Webinar Recap: AI in Collections

Watch the Webinar On Demand

With major shifts happening in the financial landscape, you may be evaluating if it's the right time to introduce AI tools in your collections process. 

In this webinar, we dive into the potential of AI and equip you with insights into its practical applications, benefits, and potential challenges. Leave this session prepared to navigate these shifts in innovation and make informed decisions for your business. 

Topics & Timestamps

03:00 | What is Artificial Intelligence
09:28 | Benefits and Risks of AI
23:00 | What to Consider Next
25:30 | Past-Due Customer Insights 
32:00 | Why Data is so Key

Key Learnings and Takeaways

In this webinar we discuss:

• The fundamental concepts of AI and how it's reshaping the financial landscape.
• The distinctions between AI and Machine Learning and their unique impacts on FI operations.
• Insights into the potential benefits, challenges, and risks AI introduces 
• A peek into the future and how upcoming advancements influence customer experiences
  
  

Read the Transcript:

[00:00:00] Michael Pupil: Good morning. Good afternoon. Good evening, everybody. And thank you for attending Lexops most recent webinar. My name is Michael Pupil. I'm going to be taking us through hopefully a fun session today. Today we are going to be talking about AI, a really relevant topic today and navigating the benefits and risks in collections and talk about great timing.

[00:00:26] Michael Pupil: I had delivered a conversation maybe about a month or so, ago, talking about AI and obviously I delivered that conversation in California and looks like the writers strike seems to be coming towards an end. So some fun updates there with this. And so as we continue just a little bit of housekeeping, we are going to be recording this session.

[00:00:50] Michael Pupil: We're going to be sending it to everybody. There is going to be a question answer period towards the end of the today's webinar. And you can, of course, post all of your questions in the Q and a [00:01:00] section of the box. But let's get kickstarted today. I think it's going to be a fun one, hopefully. And hopefully we learn a few things, including myself here.

[00:01:08] Michael Pupil: So again, my name is Michael Pupil. I'm the vice president of Lexop. This is going to be a session today just exploring some of the foundations of AI. I think that there are also some differences that we should be talking about between AI and machine learning and how that is reshaping the financial landscape.

[00:01:26] Michael Pupil: For a lot of lenders that are out there, it is wonderful to see how many credit unions regional banks, national banks that are in attendance. But we also have a couple of other you know, individuals in in today's session. So hopefully that will translate to everyone. We'll be talking about a little bit of the AI realities.

[00:01:44] Michael Pupil: And where that stands specifically for credit unions and banks. And of course, what we anticipate as the future and what is on the horizon and what we're seeing today. So the first thing that I would like to start with is that AI, like anything else is a tool [00:02:00] and tools need to be used effectively.

[00:02:03] Michael Pupil: Some bonus points for some mythological nerds that are out there. I am one of them bonus points. If you know who this is but this is Prometheus and Prometheus is. gift to humanity or tool that he brought down was fire. And for that he was punished where, you know, new tools to humanity sometimes can be used for both good and bad.

[00:02:24] Michael Pupil: And when we think of fire as a tool you know, cooking food and providing warmth and certainly a turning point for humanity. It could also burn down the village or burn down the house and it could be very, very devastating when used inappropriately. So a tool just like any other, I think that there is some fear surrounding AI, but also some excitement.

[00:02:44] Michael Pupil: And just like anything else, when tools are put into the right hands, you can create some beautiful you know, and very artistic you know, product or it could be used in a, in a really great way. And tools put into the wrong hands you know, [00:03:00] most likely myself, if I were to take a chisel or, or anything like this, probably come up with you know, the, the middle picture here when it's, when it's not being used effectively.

[00:03:09] Michael Pupil: So what is artificial intelligence? I think we need to start at the beginning. You know, what what is artificial intelligence defined as? And Most broadly, it is the simulation of intelligence by machines. It is supposed to simulate human activity with machine ability. And so what that means is that, you know, it should be able to speak.

[00:03:33] Michael Pupil: It should be able to read, see, hear, and move. And not in all cases are you going to have all five of these abilities that are going to be in the system, but certainly this is the replication of what humanity can do, and you're looking at artificial intelligence to do this. And so we're looking to match human capabilities.

[00:03:55] Michael Pupil: It should also be able to discover new outcomes, which is prevalent with our, [00:04:00] with our species. It should be able to infer and read from multiple sources to be able to come to a conclusion, and it should be able to reason to put something together. Tall task to, to mimic, but there is a lot of progress in this space, and we're going to dive into what that looks like.

