Embracing the Double Edged Sword That Is AI

If there’s one thing that businesses everywhere are clamoring for, it’s AI. Industries of all kinds are looking for ways to implement it to serve their business, whether it’s chatbots for communicating with customers or using Chat GPT to help generate answers to questions. Of course, the financial services industry has picked up on the hype, and many financial institutions have been quick to embrace the technology. Credit unions have also been jumping onboard, but for some, they still have questions and are choosing to wait to implement AI, or in some cases, ignoring AI altogether.

Host and Co-founder of The Credit Union Connection Sarah Snell Cooke sat down with Jake Tyler of Glia to discuss AI and how credit unions can find a balance between making the best of the advantages while also dealing with the risks and challenges.

Read the full transcript:

Disclosure: Transcript is automatically generated

Sarah Cooke 00:02
Hello and welcome to The Credit Union Connection. I am your host, Sarah Snell Cooke. I am here today with Jake Tyler, who I am told is Glia's AI guy. Is that true?

Jake Tyler 00:13
I am an AI guy, I guess. Many, many AI guys at Glia and girls. But you know, it's certainly an exciting place to be at the moment.

Sarah Cooke 00:23
Yeah, so give us a little background on what you do at Glia.

Jake Tyler 00:28

Yeah, sure thing. So I came to Glia through a bit of a interesting journey. So I started a company that built AI virtual assistants for banks and credit unions called Fin AI. I was one of the founders and the CEO of that company. And we were doing this way before AI was was cool, and way before it worked that well as well, and, and we battled, you know, we battled it out there for eight years or so, and it got a lot better as we were building that product. And in 2022 we, we sold our company to Glia. So, and we did that because, and then at Glia, I still work on AI. So that's why I'm the AI guy, and I work on it with lots of other AI guys and girls who, part of whom came from my company, and part of whom we've sort of grown into as part of Glia. So, so that's where we came from. Glia is a platform that is reimagining how credit unions connect with their members. So you can think of Glia as a, as an interaction management platform, so however your members want to interact with you, whether they're chatting and digital banking, whether they're calling up and chatting to the call center or texting in, that can all be, go through Glia, and it can be handled by a human or an AI, and it all you know, so some of the magic of of Glia is being able to really seamlessly move between channels and between AI and humans, and so that's how we fit into the bigger Glia picture here.

Sarah Cooke 01:57
Oh nice. Yeah, so let's, talk a little bit about, let's go head on into one of the bigger questions.

Jake Tyler 02:08
This is good. Always a great tee up.

Sarah Cooke 02:09

I'll start out little, then go to the bigger ones. Alright, so what are some of the practical applications of AI? Because you hear credit union executives and board members running around saying, we need AI, so what are the practical applications of it, and explain how that is used and how it could work to modernize the call center, for example, like you're doing?

Jake Tyler 02:32

Yeah, so let's you know, I'm not going to pretend to be an expert on every application of AI, so I'll keep it narrowly focused on what we spend our time doing here at Glia, which is, how does it impact frontline teams? How does it impact member interactions? And so there you can, you know, there's sort of three places where we're deploying AI today at some scale, and where we think there's huge opportunity as well. The first is automating member interactions. So whether a member calls up that, they're chatting in digital banking, if they have a sort of simple routine inquiry, which is probably most inquiries you're getting, we can understand what they're asking and either automate that, like help them self serve or get them to the right agent. Then the second is introducing new AI tools and capabilities for agents. So that's augmenting your frontline team members with AI. So that might be something simple, like summarizing an interaction whenever a transfer is made. So something pretty tactical to you know, at the end of every interaction or call you have to do post call work like write up what was the call about? We can automate that. And so there's a lot of, like, very tactical applications of AI, new generative AI tools that combined, can really streamline agent workflows. And then, so that's, you know, member AI, agent AI, and then the third would be AI for managers. So, you know, tools that would help you do quality assurance, tools that will help you provide better coaching to your agents, tools that will help you onboard. You know, everybody's struggling to hire frontline team members at the moment and train them up. And churn is a big issue. So tools that can help you train online team members and and make your you know, speed up the time for a new, new team member to become a great team member. So we think there's, you know, when we're looking at what we do for customers, there's three big areas, and there's huge opportunity in all three of them at the moment. So we're building tons of product across all three of those.

