Compliant Innovation: How Pioneers are Delivering AI in Banking Today

00:03 - Ben Udell (Guest)

The first. What's really exciting to me is all of these companies that are popping up that are using generative AI generally through, for instance, a chat, GPT and to some of that discussion a little bit ago. It means our employees don't need to learn great prompting. It's the agent or the wrapper around the agent that makes it a lot easier.

00:37 - Fred Cadena (Host)

Hello listeners and welcome to Banking on Disruption. I'm Fred Cadena. I could not be more excited about this week's episode. On the podcast, we talk a lot about artificial intelligence and ways it can potentially be leveraged in the banking space, and today we have the pleasure of talking to a real in-the-trenches practitioner who's been bringing AI to his bank over the last year as it's gained prominence, and really excited about digging in with him on use cases and how banks can practically leverage AI to get real results.

01:11

Today my guest is Ben Udall, and if you have not heard of him, you definitely will. He's been named one of BankBeat's rising stars in banking and, in addition to his role at Lake Ridge Bank, where he is the driving force behind future-proofing their operations, he also trains and consults with other institutions to help them transform their operations with AI. This week's episode is really going to be a masterclass in how, if your institution is considering leveraging AI, you can take your first steps and, if you've already decided as an organization, this is something you want, to dip your toes into some practical use cases that you could stand up today. So please get out your pen, get out your notebook, take some notes, sit back and listen in After the interview in our Quick Take segment. Really excited to get Josh and Eric back together for some discussions. And our discussions are going to pick up right where the conversation with Ben left off around authenticity and its role in AI, especially in leveraging AI for external communications, talking to other people, and where you draw the line. Really excited for this episode.

02:32 - Fred Cadena (Host)

Sit back and kick back and get out that notebook because we're about to get started and we're back Really excited today to welcome my friend, Ben Udall, to the podcast Now. With over 25 years of financial experience, Ben is a driving force behind future-proofing Lake Ridge Bank. Ben's a leading expert in leveraging practical applications of generative AI and is never afraid to be disruptive, especially when it allows him to find innovative ways to improve the client and associate experience. Ben was recently named one of BankBeat's rising stars in banking, and for good reason. With that and his experience, Ben's been helping train and consult with other bankers to help them see how AI can transform their organization. In Ben's free time and I can't imagine how he has any he creates content for robotsarepeoplenet. If you haven't checked it out, I highly recommend it. There's a lot of great resources and insights on the website. He also speaks on digital transformation and writes frequently about the financial industry. Ben, really excited to have you on the podcast. Welcome. Thanks, Fred.

03:38 - Ben Udell (Guest)

I'm very excited to be here and obviously love the topic and can't wait to share with your audience. Sounds great.

03:44 - Fred Cadena (Host)

Well, let's just start here. Obviously, you've been thinking and, more importantly, applying digital transformation innovation in banks over the course of your career. I'm assuming 20 years ago AI was not at the top of the charts of what you were thinking about, but maybe it was. You'll surprise me. When did you start seeing that AI was going to become an important transformative technology for banks?

04:13 - Ben Udell (Guest)

alf ago, in the early part of: 2023

04:30

And, as we know, technology certainly in the consumer space for financials has really been picking up over the last five years, and so that was kind of a natural opportunity for me to slide off into more specifically that digital space.

04:44

And that happened to be about the time that ChatGPT was clearly taking off right when you look at the pace of the downloads and the usage. And so part of my origin story is everybody's at the bank is going through this merger and you're putting fires out every day and, in many ways, the last thing anybody wants to talk about is the future and what's innovative and how do we make more changes. And so that was my chance to kind of have my role with what was exciting having the industry start coming together and so once I started to play around with it and dig into it and figure it out, all of a sudden these light bulbs start going off about how you can really apply this generative AI, not to the big picture moonshots, but just to this everyday. My niche is this everyday AI for everyone and you could start to see how that gets applied and it's just really powerful, powerful and transformative.

05:35 - Fred Cadena (Host)

Yeah, that's one of the things I love about it, and I've heard you I think we're both friends with Eric Cook, who's one of the contributors here in our roundtable, and I know that we've participated in a number of his linked banker sessions and that was one of the things that really first kind of attracted me to you and your approach. Was that democratization? And, yeah, some of the stuff that Chase is doing is really cool, right, building their own LLM and training it and setting it up. But not everybody has Jamie Dimon's pockets, right, and so what can we do today to really put this in the hands of people and get real results? And so how has that really influenced your approach to bringing generative AI to your institution and in what way has that maybe evolved over the course of time since you first started doing it to now?

06:30 - Ben Udell (Guest)

Yeah, I think the first piece that is important for people to appreciate and understand and I think what we're doing well is that generative AI, that everyday AI.

06:39

With ChatGPT or if people are using Copilot or Gemini, you're really trying to become more productive during the day.

06:46

So if I can be 10 or 20% more productive, that's an immediate win, either in the volume of work that I'm putting out or, frankly, just a work-life balance that exists. But then with the ability to have the creativity and the skills set, improvement with AI not only can I be 10 or 20% more productive, can I put 10 or 20% better quality work out, and when you start looking at how that is kind of exponentially beneficial not only for one person but across an organization, that becomes really really powerful, and we're all in a very competitive environment, whether you're in banking or otherwise. It's those types of wins that I think are really meaningful. And if people can start to do that today and be ahead of the curve when we continue to grow and get better with AI, you're only going to be ahead of the curve and that's, I think, this really powerful piece that people are not appreciating. They're looking for these moonshots or these gigantic lifts, and you don't need to start there to get people aggravated and go on with it. You really, you really don't.

07:53 - Fred Cadena (Host)

And you probably shouldn't start there, right, you know?

07:57 - Ben Udell (Guest)

I think I was going to say. I think what's interesting is that I think at you know the top of the house for any organization, it's like the executive team is looking for ways to push A out down the ranks, down that org chart, and the reality is, when we start looking at how you can use generative AI, the executive team doesn't know how a frontline person can write better, be more productive, upskill, solve problems. You can't put that out as an organization. The organization has to say green light, use it with these guardrails and give people the flexibility and opportunity to figure it out. And that's, I think, one of these things that I think organizations like us and others are starting to figure out, and that's this democratization piece that you've been talking about.

08:40 - Fred Cadena (Host)

So you're starting to lean into where I wanted to go next, which is let's make this real for our listeners. For a second here, Can you give a specific example of how you've seen generative AI improve the customer experience and or efficiency at your bank?

08:58 - Ben Udell (Guest)

Yeah, let me give you a great example. I'm literally working through right now with associates at the bank. So this is probably common in any organization where you have one group that has to call another group for help and often when they make that call, you get the group taking that call. They're saying why are people calling me for this? They should know this, they should be able to answer that, and that's not a Lakerage or a banking issue. That's a topic across every organization.

09:27

So one of the things that we've done in our world is that's the area that I'm working on is our card services team right? We have these younger bankers. They're newer, they're learning. So either they aren't sure, or they're not using the resources, or they need the support, or, in some cases, we haven't given them the support. So they're calling over this card services team and what I'm working on rolling out with them is the card services team just simply needs to take a list of these are the questions that were asked of me. That's it. You have a shared doc. This question, right? It takes five seconds to put the question together.

10:02

Well, at the end of the week, we can take all of those questions and basically go into chat, gpt, and we can take our credit card manual and we can take these questions and we can say these are the questions bankers asked, which ones are addressed in the credit card manual, which ones are not addressed.

10:18

Put that into a table for me.

10:20

So now we have this really, really fast, quick, efficient way to be able to say it's an accountability issue and we need to improve our learning or accountability, or it's not in the manual. And then we can hit a few buttons and say, for everything that's in the manual, write me a frequently asked document that I can send to the bankers to help with the education piece. Now write me an email that I can send out to the bankers and attach that to it. And then now give me a list of everything that I need to update the manual with. And so when you think about, like that timeframe, if somebody had to go in and basically take note of all these questions, look them up in the manual, type out all of this, you know that's going to take an hour or two to do that, and now we can basically put something together in five or 10 minutes and we can start changing the format, we can change the audience, we can do all of that, and that, I think, is this productivity piece that we talk about with generative AI.

11:15 - Eric Cook (Co-host)

How do we?

11:16 - Ben Udell (Guest)

use those resources. So something that would be taking us one, two, three hours, we can do it in 10, 20 minutes and move that through the system much quicker. Everybody wins because of a solution like that, and that's not a heavy lift. When you get down to learning how to use AI, that's actually a really easy one. It's just there's no manual for any of that. So that's where it's, I think, this creativity side of the world to say how can I change my process to help educate them. So I think that's a real practical one and that, you know, I'd wrap that up by saying it also helps us understand is it a training and accountability issue, or do we not have the resources to help individuals? It's not a fight back and forth between people. It's recognition of where do we apply help and support for people.

12:02 - Fred Cadena (Host)

Yeah, that's a really great example and I love the fact that it's actionable. It's very you know I don't want to. This is not a criticism, but it's very basic right. It's very like I can take it and run with it today. Let me ask you this, because I know a lot of the listeners that maybe have not leaned into this. You know, maybe they've played around with Chachapiti or other things privately, but they've not tried it for business.

12:26

Sometimes you put some stuff in and you get some stuff out, and it can be a little cagey, it can be a little questionable what you're going to get. So, on a scale of 1 to 10, or if you just want to give some color, how much human interaction do you have through that process? So I'm taking these questions from the Google Doc, I'm putting them in ChatGPT with the manual. Chatgpt comes out and says here's a table. Are you eyeballing it? Are you going back into the manual and seeing if it's real or not? How much of that are you trusting your gut and how much of that are you trusting the model?

13:04 - Ben Udell (Guest)

I think that's a really great question. I think there's two aspects of that that people have to appreciate. The first aspect is really the prompts that people are using, and I'll give the listeners a good example. If I say tell a story about Fred and Ben on a podcast, I don't know what ChatGPP is going to give us right. It could be. It'll be entertaining.

13:27 - Fred Cadena (Host)

I'll tell you that 100%.

13:28 - Ben Udell (Guest)

You know I was. I was, uh, when I was in Vegas at a at the financial brand conference, talking about this, I said tell a story about Ben and Fred at a conference in Las Vegas, like, and everybody chuckles because it's wild to think about. What is chat GPP going to think about? Right, on the other hand, with my prompts, if I take a step back and say, write a blog that takes five minutes to read, that talks about Fred and Ben talking about practical applications of chat GPT, cite two really great examples of that, how do you get started with chat GPT? Put three bullet points in there that really emphasize the power of prompting, and then have a call to action. We have this really, really focused topic, and so I think part of the answer to your question is that when I see people struggling with either hallucinations or bad answers, it's often because they are just simply not prompting well enough. They're not giving this intern as we often say you know around this topic enough direction, and then the intern slash chat GP, goes off and starts making stuff up and it's. I don't think that's a negative towards chat GP. I think that's just the reality of how you work with it. So then to really dive into your question.

