Into the Future: Credit Unions & The Next Wave of AI - TRANSCRIPTION

Webinar 1

Into the Future: Credit Unions & The Next Wave of AI

Come join us and read the transcription of Buildable’s first webinar! Discover game-changing strategies as top thought leaders reveal how credit unions can harness the power of AI for growth and success!

 

Brett: Thank you very much for joining; we're really excited. This is Buildable’s first webinar. We're kind of going between this webinar-slash-video cast – maybe podcast someday. So, our first one is really focusing on the new wave of AI and credit unions and, what are some of the strategies and things that they can utilize from AI. So that's gonna be a lot of our conversation today.

But really the premise behind this series is we have a lot of conversations; we listen to people's dreams and hopes, and software they're looking at building, and their ideas. We want to start bringing on thought leaders, and things that we think about, and make it really like a fireside chat. And I guess since we're getting close to summer, maybe a bonfire chat.

I'm Brett Wooden. I'm the facilitator today. I've got 25 years of experience in the credit union space. I started some CUSOs and worked on projects way back in the day. Kind of my claim to fame in the credit union space was I developed an application and also utilized iPads to open accounts. And so that was really exciting. Where I'm at today, I'm part of a software development team, which is really cool. And so I'm excited just to ask questions and facilitate this.

Josh Grigsby, he is going to be our person behind the scenes. So he's going to be managing the chat. Feel free to chat it up with Josh, ask questions. We will be opening it up at the end for about 10 to 15 minutes of questions as well. But Josh is really going to be kind of navigating behind the scenes. And he's also an awesome person, highly intelligent at Buildable as well. So excited to have Josh managing behind the scenes.

So, with that said, let's go to the other Josh. And Josh, tell us a little bit about yourself.

Josh: Yeah, so I've been in the software industry for, I guess, a little over 10 years, like in terms of actually working, but, I've been a computer guy since forever – you know, just part of that generation. I've spent a lot of time at bigger companies doing cloud-based engineering. So, it’s like if you were to go ask Google now the best way to build a scaled, real-world production application, they’d probably tell you some form of building in the cloud on AWS, or Google, or Azure. So I've spent a lot of time developing those kinds of applications.

In general, you know, I've always been really clued in to how to make the individual, the developer, and the team work – not only efficiently, but how to empower them to create great products. And so that's kind of one of the main reasons that I've really been dialed in on a lot of this new wave of AI coming, is because it represents such an incredible opportunity to empower individuals and organizations. I really see it as a tool, not as a replacement.

So, you know, I just came off of a paternity leave a couple of weeks ago. And, you know, I've got a lot of baby taking-care-of to do, but a lot of my time is also just like hacking around using these new tools. I have LLMs plugged into just about all my tools that I use, just so I can use them as frequently as possible to try and find those use cases that really allow me to step up my game. Then I can share them with people around me.

Because I think the jury's still really out on how to use these tools. And so it's really a matter of just getting comfortable with them, trying things out, and then consolidating those kinds of lessons into maybe more formal applications. But that's kind of, you know, why I'm excited to talk about this and why I spent so much time learning about it.

Brett: Well, thank you for joining us. And then next we'll turn it over to Miles. So Miles, tell us a little bit about yourself.

Miles: Yeah, hey everyone. Nice to see you today. My name is Miles Oliveira. I am the Sales Director at Buildable. I've been with the company for seven years and I actually used them as a vendor in a previous role. So, I've kind of been around the company for almost a decade.

Most of my role is really understanding a lot of what clients are trying to accomplish; what the best solutions might be for their specific use case. As Josh was mentioning earlier, it's how to use AI and LLMs is a lot of what we're going to be talking about today. And my role is really to kind of help facilitate that conversation and understand a little bit more of kind of what the best way to do that is.

My background comes from the insurance industry as well as a little bit of time spent in some marketing and obviously quite a bit with software as well. I'm really looking forward to our conversation today and seeing how we can utilize these LLMs and AI tools for financial institutions.

Brett: Cool. Thank you, Miles.