Sarah Cooke 04:34
Yeah, as far as transcribing those phone calls, oh, I haven't found a good one yet for the

meetings.

Jake Tyler 04:41

Yeah, I mean, I mean, look, transcription is hard, even for a human to do with accents and background and someone yelling at their cat and stuff like that. The question is, you know, because you know, we do this in production, just to go down this transcription rabbit hole here for a second. But, the question is, you know, does it need to be perfect, or does it need to be good enough to provide some useful like, quality control assurance, and so, you know, we're transcribing calls in real time, and then we're able to do that post-call work. We're able to have an assistant that can, you know, listen to the call and then prompt agents. So like, hey, you know, this member seems frustrated. Be careful, or like, hey, here's our, they're trying to order a new check. Here's how they order a new check. So there's all sorts of ways that you can use that, and you don't actually need transcription to be perfect, although there is obvious, every time you look at transcription, you'll undoubtedly see, like, some insane, bizarre thing that it's said.

Sarah Cooke 05:42

Yeah, well, in what I do, you know, a lot of times credit unions ends up being transcribed as Koreans. It's just, yeah, one of those things that it catches it up. But, so one of the things that you are, or and Glia are very big on, and I don't want to speak for you, but is the responsible use of AI. So how, what is, what is responsible AI?

Jake Tyler 06:08

Yeah, well, so, you know, we've sort of got right into this, but, you know, the reason we're here talking is because AI is this really big technology platform shift that's happening and, and it will be a very significant platform change. So everybody wants AI. We will all need AI. AI will significantly change how members interact with credit unions and how frontline teams at credit unions do work. So there's big changes underway. The challenge is it is a little bit like cowboy country out there at the moment. You know where you're still early on in the market. You don't know exactly what's going to work. Generative AI has all these new risks. Like, if you've used Chat GPT, it just makes stuff up sometimes, and it's hard to tell if it's making stuff up, unless you're an expert. You know, where, where exactly is your data going all the time with these different services? Sometimes you'll have, so if you're experimenting yourself, they're things to consider. You might have vendors who have new AI tools, but then, like, what are they using in the background? Where's that data going? So we are a little bit in the Wild West, and like a Wild West times like this crazy hype party, I guess, is like the sort of thing that's going on. So responsible AI, you know, as credit unions, we are clear, we only serve banks and credit unions. So we have 300 plus credit union customers. We deeply understand the industry, and we know that. You know, while we are excited about technology and excited about AI, we also don't want to introduce a ton of extra new risk, and we want to be very careful about data, and we don't want to make up an interest rate when we tell the member what the interest rate is, right? Like these things, this is not creative writing. This is, these things need to be bright. So for, you know, when we think about responsible AI, we want AI, but we want responsible AI, that means it's safe, first and foremost, your data is safe where it's not being shared somewhere that you wouldn't want it to be shared. You're not giving away PII. Also, it is turnkey, so it doesn't take a huge amount of time to build. It's not a big Professional Services Project. It is a product that you can turn on and that works, and it delivers proven results. So AI is solving a real problem, not just a cool demo. And again, you know, because we just work with credit unions, we can be very focused on, on, on really understanding those customers and those problems, and building tools using this technology to solve them.

Sarah Cooke 08:42

Mm, hmm. And so, as you mentioned, solving real life problems, what sort of, or do you have any data on, like, what credit unions are saving, as far as in in time or in headcount, or whatever, however they want to measure that?