14:44

You know, I think what we really talk through in my model is A, b, c and D is your professional knowledge, your direction, your prompting, your expectations that you're adding to AI to ask for good questions. And then E through S is just letting AI run, it's letting it talk, it's letting it give you output, which then means your T through Z is your human in the loop. So to your point on how much are you looking at the information, that's that T through Z right when you're going through and you're eyeballing it to say that makes sense? I'm pretty sure that's in there, fully on board with that. Or maybe I have to either iterate within my prompting or maybe I just have to take it out of chat GPT and start editing.

15:30

Obviously, I think that's where this professional side comes into play as well, where I think people who criticize some of the output, even if it's good output, you still have to read it right. You're not going to take work from an intern and just pass it off as your own. That's career suicide, right. You're not going to take work from an intern and just pass it off as your own, like that's career suicide, right. And so I think that's where AI really really, really needs to be a compliment to what we are doing, and I think if people have that mentality you know whether that table is right or not then they can have that opportunity to go back in and say no, the table is not right. I need you to do X, y and Z, look at this more closely, and then that's how you kind of massage AI to get to a better solution for you which is generally much more productive than what you're going to be able to do on your own, even with some of that back and forth.

16:16 - Fred Cadena (Host)

Yeah, I love that. And I think that, where you started, you know you really can't underscore enough. You know it all comes down to how good your prompt is, how specific your prompt is. I think that's a beautiful analogy of you've got an intern, right. You've got somebody who's eager, who knows how to go out and get some data, they've got a background in researching, but they may or may not understand your business, right? You invite the intern into the bank, right? They probably had a savings account when they were a kid, maybe a debit card in their teenage years and I think less so now, but probably got three or four credit card offers while they were in college and that's kind of their extent of understanding what a bank is. And you start asking them to do some sophisticated things. They're going to need some context, as you've been training people both inside your bank as well as more broadly, how much effort, how much lift goes into helping people become better prompters.

17:19 - Ben Udell (Guest)

Yeah, that's a really great question. What I typically find like when I do some onssite training, whether it's two hours or four hours, I think people can get really big lifts right off the bat, and the main reason for that is nobody. There's no instruction manual, right? Chatgpt rolled out and they basically said, hello world, have fun, we're going to go work and do more development, right? And so I think what's happening is people don't have good prompts.

17:44

Or what I also find is you can go out there and you can easily find the 10 greatest prompts, the 20 greatest prompts, but really, when we talk about this transformational side you know, I just put some blog posts and some LinkedIn posts out on this is you could have insanely long blog or you can have insanely long prompts. One of the prompts I shared in my example is a thousand words long, and I think that's where this eye-opening piece really comes into place. So I think that ramp-up period kind of climbing the hill, if you will goes really fast with understanding that. And then I think what really is the next step to this and the reason I really like the intern analogy and I know a lot of people use this analogy and some people maybe overuse it and it gets poo-pooed is that if you have an intern or you have a colleague or you have a tutor or you have a consultant or you have a teacher or you have a boss, you don't learn their personality immediately overnight. You just don't step into that.

18:42

And so people take that approach with chat, gpt and these other generative AI models is you have to kind of learn the personality and you have to learn how you work with it and you have to learn how you work with it within your role. That might take a month. That might take a few months, depending upon how much you use it and how well you might learn the technology. And so I think, if people partner those two things together, just really great prompts and depth of prompt and the iterative process of having a chat with AI as well as it's a personality. It's kind of weird. At times you got to really invest into spending that time with learning your new BFF for lack of a better word right, that's kind of a fun way to think about it.

19:23 - Fred Cadena (Host)

It is a fun way to think about it and you're right, every model has a slightly different personality, slightly different, you know, and you can go in and tune some of that stuff yourself, as you know. And then also there's a history in those threads, right? So as you're going through a chat and I'll be curious kind of your approach to this, there's some that I go back to all the time because there's already kind of a history and a tonality and I can say, hey, you did this for me two steps ago, now change this and this and do it again, and it shortcuts some of the prompting.

20:00 - Ben Udell (Guest)

You know he said on that real quick. What people don't appreciate is that, especially in Chat GPT, all of your chats have a web address. So I have this long running, nearly year long chat on Microsoft BI. I literally just bookmark that so, as it gets pushed way down in my history, instead of scrolling through I just go into my bookmarks and click my chat GPT, microsoft BI and to your point. Now I've got that history built there and it's one of those things I tell people. You can kind of see the look on their face like oh yeah, duh, but now you're organized, it's great.

20:35 - Fred Cadena (Host)

No, it's great, and it's a really easy tip. And I'll tell you this I'm sure you run into this too the number of people that don't realize you can go back into your old chat histories and pick up where you left off. I mean, it's just one of those things, unless you're curious and figure it out on your own, or unless somebody told you, it's not necessarily the most intuitive of thing to find. Where I wanted to go was this, and you mentioned ChatGPT a lot. I don't know if that's just because you like ChatGPT the most or because it's kind of the Kleenex or the Xerox of AI and it's just the one that everybody knows. But what is your approach to selecting models and tools? Do you experiment with other ones and how do you kind of decision what it is that you're going to leverage?

21:20 - Ben Udell (Guest)

Yeah, that's another good question. You know, somebody said this once and I wish I knew who said it so they could get proper credit. But the message was I work with Copilot but ChatGPT is my BFF and I just really love that because as organizations are starting to add different models and play around and I thought that that was kind of a funny way and it makes a lot of sense. So I would say, from my perspective, I started with ChatGPT because it was kind of the first one on the block right and I think to that personality piece, the rhythm, the personality, the vibe, how it operates. It just works really well for me.

21:57

Personally, I do also use Gemini, the advanced version. What I find with Gemini for listeners that aren't familiar, if you've seen ChatGPT work you ask it and let's say you want to write a blog. We'll use that as an example. You write a blog and then you want changes to it. You either A rewrite the entire blog or, b you move it over into Word and then you might pull pieces out of it and put it into ChatGPT. The cool thing with Gemini is when you write that blog, you can go in and select specific areas within the blog and just rewrite smaller chunks of it in one large window, and so from a writing standpoint, that makes a ton of sense. From a research standpoint, or maybe a first draft chatPT makes sense, and then I'd certainly work with Copilot as well.

22:47

I just find it, even though it's powered by ChatGPT, it just is not as robust and strong.

22:51

It's more like this basic version to get going, and what I've been doing is I've been going in to learn both of them. I go into Bing AI to generate my content and then I drop it over into Microsoft Word to edit specific parts of it, which is effectively like I'm trying to do ChatGPT and Gemini smashed together. So you know, in my view, I think the world will probably end up using Microsoft Copilot for some obvious reasons. It's embedded, it's relatively inexpensive for banks, but I do think there'll be a lot of use cases for ChatGPT and for Gemini and these other models. I think at the bank we're probably going to end up with having way more users on Copilot, but then there'll be a lot of users that will have some of these other models, just because they do different things and they're inexpensive, relatively speaking, and can get them going. So I do bounce back and forth quite a bit between all of them. Part of it's keeping current on the industry, part of it's just different personalities, just like that intern example.

23:47 - Fred Cadena (Host)

Yeah, absolutely. I think that is a really well-grounded kind of example. I'll be honest, I've used Copilot a little bit here and there. I find the same thing you do. I describe it a little bit like replicative fade. It's like somebody copy pasted chat GPT and it's just not quite as there. There, I've not played around with Gemini as much. It's just weird. I use Google a lot. I have Google works. I probably have five separate Google workspaces for different things that I'm involved in. I think what happened is I tried it out a little bit when it was new and I just didn't have a great experience with it. But I just made a note to kind of go back, because that's actually one of the use cases. I also create a lot of content and write in this and it seems like that would work a lot better from a workflow.

24:40 - Ben Udell (Guest)

you, the, only thing that's interesting with that is Gemini seems fritzy to me, meaning when I do that blog example, and then when I want to go in and rewrite sections of it every now and then it's almost like Gemini doesn't want to edit within that chat anymore and it will give me an error. And then classic Google fashion, right, you'll follow the error and it thinks it's helpful, but it's not that helpful and so it's like do I refresh it, do I start over? And it's just. There's almost like there's more friction sometimes with some of that. So it'd be interesting to get your take on if you run into that as well or if it's just me and how I use it. And it's fine, not the biggest deal. But you get into a flow and all of a sudden it doesn't want to work right for you anymore.

25:28 - Fred Cadena (Host)

Yeah, that would be really frustrating. I'll definitely let you know I tend to, like my BFF if I had one in the AI sense would be perplexity. I just really enjoy it and one of the things I like the most about it is the references. It kind of gives you references by default and so, you know, sometimes I read them. Sometimes I just kind of look and say, okay, well, it drew its information from these places and you know, it's not like Twitter and weird places for information. It's you know, some trustworthy sites and so I'm like, okay, I can feel like this is probably a pretty good answer. But I use ChatGPT a lot as well.

26:11

I was curious. I don't know if you saw this happen in the last couple of days. I think it was in. I don't know if it was in their quarterly filing or if it was in some other thing they released, but Microsoft just put OpenAI on the list of their competitors, which I found very interesting, and I don't know if it was in response to them releasing or getting ready to release the search functionality or what maybe drove that. But I was curious. Did you see that? Do you have any initial reaction to it?

26:44 - Ben Udell (Guest)

I guess I don't think anybody should be surprised by that, and I think the other thing what was it two weeks ago where Microsoft started to back away from some of their open AI relationship and the word on the street was the concern over I don't know if it's the monopolistic side of the world or just the too much tech power concentrated in one area, and you know, Microsoft gets so much heat for their past history, you know, is this, you know, is that temperature starting to pick up, especially with, you know, all the European Union I'm not even sure the right way to put it but their legislation, maybe for lack of a better word on regulating AI. So I and I think that's, you know, when I, when I do some of these speaking engagements and they're scheduling three, four months out, one of the worries I always have is I decide I'm going to do this for them, and I always sweat a little bit. Is that going to be outdated or even worthwhile three, four months from now? The world is changing so fast for this stuff. It's amazing.

27:47

I'm getting requests to talk more about Copilot, though, which is interesting to me, you know, because I think we generally poo-poo Copilot a little bit, but I just think because it's embedded in Microsoft, they trust Microsoft for training and for privacy. It's easy to turn on. So it'll be interesting to see if that's just more like a little interesting blip or if there's going to start to be some legs to it and, frankly, for me, I'll start putting out more content that's Copilot related.