 

“By far the greatest danger of AI is that people conclude too early that they understand it.” – Eliezer Yudkowski

 

Brett: So, our first question is for Josh. Josh, just kind of give us – not the Wikipedia version or the ChatGPT version – give us: how do you define AI or LLMs when someone asks you? Like, let's say they would ask you on the street.

Josh: Practically speaking, AI is capabilities that rival or are better than what we expect humans to be able to do. In the past, computers have been really good computing machines. You can have them do math and they're good at that, but in the past, you haven't just been able to talk to them like you would to a person and have them understand you. You haven't been able to give them a picture and have them know not only what's in it, but also how those things relate to one another.

So, you know, in the past year or two, probably a little longer than that, we've started to really see AI computers be able to engage with the world and with the world of meaning and words in a way that previously only humans have been able to do. And, you know, there's always the slightly scary part of like, when can they do better than us at certain things? But we also want that, right? Because humans can only understand so much complexity.

We really can use them as tools to consolidate complexity and understand more than we could just using our inbuilt biological skills. And in this case, LLMs represent kind of this latest wave of AI, which is like categorized as generative AI. You have not only being able to capture information, but being able to create information. You can create images, you can create text – as anyone who's used ChatGPT can tell.

We have these boundaries that are totally getting pushed back regularly. Like, oh, you know, computers can never do that. And so this recent wave of AI has been pushing that boundary back at a very fast pace. And so, I would say AI is anything that can start doing what we previously thought only humans could do.

Brett: That’s cool. Miles, do you have any additions or you have any different thoughts – thoughts in the way you define AI or LLMs?

Miles: I think Josh really nailed the kind of value proposition of utilizing an LLM in your kind of organization. The big one is being able to interact with data like a human being, right? How many reporting tools do you use which are really just a set of filters and searches that really kind of don't give you the full breadth of having a cross-functional kind of team? Being able to look at multiple different sources of data and combine that information together like you would if you were having a conversation around a conference table about what you might want to be able to do? I think for, for me, utilizing AI and kind of defining AI is the ability to interact with very valuable data like a human being.

Brett: It's been interesting, with the financial industry, I'm a constant reader of posts and articles that you see on LinkedIn and there's an interesting dialogue around that financial institutions have tons of data. And it is interesting, I would be curious to see how financial institutions like a bank or a credit union or even a fintech would describe AI and use of their data.

 

“AI is anything that can start doing what we previously thought only humans could do.” - Josh Melander

 

Brett: So, Miles, since we've got you kind of already chatting a little bit, you know, what are some examples? So, you do a lot of meetings with financial institutions and then you meet with those clients outside. What are some examples you've recently seen in different business sectors utilizing these LLMs and AI?

Miles: It’s been really interesting to see how creative people have been about trying to access their own data. A good example would be the best way to utilize AI within your organization is to be able to think about the questions that you can't answer right now, but always want to ask, right? These are, I call these kind of like the in the shower questions of like your great idea that you're having, which is like, ‘Oh, right, yeah, we should try to do that’. Or ‘We should try to do this’, or, ‘Oh, I always wondered about X, Y, and Z’. That's a great place to start.

Some of my clients, especially in the healthcare space, utilize a lot of information, but they're limited to the tools that they have. And they also have kind of, they're also in a highly regulated industry as well. So, a good example would be being able to identify, categorize, and match a patient's life cycle through your healthcare organization based on their disease type, their age, their location, their background, that sort of information.

All that information exists in disparate parts throughout a healthcare organization. It might be in your chart. It might be in a database. It might be in another app. There’re all these kinds of different data sources. Being able to put all that information together and actually create a clear picture of, if someone has hypertension and they're age 55 to 65, and they are from this zip code, and they're male or female – and let's see what typically, based off of the data that we have as a healthcare organization, what will their life cycle be over time? It's a great use of AI in order to take all of that slightly unstructured data, data from a bunch of different sources, apply that data set to the model and allow it to map that information out over time.