Jake Tyler 09:01

Yeah, I'll give you, I mean, it's obviously very dependent on one credit union to the next, but I'll give you some some data points on some of the capabilities we have in the market today that are being used. So one is just on the member automation side. So we have, that was the core of the business that I started and and sold to Glia. We call that our Glia Virtual Assistant now. And so that is, you know, answering calls. It's answering chats in digital banking. And it's, it's automating those simple inquiries. And typically, we'll automate, you know, 50 to 60% of inquiries. So a pretty significant impact, a little bit less than the phone, because people tend to hit zero and go straight to a human in the phone a little bit more but, but still, you know, if you're getting 10,000 calls a month, that's a significant, significant impact, and importantly, not just having an AI where that's getting in the way. If you want to chat to a human, you can just chat to a human straightaway. So it's just there to help if it can, and get you to a human if that's the best path. And then on the agent side, you know, what we've seen is that the suite of, this tends to be like more things that we can do to help, but each of them has less sort of individual value, if that makes sense. Yeah. So one example of just like a very clean time saver is, is doing this post call, wrap up work. So if you're an agent, you get a call. At the end of the call, you hang up, you have to answer a few questions about what that call is about, like, what topic, was the member frustrated? You know, maybe some other QA questions. Typically, agents spend around 90 seconds doing that post-call work, which your average call time is like, six to eight minutes. 90 seconds is a lot, and we can automate all of that with with AI now. If you make a transfer, and you have to transfer from one agent to another, you might have to write up a summary of what the calls are. So we can automate all of that. So there's just some, a few examples of of time savers. I think what we found more than saving time, which, to be honest, was my going in assumption of like, where the value was going to be. I think people have found a lot more value in being able to, like, have more personalized conversations. So take the transfer summary one. The value is not in saving time. It's in ,when you pick up, you know if it calls being transferred. So when you pick up a call, you don't have to say Hi there. How can I help you? You can just say hi, like I read the the transfer summary, I already know what the call is about. Hi there. I see you having an issue with your credit card. Let me look that, look into that. And so that's been way more value in that which is less tangible and easy to put down in simple math, but way more value in that than in the time saved.

Sarah Cooke 11:47

Huh. Interesting. No, it makes sense. So AI, obviously, you know, like you said, it's kind of the Wild West, is double edged sword and advantages. There's also a few dangers, as you mentioned. So what is the proper balance when credit unions are considering those two sides of the coin, that they should keep in mind?

Jake Tyler 12:09

Yeah, so if you take, if you take, you know, so I guess, like, what's irresponsible AI, if we put that hat on for a second. I think you got sort of two flavors of irresponsible AI. One is to put your head in the sand and just say, look, nothing's going on here. This is all going to be a big nothing burger. Don't worry about it, and, and we're not going to do anything. I think that's very irresponsible as well, because the world is changing. I think the second flavor of irresponsible AI is to, like, rush into it, do things that are untested, put those things into production. I mean, it's great to experiment, but, or work with partners who are doing things that are untested and, and so whether you're building yourself or you're experimenting with new partners, it's just about doing that in a responsible way. So, I think they're the two main flavors of irresponsible AI that we see, probably with like, equal balance, to be honest, and so responsible AI is trying to, like, you know, thread the needle there between those. So we want to change. We realize we need to change, but we also need to focus on doing things that work, and we need to focus on doing things properly so that it's a good experience for members, and it's safe and secure and that sort of thing.

Sarah Cooke 13:22

Yeah, yeah, absolutely. Because, I mean, there have been a number of concerns about, you know, FCRA issues and other ways that credit unions could get into trouble with AI. So, yeah, very importantly, I guess it's just, I don't want to say it's general vendor management, but it's just understanding, really, what you're getting into and who you're partnering with.