28:15 - Fred Cadena (Host)

I think, like so many products that are Microsoft products, it's far from best in class, but it's baked in and it's free and if that makes it to a video clip, I've got my fingers up in a quotation. I do think there's definitely bundling concerns around AI. It's very difficult to make the argument that AI is an intrinsic part of any of these different applications, and that might be part of why they did list it as a competitor. But again, I think Copilot is definitely going to get a lot of adoption just because it's there and it's bundled and it's, to your point, trusted Part of a trusted vendor.

29:02 - Ben Udell (Guest)

And I'll give Microsoft a lot of credit for how quickly they came out and embedded that in so many different applications in so many different ways in so many different areas. But I think that criticism, though, is when you get new users into AI, much less those products. It's confusing. You know it's almost like there's too much going on in that space, or you think about that. You know we're visual learners and I can go and I can use Copilot in the Copilot window. I can use it in Word embedded into a paragraph, I can use it off to the right-hand side and then oh, by the way, it's listed over here and I just I think that's overwhelming for people who are tentative to even attempt to use AI. So, pluses and minuses. It'll be interesting to see how that plays out over time.

29:51 - Fred Cadena (Host)

It definitely will.

29:52

I think part of it is and this comes out of my UX kind of background is, to your point a minute ago of how quickly things are moving.

30:01

I think part of what and not just Microsoft and Copilot, but I think everyone's trying to figure out is what is going to work well from a UI and UX perspective.

30:13

You know, and you know, I think, that they're taking a much more, you know, consumer-oriented like we're all kind of used to. You know, the classic example Facebook, like Facebook, doesn't put out release notes every time. They put out an update, right, they just change and they observe human behavior and then things that people adopt they lean into and things that people don't adopt they move away from. And I think we're seeing a lot of this in real time as companies are adjusting where the AI shows up and where the prompts are and how they kind of work. I think it's a lot of this like experimentation, because this is very different. You know this is not like Clippy on steroids. This is a very kind of different way for people to interact with their work and I just don't think, you know, UI and UX experts have a good sense of what's the best way for this to really be in the interface.

31:04 - Ben Udell (Guest)

I agree and I think for how much publicity slash, marketing slash, media slash investment that goes into it right now, realistically, are we still in spring training for how this is going to play out in a professional environment? Much less the rest of the world. And so there's people like us who are really excited about what's next, what's next, what's next. And then you've got other people saying I just I'll get to it when I get to it and see me next year, and that's confusing. And you know there's no fault on anybody for any of that because it's spring training, right.

31:41 - Eric Cook (Co-host)

It is spring training.

31:41 - Fred Cadena (Host)

Some guys show up for training, with a week or two to go, and others have been there for months.

31:47 - Fred Cadena (Host)

Well, let me turn the question back on you for a second. So you know from where you sit. You know in community banking and again you see the broad swath of folks. You know the super early adopters and the laggards and kind of everybody in between. How do you realistically see this changing on a day-to-day basis, like your average community bankers, day-to-day? In the next we'll say like year or two, right, I'm not going to ask you about 10 years in the future. Yeah, sure, but you know, near term.

32:19 - Ben Udell (Guest)

I still think this is a very slow burn in the community financial space, and I say that because when I think through what we're doing, so I've probably trained oh, maybe what's the right, probably number, I don't know 10, 15% of the bank on it, and I would say from both that percentage of people we have using generative AI, whether it be Copilot or ChatGPT, and then the fact that we're using ChatGPT's, gpts that's like the 1% of all of these community financials I talk to, and 1% is probably closer to 0% than 1%. And I don't say that to you know, I think to brag about it. I think it's just the reality because from a community financial space, there's still people that are very concerned about how do I manage the privacy and how do I manage the risk and what's the right guardrails and how do I run through the bias side of the world. And I'm on the ABA's AI working group right now and I mean there's a lot of people on that group that are just trying to learn to get their arms around it. So so when I think out a year from now or two years from now, if our 1% where we're at is even 25%, I I think that's really surprising. I just think it's going to take a while to learn and understand.

33:43

You think about the training process that goes into it. So, for instance, one of the things I'm building right now to, frankly, to sell to other banks and, frankly, non-banks is nobody knows how to train any of this. And, yeah, you can go out to other places, but the reality is it's not focused on banking, that's for sure. And so here's your four-hour training package that has all the slides you need, it has video examples, it has a facilitator guide, it has kind of a workbook style, it has direction, and that just doesn't exist right now. And so you think about all the steps that need to take place to train it well across many, many organizations in the industry, to put that material together, to train the trainer, to get the use cases, to build it through compliance and board. And then you have people that are waiting to see what the examiners look like and the auditors look like before they move forward.

34:39

Like two years from now, I just I think we're going to be surprised at how far we've come. But I think, you know, it's not going to be ubiquitous, it's not going to be widely used as maybe a way to, or it's probably. I don't think we're going to have, I don't think we're going to have giant productivity gains two years from now. At that. I think there'll be pockets of it. There'll be people like us, there'll be pockets of it and growing, but those are a lot of steps that people have to get through for this stuff. So, anyways, that's maybe a little bit of a rambling all over the map answer to it, but I think that's the reality of, you know, new technology in a new space and it's going to take time.

35:15

I'll tell you one other quick story. I was talking to a credit union kind of some friends and I have there and I was talking to they basically were promoting this younger guy to be like their AI guy and they were telling me they're like I. Just we feel like we're so far behind and you know, we got to catch up to everybody and I'm like, honestly, in a community bank space, I have not heard of anybody that has a dedicated AI person. Like I'm not dedicated, I'm innovation and it's all of this stuff. This guy's just for AI. Like you guys are so far ahead of the curve right now. It's pretty wild to think about and that was, you know, perspective where they just didn't have because they don't do the speaking and the engagement around that that I do. So it's pretty wild.

36:00 - Fred Cadena (Host)

I think more institutions could benefit from that anxiety that they feel like they're behind, because I think without that they very quickly tend to fall behind. I think your answer is fairly reasonable. I'm curious one of the things I've been thinking about more and more around AI and adoption, and we see some of this with Copilot. Salesforce has done a pretty good job of building into this interface. You know AI hooks where you can kind of like you know you're using AI but you don't have to really think about it because somebody else has already kind of wrote the prompt and you know kind of designed the guardrails around it. You just kind of click a couple of buttons and it can write product descriptions or it can do. It can do these and that different things. I'll say it this way For the last year I've heard and I've been guilty of saying AI is not going to take your job.

36:56

Somebody that knows how to use AI is going to take your job. This is dangerous to ask the person that goes around teaching people how to write prompts whether or not they should learn how to write prompts. But do you think is the answer going to be that at some point everybody is going to have to get good at running AI at this level? Or is AI going to get so baked into products that you're going to use it and you're not even going to realize that you're using it? I think back when I was a kid. I used to play around with computers probably not a surprise to anybody that's listened to the podcast. I would get into the OS and I'd figure out what's going on. You don't really need to. Nowadays I use a Mac because I don't want to have to think about any of that. I just want to turn it on. I just want it to work right. I don't want to have to think about my batch files and what's going on under the hood. So what are your thoughts on that?

37:56 - Ben Udell (Guest)

You know it's a topic I've actually thought quite a bit about. My analogy for this is Excel. You know, especially in the bank space, basically everybody at some point in time touches Excel. But when we think about the range of users, there's some people that still, frankly, don't use the sum function. They literally add it up. I mean, that's less so these days, but there's people listening here who know in the last few years they've run into people like that right. And then there's people that are power users, that are doing macros and VBA and then they're doing crazy graphs and pivot tables and filters and slicers. And then there's these systems built onto it that are effectively doing Excel-esque formulas. It's just the business model is much more advanced and analytical, et cetera, et cetera. So I think for me that's what that analogy looks like.

38:46

I think everybody's probably going to have to understand how to have some interaction with AI, but to what level, I think will be very dependent upon the role. I think in some ways that's much like, you know, the internet, you know, rolling out right Is it just is ubiquitous and things just work. And so it's probably, in my mind, going to happen is you're going to see all of these startups now that are starting to build all these agents that connect to AI. And so, to your point, you're using AI, you just don't know it. And so I was doing a training with a construction company, actually, and getting them going with generative AI, and we were just riffing on stuff.

39:23

And that's fun riffing on stuff right, and that's fun for me because I don't work with construction companies and they're super excited and game on, for this is we're just like riffing on ideas, and one of the ideas was how long until Amazon or Whole Foods or somebody like that says take a picture of your cupboard. Ai will identify all of the food in there. It will then give you recipe ideas, show you the gaps that exist to fulfill those recipes, allow you to click a couple buttons for what you want to order, and then it places the order right and imagine just click hmm, that looks good, I know I've got that in the fridge. Order this, this, this and this. Hit, send and you're done. You've basically meal planned immediately and it'll tell you calories and it'll tell you. You know, you could probably have a program for it's a meal of two or a meal of four, two adults and two kids, and so like. That's this great example where we could probably do that to some degree right now with AI.

40:24 - Fred Cadena (Host)

I'm pretty sure that exists, except the ordering interface. I'm pretty sure the rest of that is a thing.

40:30 - Ben Udell (Guest)

Yeah, 100% right. And so imagine somebody who doesn't know anything about AI, but they just click a couple buttons and they're done. And apply that to what we do in our. That's this just interesting, creative idea. I'm sure there's people working on that already, but the point is that's our future with AI, All these little agents who just do this really helpful stuff for us.

40:52 - Fred Cadena (Host)

Yeah, no, I think that makes a lot of sense and, like I said, I definitely have seen a lot of that out there, especially the consumer side. It'll be interesting to see how that plays out in business, you know, and where those use cases can take place. One of the things that I think is going to have to be a consideration you touched on this a minute ago is the compliance and regulatory aspects of this, and so you know what are those conversations like. What do you coach your institution and other institutions you work with on getting right from a compliance and regulatory standpoint?

41:25 - Ben Udell (Guest)

Yeah, that's first of all. It's a podcast in itself in many ways right. So the short version, this is what I would say. I think, from kind of what the feeling is coming from the regulatory side of the world, the examiner side of the world, the legal side, the bank side, I think, first and foremost, clearly you have to have an IT usage policy that addresses AI. That's no surprise to anybody, but I think the one helpful piece is privacy is always a big part of that we already have in most of our IT usage policies. Don't upload private information unless it's an approved vendor, and then you have a vendor process. You go through for all of that. When you think about policies around bias and around fact checking and quality material, you already have policies and procedures in place. Utilize those. Ai is part of that process. I think the other thing that we need to see and this goes back to the training piece I was mentioning if you're going to turn AI on, you have to provide some sort of training to the associates, the employees that are using AI, and I don't think it's well defined what that means exactly. But if you take a step back and say we have to make sure that it's very clear that you can't upload private information unless it's approved. You have to look for bias, you have to fact check, you have to cite sources. You can't discriminate All the stuff that we go through right now. It applies to AI, but I think there's the specificity of doing that which then leads into, I think, the next level of this, and this is where I think working with the WBA and that AI group has been really helpful.