And then you can start adding triggers on that as well. Well, if they have a comorbidity, right, let's say they have something else going on. Let's say they’ve moved, they have another event in their life that changes some other things. What are the different things that commonly come up? So, you can not only map that information, but you can also use that as a treatment plan for doctors to say like, ‘Hey, typically this might end up happening if a patient's five years into having hypertension and their symptoms are not changing, this is something that will typically end up happening’. So, you can start getting out ahead of these sorts of things.

And that example I also think applies directly to financial institutions as well. When you're working with your membership, you want to look at, ‘Okay, well, here's the breadth of the community that we support, this is the breadth of our membership’.

But they all have different needs and at different points in their life cycle. So, understanding the membership data and being able to ask questions like, ‘what is my most common customer segment doing from age 30 to 40’, right? ‘What products do they have with us?’ ‘Where are they living?’ ‘What are they doing?’ ‘How are things increasing, decreasing, staying the same?’ ‘What are their most common spending habits?’ ‘What are they inquiring into our institution about the most?’

And you can utilize a lot of this data that comes from these different places, but utilize that data set, apply it to the model to understand a lot more about how people are actually utilizing your institution and what's important to them. That's something that, from an industry standpoint, that's what I've seen quite a bit, is taking that data that exists in multiple different places, applying it together and really gaining that kind of, I'm in the shower understanding of like, ‘Wow, I didn't know that our organization, like this is how people did this’. Or, ‘I didn't understand that’. And the data, you know, doesn't lie. It's very straightforward.

But like, ‘Oh, this is actually how they behave’. And these are some opportunities we can use to go ahead and say like, ‘Well, maybe we should start reaching out to folks about this certain time’. Or, ‘This is what's important to them’. ‘How can we increase deposits, because this is what their main spending category is.’ ‘Should we offer a gift card for that sort of thing?’ All those kinds of deals there, it really makes a huge difference.

Brett: I think you bring up a good point. What’s really funny is it makes me think of, gosh, in like 2013 or 2014, I was on a project. But we did this auto loan idea where we took all the data from the credit union and said, ‘Okay, the typical, like the age of an individual, what type of car they bought, the cost of the car, the payments’.  And what we did is we matched up when a person was looking to buy a car, we said people in your demographic that currently bank at our credit union buy this kind of car for this kind of payment. And so utilizing, you know, things like that, I mean, not going too creepy.

I think like Target at one point utilized a data target of buying habits. And a dad one day went out to the mail and got this flyer that was offering baby supplies, you know, like diapers and that. And it came out that he's like, ‘Why are you marketing to my daughter?’ And, she was in there buying pregnancy tests, those kind of things. And so, AI knew pretty much what was going on before even the family did.

I love the life moments, you know. As open banking and those type of things start coming in the space it's going to be important as people start moving away from the transactional side and getting in those life events.

Miles: And my kind of follow up to that one, Brett, is in 2013 or ‘14, how long did it take to consolidate all that data and get all that information together so you could build a plan?

Brett: Oh, man. Yeah, it was pretty intense.

 

“It’s been really interesting to see how creative people have been about trying to access their own data.” – Miles Oliveira

 

Josh Grigsby: Rob just asked, he's looking for resources specific to community financial institutions that are trying to understand what the major domains of AI are. So if you guys have any recommendations on that before we move on, it'd be great to just answer that and give him the resources that he's looking for.

Miles: I can take that one for Rob if you want, Brett.

Brett: Yeah, go ahead.

Miles: I think, Rob, there's a lot of main domains of AI, but it really is applied to your actual use case of what are you trying to accomplish. A lot of AI you want to start small to begin with and then kind of grow your group there. There's really not a lot that you can not do with it, but it is a tactful approach in applying that kind of system to your organization.

 

“It’s a tactful approach in applying that kind of a system to your organization.” – Miles Oliveira

 

Brett: Let's jump on to the next question. This one's for Josh. So Josh, what are some advancements that you see in the AI potential to change or revolutionize the credit unit industry in the next one to two years? Miles had touched on a little bit of it, but what do you see in the future for financial institutions with AI?