Jake Tyler 13:42

Yeah, yeah. I think that, like, particular challenge with AI is that you, like, you use Chat GPT, and you're like, and you sort of think it's magic and, and it does, it feels human. And so you're like, Oh, you just like, carry it a little bit further and be like, Oh, it must be easy to do this. And so I think, I think, you know, what we see is people tend to get into a bit of, like, see a great demo and extrapolate that out further than it should go, or think that it's easy. If they see a demo and like, I'll just build this myself, or, you know, this is, you just plug it in and it'll work like that. And actually, I think the challenge with a lot of AI projects is that, like, getting 80% of their way there can be quite straightforward. But man, getting the rest of the 20% is super hard. And that's where, like, all the edge cases are, where all the like, importance and value is. So it's almost like, it is a double edged sword with the technology. It's like conversational, it feels human. But then, just like we do with pets, we like, we think it's, you know, we just ascribe human equalities to it, and think it can do things like humans can do. So, so that's a, that's sort of the double edge of that thing.

Sarah Cooke 14:51

Yeah, no, I busted my intern last summer using Chat GPT to write stuff. I was like, it's me learning this first. Yeah. Yeah, you gotta learn the fundamentals before you, you know, become a Chat GPT prompt writer.

Jake Tyler 15:07
That's right, yeah, but it is, you know, it is important to be learning, though, that that is, I think,

for sure, and experimenting, but not, not, but not going too crazy with it, yeah.

Sarah Cooke 15:18

Yeah, yeah. So what are some of you mentioned the two types of irresponsible AI, one of them being putting your head in the sand. What are some of the most common objections that you might hear from banks and credit unions? And how do you quell those?

Jake Tyler 15:36

Yeah, I think, well, like, you know, just to segue way off the last point. A common one, is like, you know, I'll just get this myself. I'll plug in Chat, GPT, or some, some sim, you know, some similar system, and it's easy, right? Like, it's a very powerful system, the AI will handle it all. And, you know, having built products like this for a long and maybe that'll happen at some stage. I don't know. It's, it is not the case where that happens today. It's like, it's a lot of work to get it from, to get it actually working and solving problems in the real world. There's a lot of complexity there. So that's one that, like, I'll just DIY it myself, type approach. The second, like, very, very valid concern is, is, where's this data going? Like, how do I know this is safe and secure? What's the LLM you're using underneath the hood? Where does my what, what's the data flow? Like, very, very valid, like, area to dig into, I think. And if you're not digging in there, like you're you totally should. I mean that. And then the third is, is like, is this really solving a problem? Does it work? Like, because generative AI does make stuff up occasionally. Like, doesn't matter what sort of guardrail you put on it today, there is some degree of hallucination. And so, like, will I accept that for this particular use case? In our case, we don't generate answers and, and serve them up directly to members. We would only ever use generative AI for internal facing use cases, so there's always a human in the loop before an answer is sent out, and even then, we're very careful about how we use generative responses. So, so, you know is, is it working? Is it solving a real problem? You know, as another example, we, we, you know, we've been experimenting with all sorts of things and, and one of the things we tested was a sort of generative AI assistant for agents. So it would like read a knowledge base, and it would suggest answers on the fly to agents. And, but, you know, it did have some hallucination in it, and what we found was actually like it, because it wasn't right all the time. Agents had to double check everything we said, and it just, it wasn't saving any time at all. It was adding time. So I think you've gotta like, it can look really cool, but is it really solving, making things easier for people, or are you just adding another, adding, adding noise, you know?

Sarah Cooke 18:05

And that's not, I mean, to your point, a lot of times it's like CEOs or VPs or whoever making those decisions at the top, and they aren't the ones that are going to be playing around in it, you know.

Jake Tyler 18:17

Totally. Yeah, I mean, a lot of the time it is like chairman has seen something, found Chat GPT finally, and is like, sends it over to everybody and say, hey check this thing out. Let's do it. And, yeah, I mean, obviously it's a lot of work to go from that to something that solves a real problem, but getting into the weeds on those workflows and making sure you're really saving time and not adding complexity and time is all the hard stuff and is the valuable stuff as well?

Sarah Cooke 18:45
Yeah. And so what's the future of AI look like specifically for financial institutions?