43:05

I think there's two tracks to this and I've been putting a lot of thought on this because I got to start writing some blogs about this and putting information out. I think we have to think about it from two paths generically, because there'll be some overlap. The first path is when I use AI, does it already fit into my current processes? So, for instance, if I'm going to write marketing content, if AI writes it, if the intern writes it, if the bank president writes it, that material comes through the process. A subject matter expert reviews that. You've edited it, you've branded it, you've fact-checked it, you've looked for bias. If you're talking about items that might hit HMDA regulations or fair lending regulations, you have fact-checked that. It doesn't matter how that information got to the client. It's gone through the same process, whether it was written by AI or whether it was written by a human being, and that means there's a human in the loop along the way, right?

44:02

I think demonstrating that is very, very, very important. Which then leads to the other side, which I think is going to be more challenging, and that is if you have AI making decisions and the Treasury Department love to use the phrase explainability if I am making a rate exception or a rate recommendation, can I demonstrate the explainability that went into that? And if we take a step back in, like the larger umbrella, with AI, you have robotic process automation that is arguably under AI you probably have a high degree of explainability, right, because you've literally mapped out what goes into those variables. If you have machine learning, you generally have a pretty good mapping out of that as well. If you have ChatGPT's GPT, you may not have great explainability because of the black box that exists. It's not suggesting that that's right or wrong to use it. What it is suggesting is I think there's going to be a lot of anxiety over what inputs are you giving it, what are you training it on, how do you audit that? What's your process to continue to revalidate and retest to backtest to make sure that you have consistent outputs when you're using it. So that AI agent, that's a black box. It's mimicking what you've done before. It's just increasing your productivity to get to that pace and I think that's going to be the harder piece.

45:32

But when I think about the community financial side of the world, I don't think we have a lot of things right now that fit that bucket, because those are probably vendors that we're working with. They're probably, you know, we're not making rate decisions based upon this black box. We're making it on more traditional means, and so I think we're going to be okay there for a while. But I think that's going to be really kind of this frontier, if you will, that we have to think very critically about, and I think that's what's going to start coming up more in some of the regulatory side of the world. It's what the treasury is looking at with their request for information on AI.

46:05

It's just going to take a while to get there, but I think, for the listeners if you're in this space I think that's the thought process I would go through. So when an examiner comes in and says how are you using generative AI, you're very specific around. We use generative AI complementary to what we do, the processes, the policies that all align and we are consistent, whether it's a human or an AI. So when it touches a consumer, we're good or we are. We've created this process. That is a black box, but this is what we're doing to help you know, ensure that we are not. We are accountable to the laws and the regulations that exist out there and we can validate that for you by giving you that history right.

46:46

So I think, that's a long-winded response, but that's also like the reality that we're kind of dealing with right now, that you have to have somebody that's going to be able to clearly explain that to an examiner when they come in, and you know you're comfortable with all of that. That's back to that couple questions ago right, where it's going to take a year or two at a minimum to really get.

47:06 - Fred Cadena (Host)

I think some more focus going on this area. Yeah, no, I think that's definitely, you know, very well thought out. I mean, I think the first half to your point is the easier half. I think that that is where probably at least the conversations I've had most people in institutions are is in kind of figuring out that piece of you know. How do we put this into our workflow and not have it be in a position where it's making decisions I actually don't know personally that LLMs, you know.

47:37

And again to your point of various types of AI technology, I don't necessarily see that. I will say never is a long time ever being the right technology for things like credit decisioning. It's just like could I take a screwdriver and use it to beat a nail into a wall? Absolutely, is it the right tool for doing that? No, it probably isn't, but there's a lot of things that I think about that are closer to that end of the spectrum.

48:06

So one of the things I've been very interested in is for a long time before generative AI became a buzzword is getting down to markets of one and really personalizing offers and everything from understanding what your customers need is based on everything in their profile and everything in their transaction history and everything that you know about them zero party, first party, third party data as well as then translating that into a personalized offer.

48:38

And I think if you are ever going to get to do that at scale, you're going to have to have something like an LLM generating marketing language, generating images, making those selections. And I think at that point that is an area where I can obviously see regulators coming in and saying how did you decide this offer, how did you decide this language? And really being able to walk through that thing. And I think that's an area where, before I see it going into credit decisioning or exception management, I definitely see like more of those types of use cases and I have been putting a lot of thought around how that would work and how you would be able to kind of test that and also, you know, explain it when the regulators come in.

49:23 - Ben Udell (Guest)

Yeah, I think, if anything you know, there's some more democratization of some of the machine learning tools that are out there, and there's some interesting people out there I was just talking to Savvy AI about it that's trying to democratize some of that. They were an interesting one, Obviously AI, if you're more kind of self-study on that. But what's interesting, though, is back to like those rate decisions. Interesting, though, is back to like those rate decisions. Right now. I bet you 99% of the banks and credit unions out there, if they want to do a rate exception on a deposit account, for instance with Freddie, he calls up I got this money, I'm going to move elsewhere, I need the best rate. They're just like yeah, we're in a good mood today, Good client, here you go. And so it's funny how, technically, it's not hard to get generative AI and LLM to be more consistent and fair than that random discussion. But now that we've wrapped AI around it, now we've got let's call it some significant scrutiny about being fair or not.

50:23

And I don't know if that's right or wrong but it kind of backs us up to the beginning part of our conversation today, which is forget all of that and focus on the areas that you can get greater productivity and creativity and start getting some wins there and let the rest of the world catch up on some of the other areas down the road right, exactly.

50:43 - Fred Cadena (Host)

Well, this has been a great conversation. I just want to be cognizant of time. But before we wrap up looking ahead, we've talked about a few tools Like what are you looking at today? Like, what do you think is you know on the horizon, as, whether it be, you know, a new model, a new version of a model, a new, like purpose-built application, whether it's, you know, bank-specific or more generalized, like what do you see on the horizon that you think is going to have a lot of impact?

51:12 - Ben Udell (Guest)

I think. Well, first of all, I think when we're recording this, there's a lot of talk at ChatGPT coming out with their next level model, so maybe Murphy's Law says it happens this weekend, since we're just talking about it coming out sometime. That would be awesome, right? I think there's two things that are more practical. I think the first. What's really exciting to me is all of these companies that are popping up that are using generative AI generally through, for instance, a chat, gpt and, to some of that discussion a little bit ago, it means our employees don't need to learn great prompting. It's the agent or the wrapper around the agent that makes it a lot easier, and so, for instance, one of those that I've been working with is a company called CustomGPTai. It's basically a fancy GPT builder and we can upload all of our policies and procedures into it, and now, either in a web link or embedded with HTML code on our website, our internal website, it becomes a search tool for us, and so imagine a banker going in and having to search through all of SharePoint trying to find something, and that's difficult. Or multiple versions of documents pop up because nobody uses SharePoint well, and this is a doc that's going to like, highlight the. And, oh, by the way, what's cool about it is it shows you the reference documents it came from. So if somebody questions the factuality of it or they need more info, click the document and then you can go into it. And so I think it's a lot of systems like that, where it's building around it.

52:47

And then I think the other piece is these more traditional vendors. I could give you an example. I work with a company called Marquee, which is a great company. They do a lot of CRM and marketing customer information files and some compliance stuff. To take their documents now, which are all probably more PDF orientated, drop them into a model that's connected to an LLM, so I can go in and say, walk me through all the steps to create a user defined field in the marketing data platform and it just spits all that out for me. Now give me practical ways to use that. And this is, I think, these tools, where we aren't going to build it, but our vendors are going to build it, and it makes that client experience much better for us with them. It's not hard to do, and I also think whether you'd like to go by the PDF version or you'd like to go by the AI side. They can live in harmony together, and that's now. You know, from a user experience standpoint, you get the benefits of both worlds. So I think that's part of that.

53:49

I do think the challenge, though, is I spoke to one vendor I haven't mentioned, and they were asking me all these AI questions and they're like why are people not doing more AI? And you got to get doing this, you got to do this, and I'm like listen, we don't have the technological savvy across the community financial industry to dive in like that. You are talking about AI. I am talking about how do I make search easier? That's different to a search user versus a fintech. I think these light bulbs went off of in some ways. It's like how do you talk to your non-fintech people about what you do? If you explain it to in fintech language, they have no idea if you explain this every day. That's, I think, this other gap, and I think that will be an interesting place, with some winners as well in the space who can understand what that experience looks like, not from a fintech, but from just. You know, marion Deposit Ops or Bob in the front desk who is just, I don't care, I need it easy.

54:45 - Fred Cadena (Host)

Right. Yeah, it's about figuring out what are the business outcomes you can impact, and not about here's a fancy new, faster model, more parameters type of a thing. People like you and I get jazzed up about that. People in the middle of that adoption trough just want to know how to get their job done more easily.

55:07 - Ben Udell (Guest)

As soon as you say, oh, you do AI, their eyes glaze. Everybody says they do AI right now. It's been funny with me being my role in the industry now and my engagements. We'll have vendors start to talk to us about AI and I'm just like tell me about this, this, this and this. And you can kind of see them take a step back, like tell me about this, this, this and this, and you could kind of see them like take a step back, like oh, all right, he knows what he's talking about.

55:31 - Fred Cadena (Host)

Anyway, you got to get somebody else on this call.

55:33 - Ben Udell (Guest)

Yeah, somebody else Like oh, so what you're really saying is you're using machine learning for your chatbot, not generative AI. And if AT&T can't figure out their chatbot, why are you telling me you can figure out their chatbot? And they're like well, you know, it's some programming and I'm like that's fine. I'm not dismissing your solution, I'm just saying we're not. No, thank you. No, thank you, for right now, it's not the AI that we want.

55:59 - Fred Cadena (Host)

Thanks, but no thanks.

56:01 - Josh Matthews (Co-host)

Yeah.

56:01 - Fred Cadena (Host)

Man. It's been a great conversation For listeners out there. If you need guidance, if you're trying to figure out where to get started with generative AI and really finding applicable ways to get it working with your teams, today contact Ben. He does consulting, he does training, he's able to speak and he's open to a lot of different collaborations. Bottom line is if this is something you're trying to figure out, ben can help you be more successful more quickly.

56:29 - Eric Cook (Co-host)

So thanks for the time, Ben.