Josh: Yeah, for sure. One of the biggest ones right now, well, I got a few, but when ChatGPT, these LLMs, first came out, one thing they fell really short on was reasoning. You could ask them a question; if they didn’t know, they would answer just stone-cold face like they did know. And they did not do well with mathematical questions, with questions that require rigorous logic. If you can imagine giving them some of your financial data and then asking some questions about it or helping plan out something that's really important to make sure that the pieces fit together, you could not rely on them for that. And this has been a huge point of development for the biggest AI, LLM model creators, OpenAI, and Google - to improve their ability to reason.

So when you talk about having an LLM agent that is helping you work through your data and understanding it, you need to make sure that they can reason. When you ask it a question in English or whatever language and it needs to actually translate that to code or whatever and then be able to answer the question you ask, that takes some pretty strong reasoning skills. And so we're seeing a lot of work put into that right now, and they will just be getting better.

They're probably close to being masters in math at this point; that's just like one little benchmark. It doesn't really represent how well they'll be at everything logic related, but it's one benchmark they use. The other area that's very interesting is called agents, and this is basically composing AI to talk to itself.

You have agents who take on roles. I'm talking about, like think of any situation where you have a team working together. I'm in software, so I'll use that example. We have project managers; we have engineers; we have people who test. We have the clients, the members, the people who are using the software. And the way we work and create and build value in the world of business is by working together in these different roles. And so, there's a lot of work and experimentation and frankly, really good results because you don't need a model just to be smarter if you can structure a team of agents appropriately.

You can have these agents play out these different roles and hand work to one another to create a really good end product. And so any situation where you have pretty well-defined input, output, team-playing roles, whether we're talking like back office or loans and underwriting, then you can have these, you can build these teams of AI to work together. And it's not just AI working together by itself. You can compose teams of people and AI. That would be a major theme in everything I say: using AI as a tool, not necessarily a replacement. It accentuates, like it's really a, it's a working together thing because for many reasons, that's what I believe is the best way forward.

And then lastly, I'll just say near-human interaction. I don't know if you've seen the latest stuff from open AI, but it's basically reached like movie levels in terms of real-time communication with a voice that can speak on your level and sounds human-like. And I think this is probably a pretty fine line in terms of what people, who people want to interact with. But, we've already seen this trend with call centers and chatbots, we have this theme of cheaper yet satisfying interaction for customers.

There's a whole swath of questions and answers and engagements that are pretty well-defined and routine and don't necessarily need to bring in a human. And so this will continue that trend of being able to provide members, you know, satisfaction with their questions and investigations in a relatively cheap manner.

Brett: And I think that's really funny. There's been a lot of conversations that I've had around credit unions right now, you've got those slim margins, they're needing deposits. They are looking at, ‘How can we use AI to serve the member?’ And really, I think sometimes we should look at flipping that around like, ‘How can we utilize that to help the staff?’ You've got a lot of new staff, you've got staffing shortages.

And a lot of times you have like an MSR on a call center that's looking to open, let's say an auto loan, right? And there might be something they don't know. So the first thing they do is, they either ask the employee next to them, or they reach out to their manager. Well, if both are busy, that usually creates kind of that behavior of like, ‘My manager is not ever available’. Well, you know, ‘I have no support’.

If there's some internal tool utilizing all the data procedures, the training that they've got, where they can either talk or chat and say, ‘How do I finish this?’ ‘How do I open an IRA?’ ‘What does it mean if they're a sole prop business?’ You know, having that internal tool. I think is something also that we probably will see easily in the next one to two years. I know there's a couple of fintechs working on that, but not to that degree.

Josh: If I may add to that, I think the internal, increasing effectiveness of people doing their jobs is the more attainable right now, this is something that can provide value. And then this kind of harkens back to the last question, but I have friends who work at a Big Four accounting firm and, you know, they're rolling out this big tool that literally has all the regulations in it. It has their previous work. And so when they're stepping through these, frankly, very complicated regulatory filings it's providing them a framework to work through, to know what they need to do, before they file to the SEC, you know, that kind of thing.