Jake Tyler 18:51

Yeah. I mean, I think that the, you know, if you look at how members interact with credit unions, it hasn't really changed dramatically in the last couple of decades. Like, maybe we do it in digital banking now, but mostly it's phone calls and the systems, and mostly it's talking to humans. And if there is an automation system involved, it's a Touchstone IVR, and that Touchstone IVR was built in the 90s. I mean, it hasn't changed since then. And even though, like, I mean, the telephony system you're using is basically the same thing. I mean, it's moved from on prem to the cloud, but it's still like it's doing the same thing. It's not very well synced up with what's in digital bank. So, you know, we're moving from a world where that, that whole operation, hasn't really changed dramatically in a few decades, and this, this wave of technology, is going to completely change it, like, you just like, the who you hire, how many people you hire, how people interact, where they're interacting, that whole technology suite, those frontline teams, the size of those teams, the work that they're doing, the skill set you need. That's all going to change.

Sarah Cooke 20:01

Yeah, they're kind of...

Jake Tyler 20:02

This is probably Jake's personal view more than Glia's personal view at this stage. But that's that, you know what, from what I'm seeing, that's what I think we'll have like we are set. We are coming into a decade of meaningful change, from a period where most contact center managers haven't really seen that much dramatic change in the way that they interact with members.

Sarah Cooke 20:23

And that really raises the HR issue, right? Like, who's going to stay, who's going to go, because they just can't adapt, or whatever. So I think that, yeah, there's a lot of really interesting implications, I think, for that.

Jake Tyler 20:35

Yeah. I tend to, don't, I mean, the whole like, is AI coming for a jobs thing, I tend to think, is like is not the AI, like, for sure, it will change the jobs. I tend to think, as AI automates the simple things, we may actually put more of a premium on human connection, including, you know, credit unions are built on human connection and being in their communities, right? So I think it may play to the credit union strengths, and it will actually increase the importance of human connection, and people will want to do more of it as a result, but you won't be doing simple things. And even when you are chatting to a human, it'll be very augmented by AI, so it'll still look, you know, look and feel very different. But I personally don't think we, we want, like, we still want to talk to people. I just, I don't think we'll, you know, if I want to check my balance, I don't need to talk to somebody. But if you know, there's lots of situations where I do need, like, some counseling as well as some information, you know?

Sarah Cooke 21:34

Yeah, no, I was on the board of my credit union, and I never walked in it. So I would go in and nobody, oh, you're a board member because in the, in my profile, but yeah, I mean, you just self service. I think, you know, not only do credit unions save money and time and all that, usually with that, but you know, people appreciate it too. Especially, I feel like Gen X is like, just leave me alone, yeah, and let me do my work.

Jake Tyler 22:03

I think it'll, I think it'll barbell. I think you'll want to self serve, and we'll want to move lots of stuff to self service, but then at the same time, the other side of the barbell, you'll praise, place a much bigger premium on human connection.

Sarah Cooke 22:16

Mm, hmm, exactly, when, you know, say, if somebody's getting thier first mortgage, and they need to understand what they're getting into and that kind of stuff. It's definitely important. Yeah? So anyway, we've gone from very small to very big picture.

Jake Tyler 22:31

Yeah.

Sarah Cooke 22:31
I'm going to, as I do with all my guests, I'm going to give you a final word. Final thoughts, what, what do you have to say Jake?

Jake Tyler 22:38

Fine, final what that, I mean, we, I think the biggest thing to take away is, like, this is going to be a big change, and we don't need to rush into it. Be careful in the way you tread into it or walk into it, but, but you probably do need to start getting into it, and, and, and there are things to do today that work and deliver value. Talk to your team, talk to your, you know, your trusted partners. But I would very much encourage people to get started on the journey. And my guess is things will accelerate more than decelerate, just if you look at the pace of change over the last couple of years. So, that first version of irresponsible AI where you put your head in the sand, I would counsel against that. I mean, obviously I have a vested interest in that. But, but more, more than that, I think that we're entering a period of probably rapid change for frontline teams.

Sarah Cooke 23:37
Yeah, awesome. Well, thank you so much for time today, Jake, I appreciate it. Yeah.

Jake Tyler 23:42
Thank you very much. Thanks for having me.