56:32 - Fred Cadena (Host)

We'll talk again, thank you, and we're back Super excited to get the Quick Takes Roundtable team back together after a little bit of a summer hiatus. Josh, eric, welcome back. How's the summer been treating you folks?

56:50 - Josh Matthews (Co-host)

Thanks, buddy. It's good to be back so far, so good man.

56:53 - Eric Cook (Co-host)

It feels like it's almost over. They say when you get older, stuff goes faster, and I didn't think that that was real, but it happened this summer.

57:01 - Josh Matthews (Co-host)

Zoom. That's because your summer camp let you out early for bad behavior.

57:05 - Eric Cook (Co-host)

They did. Let you out early for bad behavior. They did, they did. But that's okay by me.

57:15 - Fred Cadena (Host)

Well, the highlight of the summer for some people is the Summer Olympics. Summer Olympics are recording Monday. I think Today's the last day, or yesterday or the last day is coming up. Yesterday was the last day. There we go. Somebody that followed it closer than me there you go. So just start off a little bit off topic for the podcast. But what are your all thoughts on the Olympics this year?

57:31 - Josh Matthews (Co-host)

I got to tell you, look, I I'm more of a winter Olympics guy.

57:35

You know all that agony of defeat, ski jumping and stuff like that, Plus having lived in park city and having gone to the Olympics. But ever since I was little I was a winter, winter Olympics fan. So I kind of approached this this year's Olympics with a little bit of trepidation, a little bit of like oh God, I don't know whatever, a bunch of whatever stuff. And then the more I spent some time with it, the more I started to get into it and really get a sense of, I think in a more adult way, be able to wrap my brain around different aspects, the idea that there are certain sports out there where the winner is clear and then anytime you involve a judge it's not right, which is proof that human input, human interaction of scoring can get really messed up. I remember going to the Pairs ice skating event when the Russians were colluding and trying to screw the French out of a medal, and we've seen some weird stuff with rules around one of the USA gymnasts, the Jordan Childs.

58:39

That's right, that's right. So, we've seen some controversy. The controversy almost always stems from judged events.

58:46 - Fred Cadena (Host)

I'm with you and I won't take credit for this as an original thought but I'm with you in having some concern about the fairness in completely judged events. And yeah, I was disappointed and we just actually just got in my phone about 20 minutes ago I guess the final, final, final appeal for the gymnastics was rejected and so, even though the scoring was wrong to begin with because the appeal wasn't filed in a minute, you know we lose the medal and you know, again, it's one of those things where anytime you can't go out and like objectively prove that one person or one team has beaten the other team with with objective measures. It's, it's troubling to me. There's lots of things that are super athletic I love. I love gymnastics, I love swimming, I love a lot of these events, but if, if you don't have an objective winner, it is difficult and there has been a lot of manipulation. I kind of hope and you know we're going to make it a whole like four minutes before I mention AI I would hope in the next couple of years that some of these places introduce some AI. Right, it should be very easy to like set up a bunch of cameras and just say like, hey, was this element completed or not? Was this person straight when they landed or not? Were they inside the line or not? And just eliminate some of that uncertainty, because a lot of these people work really hard, like that's. The thing I want to celebrate is like how hard a lot of these athletes work.

:: 00:19

It's incredible and I'll be honest, I probably didn't watch very much Olympics this time, but I did flip on Peacock from time to time. Did flip on peacock from time to time. The. The thing that I always like about both both winter and summer olympics, is just the number of obscure sports out there. Sure you know the the handball is probably the one that I every year. I'm like what is a giant net and they've got like a volleyball and they're smacking it and like it just looks like something I would have come up with with like a bunch of my friends when I was in the fourth grade. But it's a sport and and I mean I'm sure the people that played at that level are incredibly athletic it's fun to watch and so that was cool this year just to be able to kind of like you know. Hey, here's something I'd never think about watching fencing. I watched fencing twice.

:: 01:03

So used to be a fencer, a competitive fencer, yeah yeah. And that's like I was talking about the unfairness of judging the other day with her and she was saying, and I was saying it's not like fencing Beep, you got a point, man, that buzzer goes off, you're wired up. It knows we're already using technology, whether it's whether it's I mean it's not, it's almost like a radar thing that they use for the finish line on sprint races and stuff like that. Right, I mean it's crazy. So why not use AI when you can, where you can?

:: 01:35

I'm going to just throw a dissenting thought out there, just to poke the bear a little bit. Do we want to get to a point where human judge and the element of human involvement at the judge, the officiation, the evaluation process is completely eliminated from sports like this and it's all based on a machine? Or I mean, where's the happy medium? Because I think of tennis. You know you've got the line judge there that's calling everything, but then you've got high-speed stop motion cameras that will tell you exactly where that ball hits yeah, it's referees and replays right right, right, so is is.

:: 02:19

I don't know, I just I, I don't know if that's a slippery slope where we don't trust the human anymore, so we just want to get the human out of the loop so that we can just go do our own thing, or if that is one of the elements that makes sporting and competitive events. You know, there's been countless football games that have been cost based off of a call from an official and you know, is that just the nature of the beast and that's what makes sports somewhat of an unpredictable experience.

:: 02:52

I don't think so. I think sports have rules. Again, the unpredictability of sports, I think, comes from lots of other things. Right, like no balls completely round or completely the right shape or weather for outdoor sports or baseball from the size of the field. There's a lot of places where unpredictability can be brought in. I don't think fairness and accuracy in judging is one of the places. We want that principles.

:: 03:26

If we were inventing gymnastics today or any of these judge-oriented sports, I think they would be built with as much automation in the scoring as possible, but obviously that didn't exist back then. Now there's some things that I'm certainly not an expert. There's some sports and there's some competitions where there's like an artistic element to it and, like you know, like those kinds of things, I think there's still room. I mean, I don't think an ai is going to judge artistic elements and that kind of stuff. And I think if it's something like gymnastics where, like there's a scoring sheet, it's you know. Here's the difficulty of the elements. You need these elements. You know they have to be executed to this degree of accuracy or you get penalized. I think that can be all done in AI and I think that would make for a much more fair and consistent outcome.

:: 04:14

It probably would, and I do agree with you. But I'll say this the challenge isn't really the judging in the Child's case. The challenge is in the ability to contest it and what I will call letter of the law over spirit of the law, because they missed it by four seconds. Right, and, let's face it, it's not a fricking F1 race. Okay, like, use your judgment. Yeah, you're supposed to do it in a minute. Oh, it was a minute. Four seconds. Close enough, we erred, let's fix it. You know what I mean? This is spirit of the law versus letter of the law, and you have to have letter of the law, but when it comes to review, I think there should be some spirit involved as well. I think that the goal is to get the person who won the medal. That's it, and if your rules prohibited that, then there's a problem with your rules. Change it on the fly and apologize yeah, it's.

:: 05:05

I mean it's a both end, like they would have never had to file an appeal had the original judgment been accurate.

:: 05:12

Right, and you know what? They never would have had to judge it if they just got rid of the sport. So we could do that too, ouch. Let's not do that I like gymnastics. I'm just kidding.

:: 05:20

I know Eric has a sport he wants to get rid of Speaking of sports to get rid of it.

:: 05:24

And I saw some of the good ones and I still was getting my head wrapped around it because maybe and I was never a break dancer myself, but I had friends that were I'm I'm just, I was Prince.

:: 05:34

Popper baby, there you go I.

:: 05:36

I'm and this will probably generate some hate from our break dancing friends, but I just don't know is if that's an Olympic sport and let's not so you know they got rid of it for LA.

:: 05:46

They've already gotten rid of it, have they? Yeah, well, and I don't know if I have.

:: 05:50

Australian last place Raygun. I mean that got you know. Somebody says you know they never remember the last place finisher, but in this instance they did. So it's kind of going all over the I don't know. It just seemed.

:: 06:05

I don't know, I was glad to see it. I'm with you.

:: 06:07

I'm with you, that was much more exciting.

:: 06:11

It's one of those things to me. And it's not an Olympic sport, but it is a competition, at least here in the US, that's very competitive at the high school collegiate level, like cheerleading. It's very athletic, right. I'm not taking anything away from the hard work and the athleticism that goes into cheerleading. I don't know that it's a sport. I'm not taking anything away with the hard work and athleticism that goes into break dancing. I just think there's too much like artistic and and too much room for subjectivity for it to be a sport. Not saying it can't be a competition, not saying you can't judge it on its own merits. But I'm kind of with you.

:: 06:49

I don't know that it's sports, other than being the source of some really good memes.

:: 06:54

I saw this I don't know what it was, but it wasn't Ray Gunn's thing, but it was one of the women's competitions and I'm watching it and I'm thinking okay, so these people have to compete. First of all, they're introducing the element of music and they don't know what song they're going to get. Well, great. So now it's really about improvisation. What are you going to do based on the beat and the rhythm of the song? Okay, I guess, fair enough. But what I couldn't stand, what I absolutely abhorred, was the idea that you're doing this competition and while it's your turn to do your break dancing, your competitor is standing there, pacing around, licking their lips, shaking their head.

:: 07:33

I just thought get this bullshit out of the Olympics. That is the most unsportsmanlike BS. And you're forced to watch it on camera. You're forced to watch it and I literally thought and this was before Ray Gunn came up this is unwatchable. Now I did watch the gold medal competition, the final battle, and those guys. It didn't bother me as much because they do a routine and they'd nod at each other like hey, good job man, like hey, that was pretty cool. So I like that supportive element, but this sort of like nasty you know what's that show Tattoo master, ink masters, where everyone's dissing on everybody element is like there's no place in the in the Olympics for that kind of garbage. Keep, keep that to TikTok.

:: 08:24

Yeah, I'm, I'm with you, like, and I think it's part of breakdance culture, and I'll be the first one to tell you I'm not part of breakdance culture, but I'm with you, that's because you're successful.

:: 08:33

It's not part of Olympic culture.

:: 08:35

And I think that's part of the reason why it doesn't really belong in the games.

:: 08:40

It's unsportsmanlike is what it is in what they're calling a sport, so get rid of it.

:: 08:46

Yep and they did.

:: 08:49

So, getting back to the AI topic, there was an article that came out just a little bit after our last episode after our last episode we didn't talk about it then on a Citigroup report that showed that the banking industry is predicted to be the hardest hit from the deployment of AI. Their prediction is 54% of roles at banks are at risk for AI-led job displacement, and I kind of want to talk about that, but I want to talk about it in the context, Eric, of that LinkedIn post that you shared with us around authenticity. So do you want to introduce a little bit of that into the conversation?

