. One, educating the EVPs, getting them involved upfront as far as what it is, and agreeing to a decision.
Like, okay, let’s experiment. Let’s do a proof of value. So that was, uh, that [00:51:00] was critical because as that proof of value in taking it, you know, one step at a time, that four phase method worked really, really well. And now we’re taking that, uh, as we’re bringing in new technology and new fintechs. The other thing is having, having a sponsor, um, that is in the business, right?
So a technology, you know, coming out of it, right? Being the CTO, especially not being within, um, you know, long credit union history, right? We had a sponsor who was basically, she was the SVP, a regional director of, uh, of retail. So she was responsible for half the branches in Minnesota. Um, she just. Embraced it and she was our executive sponsor that was on the ground that was really driving it.
So that, that is key. It can’t be an IT initiative. It’s really, you gotta have that, that that business sponsor that’s, that’s driving it and embracing it, right. Um, and willing to learn. So those were, those were really successful. So it’s that collaboration.
Kirk Drake: Yeah, I think that’s a [00:52:00] great point. ’cause, uh, one credit and I talked to, had figured out the IT team had gone off and done a bunch of the ML analytics and figured out that basically check holds don’t actually accomplish much of anything and they could never find a, a single instance where it actually, it prevented any particular fraud or loss on behalf of the member.
And they could prove it data wise because they had no sponsor on the branch ops side. No one would implement it. Yeah. Right. Uh, and, and, and so, you know, any, it, it, we like to believe that if we show the data that people are gonna be willing to do it, but there’s an emotional connection to this. And if you don’t have that sponsor that’s bought in that wants to also.
Be along for the ride and the change. No amount of data and proof is gonna actually convince ’em of it.
Gary Jeter: Yeah. And somebody with three cred that can carry that forward. And so that’s, that’s huge. And we’re, we’re, uh, you know, as we bring in these new technologies, we’re doing the same thing.
Doug English: Excellent. Well, Gary Jeter and Kirk Drake, thank you for your work in the Credit Union Movement, [00:53:00] Kirk, uh, particularly your date night innovator, uh, GPT.
Uh, we will be looking for that on the, feel free to email me. I’ll send you the prompts. Yes, yes. Give us the, uh, the prompts in the notes for this podcast so everyone can have a, uh, a better date night, an ever improving date night, uh, as a result of Kirk Drake’s, uh, uh, forward-looking ideas. Thank you so much.
Thanks everybody in the Credit Union movement. We’ll talk again soon. Thanks Doug.