[00:04:16] Michael Pupil: So machine learning as a counterpart to this is a subset of AI and what machine learning does is that it uses a primary set of data algorithms. And the reason why I'm kind of pointing into this is not to get too too deep into you know, the technical side of A. I. And machine learning. But More of a superficial understanding of how this is used and where this is used and you know, gradually helping organizations modernize their communication tools, especially when it comes to things like collections in order to be able to use these type of tools in the most effective manner.

[00:04:54] Michael Pupil: possible. And I think that there's a little bit of an overlay between machine learning and AI. And so we're [00:05:00] going to have a great example as to what machine learning is and how it's implemented today. We can make some inferences in terms of lending or in the collection space and gradually moving forward from supervised to unsupervised.

[00:05:14] Michael Pupil: Machine learning within the system. So we'll start today. And since we're, I'm on the East coast and it's noon I think pizza ends up being a great analogy for what we can do. I can't take total credit for this. This was something that I learned from Microsoft when they were going through these, but machine learning is essentially a series of inputs.

[00:05:34] Michael Pupil: We were talking about data inputs before about what machine learning is and inputs could be zeros and ones. Yeses and nos, and assigning weight factors to those zeros and ones in order for the software to make a prediction or a conclusion. So let's run the rabbit down the hole, so to speak. If I wanted to order pizza, I would input a number of [00:06:00] series of values of what that is going to look like.

[00:06:03] Michael Pupil: And so with yeses and nos or ones and zeros and weight factors to the three questions that I'm going to ask whether or whether or not I should order pizza, it'll tell us whether I'm going to be making that phone call or ordering it through the app. And so the first question that I want to ask myself is if I order pizza, will it save me time as opposed to making a meal for myself?

[00:06:25] Michael Pupil: And probably the answer is yes, I could probably order a pizza faster than I can make myself you know, a relatively virtually the same type of meal or make a pizza myself. So I'm going to input that as a, as a one Will that Help me lose weight. Maybe that's one of the goals that I want to have here.

[00:06:44] Michael Pupil: And of course I don't know if pizza is going to help me lose weight, not to get into a philosophical debate about how healthy or unhealthy it is. I like my pizza fully loaded. And so I know that that's probably not going to shed a few pounds for myself. And then the third input that I want, or the third [00:07:00] question or value that I'm going to enter is, is it going to save me money?

[00:07:03] Michael Pupil: Will ordering a pizza cost less than perhaps me doing some other things? And I've got some coupons, so I'm going to say, yes, it's going to save me money. So you can see the series of yeses and nos across the board. Now we can easily make. A comparative analysis to let's say lending where, you know, do you have the right you know, factors in play and yeses and knows that you are plugging into this to help move forward.

[00:07:30] Michael Pupil: And then, of course, we want to assign weight factors to this. And so I highly value time and on a scale of 1 to 5, 1 being the lowest, 5 being the highest, I want to definitely be able to you know, save some of that time for, for ourselves. Will it help me lose weight? Well, while I like to lose weight. It probably is not the highest level of priority.

[00:07:52] Michael Pupil: I'm not in the gym every single day, although I probably should be. I'm going to assign it a value of three, but it's not going to be the highest value [00:08:00] that's in there because I definitely have my cheat days and losing weight isn't exactly the highest, but it's not a zero across the board. And of course, will it save me money?

[00:08:07] Michael Pupil: Well, Ordering a pizza or making a meal isn't too cost prohibitive in either case, and so I'm going to assign a weight factor a little bit lower in that regard. And as you can see, what we're doing here is, if you're entering in the data outputs that we're looking for, ones and zeros, yeses and nos, and assigning a weight factor, If the ultimate answer is that through this exercise, if we can get to a weighted value or a threshold of seven, it'll mean that I'm going to order that pizza.

[00:08:36] Michael Pupil: If it is below, it's going to be declined. If it is equal or above, we're going to go forward with it. And so hopefully I've made everybody very, very hungry with this example, but maybe We've you know, kind of simplified what machine learning is in terms of what we have. And some of the organizations that I know that are in attendance today probably have some sort of machine learning or process built in place that this [00:09:00] accurately represents.

[00:09:01] Michael Pupil: Maybe not so much with pizza. So as we do the multiplications of 1x5, 0x3, and then 1x2 and add those all up, we see that I come to a threshold of 7. Which means I'm going to be eating pizza later today. So that was a fun little exercise. Now let's talk a little bit about AI specifically in credit union collections, banking collections, you know, primarily the financial institutions that are there.

[00:09:28] Michael Pupil: And we're going to. Lean on a movie that I absolutely love, which is the good, the bad, the ugly and for various reasons, I, I love that theme music, and it's been sticking with me since my childhood but totally applicable in today's conversation, so let's pick on the good you know, where does A.