Brett: That's great. Well, Miles, do you have anything to add?

Miles: Yeah, Josh touched on quite a bit of it. I think it's that value to the internal team to make them more efficient. I talk about this a lot with our clients. There is a massive amount of, obviously the data that they have available. There's also that kind of knowledge of where to find that data, right? Of, ‘Oh, well, where does this exist?’ ‘Is it in this folder?’ ‘Is it in here?’

And if you're new and you're unsure, or if you're running into a question that you haven't experienced before utilizing, instead of just kind of like, ‘Okay, I need to ask someone or I need to do that’, you can have that information right at your fingertips by just asking a simple question. That improves your customer satisfaction and employee satisfaction, because it empowers them to be able to be more effective at their job, a lot quicker in onboarding time, but also in the actual experience for the membership and for them as an employee. So it's a very, very, very valuable tool.

 

“You don’t need a model just to be smarter if you can structure a team of agents appropriately.” – Josh Melander

 

Brett: Yeah, nice. Yeah, that's good. So the next question is, is we, we touched on this a little bit, but how can this AI, LLM analytic power help credit units better understand member behaviors, preferences to tailor offers, you know, offerings and services. We touched on it a little bit, but give us some more Miles.

Miles: Yeah, absolutely. I think it comes down to looking at some of the most valuable information, which is transaction data, right? How deep a level of data can you get to like a level three aspect of that, where you're starting to see kind of where was something bought, what size, all that kind of stuff. You can really start seeing a pretty deep level of how people spend their time and their money.

For a credit union that maybe has just information on like, maybe they only have an auto loan and a savings loan with you, or they only have one different opportunity for you. Seeing how maybe historically accounts that started in a similar place grew over time would give you that opportunity of like, ‘Okay, well, this is maybe a good time to reach out to this person’. Or, ‘This is a service and an offering that might be more valuable to them’. Instead of kind of just looking, like, ‘Oh, well, I don't really know how to interact with this person, because they only have this, this certain thing there’.

I think that's very, very valuable. Same thing as Josh was touching on earlier, the level of transcription that AI is able to do now. Josh, you mentioned like near movie level. It's actually pretty incredible what the transcription can provide, suggestions of what to do next. It can imply a little bit of tone and kind of how people are saying the words, not just what the words are themselves. It’s very interesting.

So if your customer service, your MSR folks are on the phone, you can actually have AI just transcribe and suggest, ‘Hey, this might be something that would be valuable to this person based off of who they are’. Or kind of, ‘Hey, this might be a good way to lead that conversation next and give them that extra support’. For merely providing a suggestion can help enhance that experience.

It can also be used for training purposes later on as well, where you have that full transcription. It's not just listening to it on audio, but actually being able to see that reaction in real time. So that's kind of some of the ways that make a lot of sense, at least to me for credit unions, as well as kind of what I talked about earlier, which is member life cycle, member segmentation as well. ‘Okay, if we have an X amount of folks – this is the breadth of people who have home and auto loans.’ ‘This is the amount of folks who just have maybe checking and savings with us, or just have a credit card with us.’ Being able to see the cross-section of all that data as well, and ask questions of it to help build some products and services that makes more sense to the folks of, ‘Hey, if we want to get more people joining our credit union, how do we make a pitch to them?’ ‘What’s going to compel them to switch or to just join up with us in the first place.’

Brett: Those are great points. I was having a conversation with a gentleman from Dade County in Miami, and he essentially utilized something similar. He looked at it in a different light of like, a lot of times we look at like products and services of like, ‘How can we enhance these?’ I know that the big fee thing is a big topic right now in financial institutions in general. He utilized it.

So what he did was he took all of the members that had overdrafts for the year. And then he took the top 15 and said, ‘We are going to take these members. We're going to refund half the money into a CD with the promise that they'll do financial education. We're going to get them out of this cycle. And then in six months, they'll get these fees back.’ And it was unbelievable because the number one individual that got the most overdraft fees, it was $15,000. So essentially that individual got half of that back and with that, went through these courses and was able to purchase a vehicle, which got them to and from. So not only were the tools, the AI, utilized to look at some of these behaviors, it also helped that individual become financially well, get a vehicle, being able to go to a job to and from.