:: 09:29

Yeah, I do. And as it relates to the impact on banking sector, I've got some interesting thoughts tying that back to the community bank world as well. But the post from LinkedIn, john Munsell, his friend of mine, does some AI training and is teaching for LSU shared a post from Michael I think it's Hass or Hassie H-A-S-S-E. Shared a post that he put out just kind of a hey I want to know what you think. And kicked off with this is becoming more of a concern to him and essentially the nutshell version TLDR is he had an exchange through email with a salesperson not directly on his team but part of his organization, who is not a native English speaker, so it is their second language. But there was an email that went out that was obviously written by Chad GPT because it was in perfect English and this salesperson in question just doesn't speak perfect English. Nothing wrong with that.

:: 10:34

In fact, that's kind of what makes that individual unique. When you get an email, you know that English isn't their first language and you would expect that, and he responded with, you know, being very delicate and basically just joked you know, was it really you and proceeded to get lit up by this individual that you need to get on board with this AI thing and this is the way things are going to happen and you need to be able to embrace it. And if you're fighting it and completely missed the boat, that it wasn't genuine in his own voice and authentic and so to, I guess, put a little fuel on the fire just to see, he threw some responses into chat GPT and had jet GPT write a response for him and then, as he put it, you know, they've just basically generated a couple of paragraphs of word salad back and forth at each other and the gist of it is that's a new sport for the Olympics.

:: 11:28

is AI chat battling?

:: 11:30

AI chat battle and I'm going to stand there and look at you and lick my lips while you type your prompt and see what's up.

:: 11:37

I wonder a little bit if they were even reading each other's responses or just feeding them right into chat. Gpt.

:: 11:43

Well, and Michael even said to an extent there was a little bit of offense taken because I don't even warrant you taking the time to write a response to me. You just throw it into chat GPT and let chat GPT respond to me. Response to me, you just throw it into Chad GPT and let Chad GPT respond to me. And I've been saying this ever since the early days of social media, where people have said well, can't I just hire somebody to write all my social media posts? I don't have time to be on Twitter, I'm just going to hire somebody else to do it.

:: 12:11

And there's always been that fear from my perspective and that's why I'm a big advocate of personal branding being you warts and all is. If somebody meets you in real life and you don't talk the same way, you don't share the same information, you don't have the same perspective because your one persona online is different than your real persona in person, then that breaks down. And it didn't happen a lot, but it certainly did. But now, with everybody and their brother having access to Chad GPT or Claude or perplexity or copilot or Gemini, whatever you can go in and have it write you eloquent prose that comes across great and I'm even fighting back now because I use Grammarly when I type my emails and you never read a book you know, like you never read a book.

:: 13:01

And I will ignore or uncorrect stuff that Grammarly is recommending, because that's not the way I talk, that's not the way I write. I write the way I talk and I want that to come across that way. So the wanas and the ums and the other sorts of things, I weave that into the way that I actually am communicating, whether it's written or verbal or whatever. So I just think and I've reached out to the author of that post directly because I didn't want to stir the pot in public, although maybe I will Curious what the demographic of this individual.

:: 13:36

You know, this is a podcast, right? Yeah, I know.

:: 13:39

So maybe he'll hear it. This is going to go out, maybe he'll hear it. Millions of people, millions of people, all bazillions of people. I don't know if there's any demographic, if it is what the age is. And there's all this thought that the younger generation embraces technology and they're not as concerned. They're always on their phones. They don't communicate in real person. You can't get them to talk to you on the telephone. They all want to work from home. All those you know things that are going on generationally speaking. I say that sounding like the old guy in the room, but it is what it is.

:: 14:16

Yeah, so I think that authenticity is fairly universal. I could be wrong here, but I think that and if anything you know, I've got in the younger generations rather than a bias against it. So I think in the long run, the sales manager whoever the person was pushing back is in the right from what I think is being equal. The salesperson that's writing an authentic email, that's connected and comes across as them is going to do better than the person that puts it all in chat, gpt, all other things being equal. I don't know, we'll see how it bears out.

:: 15:12

I use AI a lot. I use AI a lot, a lot in all aspects. I use it to do research. Quite frankly, a lot of the stuff that went into the prep of this Quick Takes Roundtable was all prepared and summarized using perplexity, but this is my rule of thumb.

:: 15:31

To date, I have yet to have a single thing that I have put into any LLM come out in a way that I want to put it out in the public.

:: 15:41

Now, do I use it for internal documents, like if I'm doing like account plans when I'm working with accounts, or do I use it for ideation and for drafts? Absolutely, I use it all day long but nothing has come out sounding enough like me that I'm ready to just copy and paste it and put it on LinkedIn or copy and paste it and throw it in an email, and I'm not saying that's never going to happen. I've been working recently, inspired by Scott Galloway and some of the stuff that he's talked about, and taking a lot of his writings and a lot of his output and training in LLM on it. I've been working on trying to train something that will sound like me and I don't know that again, I'll ever feel enough right about it that it's just going to be a straight through pipe from the LLM to the public. I'd like something to be able to produce things more in my voice, but today that's not there and me reputationally I'm not going to put stuff out there that is not in my voice.

:: 16:39

Right, I've got a perspective here. I get the concern about AI affecting authenticity on LinkedIn. While AI can help automate tasks and improve efficiency sorry if you're getting this already it's crucial- what is that?

:: 16:54

Is that, claude?

:: 16:55

It's crucial we don't let it replace genuine human connections. The key is to use AI to handle routine tasks so we can focus on authentic, meaningful interactions that build real relationships.

:: 17:11

That's going to be, claude.

:: 17:14

It doesn't freaking work. That's perplexity. It just doesn't work, guys. You just have to test it once to tell that it doesn't work for human interaction back and forth, unless that other human understands they're talking to a chatbot. I know I'm talking to a chatbot. I am expecting it to sound like a chatbot and even when it tries to be cool, when it's like right on, we'll get right on that, just hang tight. It's like I still know it's chatbot, right, because a robot wouldn't talk like that.

:: 17:49

What's that? Maybe it threw RIS in or some of the other updated slang.

:: 17:54

Yeah, I'm so not in touch with any of that. No, I'm really not, like I'm so not in touch with any of that. No, I'm really not, like I'm not. But it's clear that anyone who's trying to communicate with somebody, especially ESL, it's like why don't you use the AI to go learn another language instead of using it to communicate through that language on something that's going to get blown wide open?

:: 18:16

Yeah.

:: 18:17

I think the person who used the chatbot to communicate and comment should be ashamed of themselves. Just briefly, as long as they learn from it and then move on and don't do it again. It's terrible. I don't want to not underline this. It's terrible.

:: 18:36

Don't hold back John. When have I ever Fred? I'm curious Here's't hold back John. Yeah, like really really have I ever, fred?

:: 18:40

yeah, I'm curious. Here's the thing there are too many. There are too many people pussyfooting around stuff and being soft and trying not to divide the herd. Well, sorry, the herd is divided and you've got to know what side you fall on. I'm clearly I know what side I fall on with almost everything, not all things, but almost everything. And I I'm fine with that because, guess what? There's plenty of people on this side. There's plenty of people that can hear me and want to move over to this side and be like, oh shit, someone's making sense finally. So fair enough. But I don't think not having a strong opinion on this stuff. It's like being McDonald's it's fine, good for the masses, but think about who your customers are. They're probably not my actual customers. You know what I mean.

:: 19:24

I'm curious from a recruiting perspective. Yeah, well, from a recruiting perspective, cause you live in that world, josh, are you? Are you seeing more cover letters, more resumes, more letters of interest coming across, where you read that? Yeah?

:: 19:40

Do you just pitch those? I see it every day.

:: 19:43

Do you? Do you pitch those? Do you provide feedback? Do you? To what extent are you trying, if at all, to to implement any sort of a change in belief? Or you know, hey, you probably shouldn't do this, or do you just not worry about it and let them move on and get hired by somebody else?

:: 20:01

No, so it's a really good question. So, yes, I see it and I see it every single day. I get a lot of people message me every single day and I can't ever get back to all of them. I do my best and the reality is is these messages come through and what they don't know? They create this post or not a post. They create this message to me or to whomever. Maybe they're sending it to 50 employers. They create this thing and they think that it's helping them stand out and be unique, because they took the time to actually explain things and talk about themselves. What they don't know is that I'm seeing 50 of these a day or more, just like resumes, just like how people respond to interview questions, just like sales letters with a subject line quick question, quick question.

:: 20:45

Josh, I mean, how many of those Like, hey, if I could get you 15 to 20 more clients every month, is that something you'd be interested in? It's like I see these so many times a day that you develop. We've got these computers. Our computers are better than AI. They really are. They're better than AI. They're faster. Maybe they're not faster, I don't know.

:: 21:07

We're so fast at predicting patterns, right? We are pattern recognition machines, just like crocodiles, show up two times to the water hole at five o'clock in a row. Guess what, on the third day, the croc's going to be there and it's going to eat you. They're pattern recognition machines, and so I pick it up. You guys pick it up Now. Will I automatically discount someone or disqualify them because they wrote that? The answer is no, and the reason is they actually took the time to send me something Like. It shows a moderate amount of initiative and an attempt at personalization and truly, at the end of the day, what they're trying to do is communicate something to me. They're trying to communicate their value to me. They took time to do it. They didn't take enough time, but can I blame them when they've been out of work for four months and they're maxed out on their credit cards and their mortgage deferral program's over and they are desperate and they're pounding this out all day long? No, I don't blame them, it's fine.

:: 22:03

And that's where I was going to go. How much of it is just a numbers game, right? And again, I know we've had discussion on it before. Josh and you're a little different and there's a lot of rules, but I know a lot of people bend the rules when you've got people on the hiring side using AI to screen and keywords to screen, like how much of it is like if I don't, as a job applicant, start using some of these same machine learning AI tools on the application side, I'm just not going to keep up with what's happening on the screening and recruiting side For large companies.

:: 22:38

Yes, now, large companies employ about half of all Americans that are working and small companies employ the other half. So if you're motivated to work for a large company, that is absolutely something that you should be mindful of. How many times the keyword appears in your resume matters. If it doesn't appear enough and if it appears too much, if you're saturating it with keywords, that will show up, that will downgrade you, that will kick you out of the top 10 or top 20 applicants and they might never even read your stuff. So I would say, look, if someone set the rules for the game and they set them differently, you better play that game and play it to win. So fair enough.

:: 23:14

But if you're applying and working with people like me, or smaller companies and smaller businesses, individually owned companies, that sort of thing, I'd say companies, two, 300 people or less then it should be personal. And I'll tell you, while I don't give someone bad marks for sending me some AI written LinkedIn introduction message, I give a lot of props to the person who literally took time to write it themselves and you can tell the difference. And I'll tell you I'm running a search right now for a project manager for a Salesforce partner right Now. I haven't actually done hands-on recruiting in quite some time, but I was like I like PMs. I'm going to work on this one myself.