I mean, you know, there's just so much to it, right, that you can look at. And so, with that, Josh, do you have anything to add to this question?

Josh: Oh, man, I think y'all just hit a bunch of really specific ideas on the head. I don't think I could add anything to that. I would just, maybe a more general point to make is like having your own personal data scientist, someone that, something that can be by your side and answer your questions. ‘I wonder how many overdrafts we have per year?’ ‘Who are the…?’ You know, the way you can more directly engage with data based on how you're thinking about the problem.

 

“You can actually just have AI transcribe and suggest, ‘Hey, this might be something that might be valuable to this person.’” – Miles Oliveira

 

Brett: So, next question is Josh, what do you see as far as some challenges or barriers of implementing AI in financial institutions or credit unions? And what do you recommend? How do they overcome it?

Josh: Yeah, so one of the ones I think will be pretty big is like identifying these, what I call supporting workflows. So, identifying these opportunities to implement, you know, a partner, an AI partner in some workflow. And really, that's because this is really a new paradigm and we're still figuring out all the ways that they can be of help. And I mean, I think we've got some great examples here, but there's almost certainly many more.

The way I would approach that is by really empowering people on the ground level to have the basic general tools. People a lot right now are just using ChatGPT like they use Google. They're just asking questions. And a lot of people are seeing value from that. And then you might hear from them, ‘Well, this helps, but it'd be really great if it knew about this’. And then you can iterate and develop from the ground up a tool that is supporting your people on the ground doing what they do day in and day out. I think that's a way, a methodology for discovering the places that can provide a lot of value.

Second, cost. This is, I mean, how do I say this? All the new AI tools right now are heavily subsidized. It is incredibly expensive and I'd say no one's making money right now except for the people selling the hardware. And so I would not be surprised if this is an Uber type thing where you get a lot of subsidies and it's really cheap now to start building on it. But at some point the cost will come due.

And that means be thoughtful. And we still need to engage in a cost benefit analysis. We've got to know, okay, this is great. This is helpful. How much money and time are we saving ourselves? And does this actually pay for it? We can't just go in and build everything that we want because that will become pretty expensive.

That said, lots and lots of improvements. I mean, things are improving so fast. I think every week or two there's something headline worthy. Things are getting way more efficient and maybe what I just said won't be an issue, but right now it is. You can quickly find yourself spending a lot of money on these kinds of tools.

Lastly, I'll just put it out – this is kind of a give me in this space – is data sensitivity and privacy. This is some of the most sacred information in people's lives. And so we need to be thoughtful about how we use that.

Is it okay to say we're building a tool that customers can chat with about their financial history? Are we just going to package that up and send it over to OpenAI so that they can answer? It's like probably not, maybe, but these are the things to think through right now because all the biggest ones are hosted by other people. Okay, so if you want to host it yourself, then we're in the cost realm because now you're having to pay for the compute and that kind of thing.

And the last bit I'll say about this is that these models are not bulletproof and we don't fully understand the way that they introduce security risks. There's a whole new industry in the security industry starting up around the way that these models can be exploited. It's a new technological frontier with a lot of value, but it also provides opportunities for people with malicious intent. And so that's a big risk when it comes to some of the information that's most important in people's lives.

Brett: Definitely, that's the big privacy aspect of it. We were at a conference a couple weeks ago and we brought up who in the room – it was all credit union folks – and we asked who in the room had used ChatGPT. They all raised their hands and we just asked for some examples. Boy, it was really interesting. Some of the examples we heard, one had uploaded their auto loan data to see what was the average auto loan at their credit union. We're like, ‘Whoa!’

But there are frontline employees using this stuff and so that's going to definitely be a challenge. You've even seen some credit unions and some that are even highly innovative that have put in policies and procedures in place that you cannot use it when responding or working at work. And so that's been pretty fascinating too.