:: 23:54

Humble brag.

:: 23:55

Humble brag. Well, I don't know if it's a brag or not. I have good recruiters and they stay busy, but now we're really busy. We just got really busy in the last 10 days. So, okay, we want things. Time kills all deals. I'm jumping in. So that's what I'm doing and I'm so glad that I'm doing it, because when I'm going through, like within 30 hours, I had 174 applicants, of which 94 didn't even read the description, because they're asking me all sorts of questions that were in the job description, even though I put in bold note do not apply if blank and blank. They clearly ignored it, didn't read it, whatever. But I could see the people who in the application and on my website they apply and there's a place for notes. They took time to put in a note. Hi, josh, I just want to say blah, blah, blah. Thanks for considering me and, by the way, I have four years of this and three years of that and I've done it here and this is the size of my projects. I hope that helps you in determining whether or not you want to have a conversation with me. Like great, every single one of those people, as long as what they said made sense to the job. I have called every single one of them the other one, so they take priority instantly.

:: 25:07

And guess what? My time is so limited. I mean I'm going to work till 10 o'clock tonight, 11 o'clock tonight, and I still won't get back to everybody that I need to today. I've already got it's Monday and I'm already a full day behind. Have you considered using chat GPT? I have considered. I do use chat GPT, but yeah, that's an excellent point. So the point is that, at the end of the day, real individual personalization will help you stand out from the people who use AI, and the people who use AI will help you stand out from the people who aren't communicating at all. So it's a tiered system and you have to determine just how much energy do you want to put into the thing that might matter most, or second most or third most to you in your world, which is your career and your income, and where you're going to spend 40 or 50 hours of your work life. 50 weeks, 48 weeks a year, right?

:: 25:58

If you put the time in. Yeah, yeah, no, I totally hear that and I I'm not going to out them because I don't remember if it was something that I read on his LinkedIn posts or something that he told me one-on-one. But a friend of mine is a CEO of a, of an ISV and the ecosystem, and either we were catching up or I read this. But he's he's been trying to hire for an associate partner manager kind of role, like he's got a partner manager, but this is like somebody that's coming in underneath because they've got a lot of growth and you know, small company, I think overall like 100 employees, maybe a little less than 100 employees total in this organization.

:: 26:35

And he said, like I posted this thing and like eight hours later it might have been six hours later he had like 1500 applications and most of them were like absolute garbage, like they were. They didn't qualify, didn't have you know, like nowhere close. Like we've talked before, like you don't have to hit 100 of the job requirements but you can't, you can't just hit 10, right, you've got to be, yeah, in the ballpark, right. And and like he was like I don't know what 10%, right, you've got to be in the ballpark right and he was like I don't know what to do, right, how do I separate the wheat from the chaff?

:: 27:05

And so it is a real problem. Use a recruiter. No, it's the simple answer. Use a recruiter because no offense to your friend, but if he wrote a post that attracted that many people and he didn't put in the parameters and the questions to prevent certain people from applying, that he is inexperienced in posting job descriptions and doing recruiting, clearly no offense to him. I mean, look, he's done something right. If he's got 100 employees in his company, so kudos to that. But if he hasn't recruited in a down market like this, he probably had no idea that this was coming and now he's got it.

:: 27:41

Well, now I'm definitely not going to add him on the show, but I will connect you guys if you're interested in talking? Yeah, look, he needs a recruiter. Well, and I don't know like, again, I didn't read the job description. I think it was well written. I just think, people, you know to the point earlier that all these tools make it so easy to apply.

:: 27:59

Yeah, one-click apply on LinkedIn Right.

:: 28:02

Well, not just that, but I mean, like ChatGPT makes it easy to personalize the cover sheet or the interest relator, it's easy to pump this stuff out. I think that was it more than anything else. Well, hey, I'm not really in the ballpark, but what the hell? I've got to get 50 applications out today if I'm going to hit my target and there's just a lot of that kind of scattershot going. I'm curious, eric, and we've talked a lot about it in the recruiting perspective, but from a bank perspective and I talk to a lot of institutions I think I'm pretty sure that this is going to air the same episode where I interviewed Ben Udall, who obviously talks a lot about using AI in a banking perspective. But, like, have you come across like authenticity and I know a lot of the stuff he talks about is more internal than kind of external facing but like, are there any banks you've worked with that are like using AI and putting that stuff out in the wild without it going through some kind of human review?

:: 29:00

Short answer is absolutely not. The even longer answer is a lot of banks, and Ben's unique in that sense that he's been able to move forward and using it for a lot of his creation that doesn't involve personally identifiable information, aka PII, sensitive stuff. He's doing it for internal things, as well as any public facing content, but doesn't talk about customers, products, services, and he's also as you probably heard, if this airs in partnership with that one a big fan of human in the loop and making sure, just like anything else, whatever we produce as a bank has to be looked at by a human, whether that's a marketing person, a product specialist because a marketer writes a blog post about how to get a mortgage, you've got to have a mortgage loan officer look at that to make sure that that's right, because a marketer is not a mortgage person. And then a compliance person is going to look at it to make sure that there's no trigger terms or other sorts of things that are involved. It's that very same process as it relates to AI. You're just using a technological intern of super intelligence capacity to be able to produce that for you, but being able to do that Now.

:: 30:12

That being said, there are a lot of banks that have forbidden AI within the organization and they don't even take a look at the possibility of using it for non-PII, non-private, sensitive, non-public information. They just don't want to use it at all. And one of the things and I think Ben may have even said this on a joint call we had with a client or just in conversation with him but what does a bank have in place that prevents somebody from posting that information out on Facebook or LinkedIn or Twitter or emailing information to themselves at home and then having that information on their home computer? Not a lot of banks have protocols in place to keep that from happening, other than hey, this is a violation of our policy and if you do it you're going to get discipline, probably fired. And it's the same thing as it relates to AI.

:: 31:01

But because this technology is so instantaneous and immediate, it has a lot of institutions, especially at the smaller level, freaked out, because security and cyber and IT and compliance are already overburdened by a crap ton of other stuff, and this is one more thing that they don't understand are going to have to learn and now have to remediate and manage and monitor, and that's beyond you know.

:: 31:29

So that kind of goes back to the article that you had referenced earlier about how 50% of the banking industry is going to be impacted by AI At the community bank level.

:: 31:38

What I'm envisioning is they will hopefully embrace it and recognize that it will empower them to be able to do stuff that they can't do now, because it'll give them that superpower of automation and being able to take care of a lot of the repetitive, data-intensive, mundane, compliance-driven things.

:: 32:00

To be able, almost like the judging, going back into the Olympics review the contract, take a look at the material, just give it a good set of eyeballs to make sure that the stuff you're producing checks all the boxes and you don't have the luxury of excess staff in many community banks, and it's going to make them that much more productive so they can grow and prosper without having to worry about adding staff. And at the bigger level, you know Klarna came out a few months ago with their big announcement that they let 700 customer service representatives go in their call center because an AI chatbot could take all those phone calls. I don't know any community bank that's got 700 people in a call center just taking phone calls, and so you know there are a lot of things in a larger, bigger institution, that will likely be impacted.

:: 32:52

But I think I mean they may not have 700, right. But I mean that's the thing about percentages, it's not hard numbers. They might have seven or 70. Maybe they have, and if they can let go of half of them, I don't know, I don't know if that's realistic point. I still think we're a long way before there's going to be a lot of level of comfort in letting truly generative AI produce content that then goes straight out without some kind of human in the loop, to steal Ben's phrasing, right Like. I think there's a lot of like behind the scenes work that can happen. I think you know.

:: 33:35

But when I think about like AI more broadly, beyond just generative AI, there's a lot of things I've been doing for five or ten years or more. So using OCR to take people's financial records and put it into your commercial lending system and automatically completing all the spreading without somebody having to manually put all that information in. How many people is that going to displace over time? Underwriting decisions, account review decisions right.

:: 34:05

When I was at Options Express, one of the things that we stood up, I mean, there was still a human in the loop, but rather than having a human have to look at every online application and approve it like a licensed principal, approve it for trading and assign it at a trading level. We created a rule set. This was gosh 15 years ago. We created a rule set. We read every account through that rule set. It was presented to the principal on a screen where they kind of scroll through and just look here's the key points of the account. Here's the trading level. Approved, not approved. Approved not approved. These days, some of those could probably be made completely automated with a level of comfort that they're going to be right.

:: 34:44

Well, let me ask you this, then. I do, let me ask you this then Would the global financial crisis have been averted if there was AI back in 2008?

:: 34:53

I mean, there was AI in 2008. No, no, no, come on I mean at scale right. I mean, you don't have a bunch of 24-year-olds writing.

:: 35:02

Let me finish the response. A big part of what caused the destabilization was automated trading algorithms. So I mean that was absolutely a big part of what caused some of the early market swings that helped undercut some of the resiliency and comfort in where the market levels were. So I think the answer is yes and no. Do I think if some of the big banks has set up AI to put better risk controls around their mortgage approvals, do I think that could have helped Potentially? But it all depends on what those parameters were. If they were setting those parameters in a way that they wanted to approve a lot of loans so they could make a lot of money on mortgage fees and on underwriting those mortgages, I don't think it would have helped alleviate it. I think it could have helped compound it. It's not about using the AI or not. I think it's about how you use the.

:: 35:59

AI. I think that's a good point, fred, but it's also how the AI is written, how it's coded, because there is bias in AI. It's misinformation, bad information or limiting one's access to good information, while being tied to the technology. It's like you ask a business person to live without their phone. Ask a high school or college student to live without their phone. Forget it, it's not going to happen. So now there's such a heavy reliance right, and now there's becoming a heavy reliance on AI, and once that reliance becomes absolutely immovable, when everyone must use it and there will be a day likely when that will happen, when everybody must use it and there will be a day likely when that will happen when everybody must use it, or if you don't use it, you're left way behind. You're a Luddite. Have fun splitting logs with your knife in the woods. Have fun with that for real.

:: 36:53

So when that becomes absolutely, when we're 100% relying on AI for certain things, the way we're 100% relying on AI for certain things, the way we're 100% relying on our phones and our pockets, what concerns me the most is information and it not being accurate Because you want to control the masses. Control that. That was proven four years ago and now it's almost getting proven again, so that's a little bit scary. So, when it comes to banking and now we're talking about things that can happen at scale, and you just did, I think, an incredible job of describing some of technology's influence in the global financial crisis back in 1809. You did an amazing job. I'm very impressed with that.

:: 37:35

Yeah, I mean, I think again, it's one of those things where, like AI is a big concept, there's a lot of specific tools inside of AI, some of which are designed for better things than others. So, like everybody thinks about chat, gpt and you know the new sexy LLM generative AI whether it's, you know, image or text or anything like that and by design, those are coded to hallucinate. Part of what makes the magic in the responses sounding as natural as they do and we just had a discussion of it doesn't yet sound as natural as a human, but it does sound a lot more natural than a Google search return is in the uncertainty in what word it's going to pick next in its response, and some of that uncertainty. No matter how much good training data you've put in and that's the other issue is not all training data is good you're going to end up with some bad responses, and so AI has to be deployed with some guardrails on either end, and today I tend to think a lot of those guardrails still need to be human, but some of it can be automated. So, like perplexity, which we all know and use frequently, has some guardrails in that it goes and looks for sources and then it gives you those sources at the end that it used in the response and you can go look at them yourself and say, well, yes, this is right. Yes, this is right. No, this is a garbage website. I don't know why I pulled this in.

:: 39:11

And there's even a form where you can put in and give feedback on what those responses look like. And that could be in an institution, you know, like a banking institution or anybody that wants to stand it up for business purposes. It could be like in customer service. You know we have it generate responses. They don't go right out to the customer, it goes to some customer service manager.

:: 39:33

They kind of grade in how good it is and then over time, there can be some handlers on either side of the AI that can improve it and make sure it's accurate and make sure, like you can go back and trace the answer back to something that sounds, you know, that's based in fact and that's based in what's actual. But we're a long time, in my opinion, from that being prime time for things that are absolutely mission critical being prime time for things that are absolutely mission critical, and we've already seen courts are very open to holding companies legally liable for things that its AI puts out in the public, and so, for that reason alone, I don't think that regulators are going to be any softer on it than the courts have been. I think anybody that wants to deploy AI in a business setting needs to have those buffers.

:: 40:24

The circle of life, it's all coming back around.

:: 40:26

I wanted to ask a question to you guys, because this just popped in my head. I'm curious what you think, let's say, ai gets really good, like really good, when you and you're going to use AI again as ChatGPT just shared with me. It's better to use it for the mundane tasks. Do you ever envision where people are on the front line and, when something gets really complicated, that they then send that? Let's say, in customer service, they send that person oh, this person qualifies for the AI chatbot now because it's way better than us. Are they ever going to be third-tier support? Ai Third-tier support is that ever going to exist?

:: 41:09

I'll go first and just say from my days in risk, I never like to take ever as a timeframe for any type of a risk. Right, Ever is a very long time. Do I think that's happening in the near term? No, I don't think in the near term that the ultimate point of escalation for any customer inquiry is going to be AI. Now what I do think and I can point to some examples where it's happening today you know a lot of things in like a contact center, as an example are like 80-20, right, 80% of your calls are things that you handle all the time, 20% are things that come up very rarely.

:: 41:56

I have seen institutions stand up tools that say hey for the rep, instead of having to go pull the binder out or go to SharePoint and try to read all these policies, we're going to give you a tool internally that helps you surface the answer, but then it still goes through the filter of that rep to make sure what they're saying is accurate, right. So that I think, is happening now in certain spots. I think it's going to happen more widely. I think it is a long, long time before those hard, hard, hard cases go straight through to AI. I don't know, eric, your thoughts.

:: 42:32

I think AI trained and empowered second or third tier individuals. I mean, I can see you know Mark from our team uses Claude quite often for programming and other sorts of stuff when our websites aren't doing what we want it to do. Module isn't performing. We want to do something that is outside of the box. You have to have someone running the machine that understands what the objective is and understands the machine and can work with it. It's almost like you know in in avatar where you've got these humans that jump into these giant robotic. You know you're sitting in the head but you're still controlling the arms. But it's making you walk faster, it's, you're more powerful, but you still have to understand the basics of hand-to-hand combat. It just amplifies your capabilities. You still have to understand HTML and JavaScript and programming language and customer experience and functionality and banking. You have to know all of those things. But now you can use that in conjunction with an AI. That makes you that much more powerful.

:: 43:43

And to your point, fred, having a bot that's trained inside of your organization that gives you fast access, but also knowing enough about your organization so that if what you get access to smells fishy and you know that this might be a hallucination. You can check yourself before it comes out of your mouth, and it still may not be 100% of the answers. But of the 20% that needs to be escalated to that 80% of those will be able to be automated through this interface and then 20% of the 20% might need a. You know what, fred, that's a really good question. I don't have the answer for you right now, but let me do some research. But that only happens a fraction of a fraction of a time. Yeah.

:: 44:31

Yeah, that you know that's more reality. It's another good use case for those. That's another good use case for those handlers, right? So you get the answer, but you get the answer back with a citation or maybe it even surfaces the actual policy. You know, here's the response and here's the policy, right? Like there's a lot, of, a lot of room for that.

:: 44:49

You know, we started the conversation on the banking article that Forbes produced about the AI taking, you know, 52% of jobs. And then you said and it made me think of the article, fred, because you said you never say never or you don't say ever, you're not putting a timeline on something. And then what, eric? You just said something, or what did you say something about? Something seeming fishy. Well, both of those thoughts came into my mind when I was thinking about this Forbes article, because 52% could 52% could be replaced when the next 200 years.

:: 45:28

What's the timeline you?

:: 45:29

know? Okay, when, what are the chances? What's the solid fact, likelihood? How did you measure that? Who figured this out? How much of AI did you use to calculate this? How much of AI did you get the answer about AI? Is there self-interest in any of the algorithms that are being used to come up with this? Because you have to look for those bias, and I believe that there's going to be bias in machine large language models, in LLMs. There's going to be bias. We already know that there's bias. It's been proven right and we've already cited it on this show before, so I'm not going to bring it up again, but you know so when you take an article like this and they use the word could and could is the hinge of the entire whatever 600 words it's like, okay, yeah, I mean that's, that's every, that's every analyst report ever written.

:: 46:27

I've never read an analyst report where they say mark it down in blood, this is a hundred percent gonna happen.

:: 46:33

Okay, but you know, analysts have, it's just not how that industry works right, but analysts are wrong half the time. That's been proven that it has been Right, but analysts are wrong half the time. That's been proven.

:: 46:42

That has been proven At least.

:: 46:43

At least half the time, and so why pay attention to the analysts at all if it's a flip of the coin?

:: 46:48

Yeah.

:: 46:50

Well, I mean yeah, I hear where you're coming from. It's thought-provoking, I think, is the long and the short of it, and I think it's one of those things where I think institutions need to look long and hard about how they can leverage technology. I think individuals are to look long and hard about how much value are they bringing to that role and if it's something that feels like it could be easily automated, maybe it's time to upskill, reskill and figure out how to be on the winning edge of that, because it may be 54%, it may be 75%, it may be 10%, but there is going to be a percent that can be automated out. It's already happened and I think the answer is just to be as prepared as possible. I do want to bring us full circle. I feel bad, josh, for you on the East Coast. We've gone well into the evening. Eric and I are in earlier time zones today, but we are getting up near close to an hour. Long takes. Before I close it out, we're in very long takes.

:: 47:45

Welcome to long takes.

:: 47:46

We're not going to get fully into it, but where are you guys going to be? Josh, I know you and I will be at Dreamforce. Eric, I'm assuming you're not going to Dreamforce this year?

:: 47:56

I am not going to Dreamforce. I just returned from the graduate school of banking in Madison, wisconsin, which was an awesome visit. Once again I'm in Minnesota and we'll be presenting although when this drops it'll be done their bankers conference, and then I'm off to Springfield, illinois, and then to Missouri for the Midwest Independent Bankers Community Conference. And then I think the next one is the Pacific Coast Banking School in Seattle. So any of my West Coast bankers that may be out there, if you're going to be in Seattle attending PCBS hopefully you're taking my elective the second week, but if not, still pop by and say hello.

:: 48:35

All right, Josh. So I am going to be at Dreamforce 16th through the 20th or whatever it is, and then a couple 10 days, 12 days later I'll be at Florida Dreamin' in Clearwater, Florida, and then later in October it's not guaranteed yet, but I'll probably be at Life Sciences Dreamin' up in Philly, which is later on in the month. So it's going to be a busy five, six weeks here for me, which is about all the travel. I do for work. Lots of dreaming, going on, lots of dreaming, lots of dreaming, lots of dreaming.

:: 49:08

Very, very, very cool. I was remiss this year. I made a rule for myself that I was only going to go to dreaming conferences where I spoke and that I've not had time to submit a single submission so far this year to any dream and conference. So no dream and conferences right now on the docket for me. I will be at Dreamforce, looking forward to it, Definitely working on putting some some dinners together, some networking dinners for for institutions and, Josh, we should definitely talk again about having maybe a joint event or something.

:: 49:41

I'd love to do that Our podcast audience. Yeah, and I got to figure out how to sign up for that cigar bar party at the end too. This time I plan to not get COVID. That's my plan.

:: 49:50

I send you the link. I'll send you the link for the cigar party. It's a phenomenal way to end the Dreamforce week. Other than that, the next conference I'll be at is the Go West Max conference in October in Denver, so if anybody's attending that, please come check out my session.

:: 50:08

Actually that might be incorrect. I think you and I are going to see each other in Cleveland, at Macon, if I'm not mistaken.

:: 50:14

Oh, that's right. When is that?

:: 50:16

That is September 10th, 11th and 12th in Cleveland the Marketing AI Institute. If anybody listening is interested in attending that event, I do have a coupon code of COOK200. That will save you $200. And any one of the registration options. I think they're looking at close to over 1,500 participants and the speakers are on their website. I know we jaw a lot about AI and then they do it way better than I ever could, and so Paul and Mike and their crew are putting on, I think, the fifth annual Maycon event. It's going to be pretty awesome and I'm looking forward to getting some IRL time with you in Cleveland.

:: 50:56

Sounds fantastic. I'm definitely looking forward to that. One of these days, we're going to have to find a conference that all three of us want to go to, and then we can do one of these in person.

:: 51:04

That would be awesome. I think that would be awesome. Or just a golf trip.

:: 51:08

That'd be fine too. I'd be down with that. Well, gentlemen, it's been great. Look forward to talk with you all in two weeks and take care.

:: 51:14

Excellent. Thank you, everybody.

:: 51:15

See you. See you soon, bye-bye.

:: 51:18

Bye. Well, folks, I hope you enjoyed episode 31 of Banking on Disruption. I think I delivered on everything I promised during the intro. That was a phenomenal episode. Don't forget, if you didn't catch it all. You can find show notes and a full transcript of the show on our website, bankingondisruptioncom. And remember, new episodes drop every other Thursday. So we'll see you in two weeks and in the meantime, don't forget to follow us on LinkedIn and Instagram at at Banking on Disruption. Until next time, this is Fred Cadena, wishing you success in your digital pursuits.