Josh: I think that's tough because when people, individuals, when they have a tool that is really useful they're going to work around restrictions. They ask someone not to use Google; it's a tool and if something provides enough value people are going to go against the rules. And so, if there's a big enough wave it's like, all right, how do we encourage this in a safe way rather than just saying, don't do that.

Miles: Yeah, I think a lot of it kind of comes back to some of the processes that Josh talked about, that starting small and understanding exactly what data do we want to utilize. And then I can get into this a little bit later, but the first step is cleaning the data, making sure the data that we're going to actually use to apply to a model is definitely only the information we want to have and that's not being used to be trained elsewhere.

A lot of AI, especially LLM models, will give you essentially a card that says this is how this model is trained, this is the data that it's trained on, this is kind of how this works there as well. And so part of that policy for an organization is to be like, well, this is how you should use it. These are things that are absolutely no-go's and a lot of that, if you're running your own model against it, you can actually enforce that from prompts.

It will not allow certain prompts, it will not allow certain data and just be like, ‘Nope, we cannot utilize that’. The only thing it's trained to do is say, anytime someone puts in a prompt with, let's say, an account number or kind of more personally identifiable information of that sort, absolutely not, depending on how you're utilizing that model. So there's different tricks and ways to kind of make sure that the data that you are using is safe and secure. And there is still extremely valuable ways for your employees to use it internally.

 

“All of the AI tools right now are heavily subsidized.” – Josh Melander

 

Brett: Cool. You kind of like segued to our next question for you. Let's talk ethics or ethical considerations that we need to keep in mind when utilizing AI and that sensitivity of financial information, privacy, those type of things.

Miles: Yeah, absolutely. Well, definitely privacy is kind of where you want to start. So you want to make sure that, again kind of coming back to it, you want to make sure that the data that you're applying to a model is the only data that that model will essentially be trained on and that it's not sending that data elsewhere. You're not utilizing it in a way that is going to expose any sort of PII, any kind of PCI compliance. Well, anything that you just really wouldn't want that information getting out into, somewhere someone else could use it who maybe shouldn't be utilizing that information. So that's kind of step one.

And then step two is cleaning the data itself and making sure that you understand a little about kind of what your data is. Again, this is still software. So bad data in means you get bad data out. That truism is still very much applied to this tool as well.

So make sure you have good data, making sure you understand what that data is, what it means. Start small with some tests. A good example of kind of what not to do, Brett, if you'll let me kind of utilize this example?

Brett: Do it.

Miles: Our good friends over at Wells Fargo who have a lovely colorful history of just excellent practices and they've never been in the news for kind of anything about what they've done in the past, what have you. So instead of kind of utilizing some of the privacy concerns and making sure they're cleaning their data, making sure they understand a little bit more about kind of what their actual data is and what it says and doing some tests, they utilized a pre-authorization model to pre-authorize auto loans for certain individuals. What they found was that the model was very good at lending to certain demographics and very bad at lending to other demographics.

And that was based on how that model was trained and weighted, which means they did not either clean the data or they did not train the model appropriately to be a little bit more inclusive. So they immediately shut that down because it was not great for their individuals. And it was also denying the business that they would have likely wanted to have outside of that.

So that's a good example of the ethical context behind it of, you should always understand what your data is and how you want to train your model against it, to make sure that it's getting you the results that you actually want.

Brett: Nice. Thank you. Josh, do you have anything to add?

Josh: Yeah, I think the only thing I'd add is just like there are laws that have been passed, GDPR and whatnot, that give an individual the right to opt out and to request their data, the way things seem to be going policy-wise in Washington is that, that right will be passed down in terms of having  their data trained on. So, it very well could be the case that individuals can say, I don't want my data trained on.

And we've seen all kinds of examples of these models regurgitating. They're not always just creating new things. Sometimes they just regurgitate what they were trained on. And so, yeah, I think that's likely going to be a consideration going forward into how these tools are implemented as any one person can have the right to say, ‘No, not my stuff, please’.

 

“You should always understand what your data is and how you want to train your model against it.” – Miles Oliveira

 

Brett: You look at, right now, credit unions – one of the big things we're seeing is CUSOs and CUSOs starting to partner up with technology providers or create their own software or service. How can credit unions start collaborating with individuals, fintech startups to stay competitive?

Miles: I think the best place to do it, and we can kind of talk a little bit more about this, but I feel like if you're not utilizing artificial intelligence in some form, most likely is that your competitor is. It’s going to be a part of our lives for pretty much the rest of our existence, at least, you know, for quite a while.

And so, because of that, I think it's important to start small and start finding different ways where it's going to be valuable. Each institution has some similarities and they have a lot of differences as well. It's really important just to kind of start having those conversations internally about like, ‘Hey, this is something we might want to take a look at’. Or, ‘This is somewhere where we think it could be valuable to us’. Or, ‘We have inadequate reporting on this data set’. ‘We would like to maybe apply an AI model to it to see if there might be something more valuable that we can learn out of that, that we aren't right now’.

And so I think the first thing you want to do is make sure that you kind of engage with maybe a company like ours or another consulting company and really understand more about like, ‘Well, what do we think the actual value is?’ And start building your kind of goals and dreams around that. So understanding that you want to start at the end in mind, this is what we're looking to accomplish, or this is what we want to explore, and start small with some test information and start building something that actually has tangible value.

I think a lot of that was what Josh was talking about earlier. Just like, here's something that actually has like real tangible value with some key performance indicators of like, this is actually how it's going to save us a bunch of time, or this is how it's going to support our employees be more effective, or this is how it's going to improve our membership experience or grow our membership itself. And really kind of looking at those kind of things there and finding the solution by working backwards.

Brett: Cool. Any additions, Josh?

Josh: Yeah, I mean, everything that Miles said, and I'll just be really specific. I think a great place to start is to have, you know, every company has a repository of information and standard operating procedures and all this information that can be a little hard to go through specifically. So, I would start with integrating that into a personalized or maybe not personalized – credit unionized bot that has all your information and you can start people using it now, asking questions and increasing efficiency and then let us iterate on it, see how people are using it and we can build out from there.

 

“If you’re not using AI in some form, most likely your competitor is. – Miles Oliveira

 

Brett: Nice. Yeah, so with that, this is another great segue. I had asked Josh and Miles when we were kind of having our prep call for this webinar, ‘where does a credit union start?’ ‘Where would you recommend?’ And these are some of the notes that I took. So first one is: Align your AI strategy with your business goals. Kind of your roadmap, like, you know, is it employee training retention? Is it finding that, essentially that life cycle of your members?

And the other one is: Upscale your employees for AI. So, if we know that they're using it, put things in place like procedures, policies, get training in place, maybe even adopt a specific AI for your institution.

And then: Collaborate cross-functionally. So, throughout the credit unions, maybe with other credit unions, associations, partners.

And then as we talked about: Implement data governance. So really looking at your data, having it maybe even human-led AI, but just making sure you know your data and you're governing that data.

And then really starting to: Standardize and automate your business practices. So as Josh had mentioned, starting small and looking internally at all your repository of information for your staff.

And then the big one is if there is cost to it, really: Ensure that there's adoption and there's value to it. We see a lot of times that institutions get intrigued with that shiny ball or the new thing and they invest in it and then it doesn't get utilized with their staff or their credit unions. So, you know, really ensuring that there's value and opportunity for you.

 

End Interview

 

Brett: We do have continued webinars coming up. So we've got our next one – we are going to have the infamous Max on the line to talk about buy versus build. And so that'll be a really fun conversation. He's also going to talk a little bit about what it takes to build a good software development team. And after talking with Josh and Miles and the other Josh, I'm pretty sure I know where he's going with that.

And then we're going to have some guest speakers that come on with us and really talk about their fintechs and their ideas as well. So with that said, I want to just thank Josh and Miles today for spending time to talk to all of us. And then Josh Grigsby, thank you very much for managing the chat and letting people in and out.

So have a great Friday, everybody. Thank you.

 

Thank you for joining us!