[00:09:47] Michael Pupil: I. Kind of fit in for collections. And of course, where it's being applied to today is in text messaging and a I powered chat box and communication that's going back and forth. Now that [00:10:00] could be good. And we'll get to the other two characters in the movie shortly. But what this does allow is service for 24 7 access for any member that is looking to reach out or any customer that is looking to have questions answered.

[00:10:15] Michael Pupil: And as long as You know, the data that you are inputting into the AI algorithm is respectable and is accurate. You should get a positive outcome to this and we're seeing this in large part today. You can translate that into email and where AI can analyze a lot of the borrower data and can personalize those segments.

[00:10:38] Michael Pupil: So, you know, obviously things like date due date the amounts, things like that, what the time of day and day itself in a seven day week has a higher hit rate in terms of an open rate and a reply rate can be attributed here resulting in better communication or more effective communication.

[00:10:58] Michael Pupil: And of course there's always [00:11:00] voice. The AI powered voice assistance. Those have been around for a long time. I've disconnected, you know, Google in the background because every once in a while when I start having these conversations, I think we've all fallen fallen victim to hearing them in the back when they're when they're listening to us.

[00:11:14] Michael Pupil: But many of these, you know, scenarios is what we were talking about before about. You know, the, the definition of AI and where it could be applied. Now, many of you guys will say, well, Mike, that is a double edged sword because it comes with risks and not all of it is good. And you're right. Second character in the movie is the bad you know, where are the risks with implementation of AI?

[00:11:38] Michael Pupil: And this becomes very, very serious for everybody across the board because of both the people and the technical risks. So. Let's tackle the people risks to start the limited ability to solve complex problems. It's not perfect yet. But the tool is growing. The tool is being used and being pumped in with a lot of data in order to be able to solve more and more [00:12:00] complex problems.

[00:12:01] Michael Pupil: At the moment, it probably needs a little bit more time, but that is a limitation of the system. And it has people risk because if it has the inability to solve all of these complex problems, there is a difficulty in recognizing. and resolving these disputes. And of course, if you're hindering access to human interaction, the outcome of that communication could possibly be a negative one or not a, not exactly a great one, not what we're looking to go for and not what most centers of service are looking to achieve.

[00:12:34] Michael Pupil: And so if we're failing to provide meaningful customer assistance, you know, what are we really doing here? And is it ready and do we want to keep it in play or do we want to build off of it? And that is a decision that will be taken on by every organization across the board. And everybody is going to be a little bit different with the risk appetite that they have.

[00:12:55] Michael Pupil: Speaking of risks, let's jump into the technical side of things. [00:13:00] You know, of course, there is a security risk for impersonation. Lots of organizations work very hard at trying to implement biometrics different types of passwords you know, both two factor authentication, but impersonation continues to be you know, something, whether it is AI machine learning or otherwise.

[00:13:19] Michael Pupil: Impersonation and fraud is obviously something that is on the plate and has been on the plate and will continue to be on the plate, unfortunately, for everybody. And while AI... May not necessarily have sick days or holidays or you know, change roles and have, you know, those type of issues. There is, of course, system reliability and downtime.

[00:13:39] Michael Pupil: Not all systems stay online all day, all the time. And so system reliability, if it does go down, what is the remediation you know, to those systems? Keeping PII safe. I think you know, I referenced that, you know, quite a bit in the the first one with impersonation. I think that is at the core of every [00:14:00] organization is to keep their customers, their members, and all of that information away from, from some bad players that are out there.

[00:14:07] Michael Pupil: And the one that I'm going to pick on the most is how decisions are made. That to me is where some of the biggest risk is in play. And I'll give you guys a couple of real world examples from some of the credit unions that I've been speaking with as to how decisions were made using things like machine learning or data sets that ultimately led with the best of intentions to some.

[00:14:30] Michael Pupil: pretty nasty outcomes. And of course, the third and final and probably funniest character of this movie. And if you guys have not seen this, you probably should cause it's amazing. But the ugly and sorry, CFPB for putting you into this category. No no disrespect to you guys. But CFPB came out in the month of June, I think June 6th, they released some recommendations.

[00:14:54] Michael Pupil: For organizations specifically in financial services that are going to be using a I and they broke it [00:15:00] down into four categories. And the first one is to be transparent and transparency, meaning how are they using a I and debt collection practices and the explanation to the consumers to how this is being used.

[00:15:12] Michael Pupil: I think that's a little bit of a gray area today. And not widely used across the board, but these are the recommendations coming in that transparency saying that this is engagement with AI and being forthcoming about it. And the honesty and transparency is always a good thing when it comes to communication.

[00:15:30] Michael Pupil: And so I like this one. The second one was accuracy to be able to ensure that the AI models are tested. Regularly to validate the models, whether they are producing or not producing biased or discriminatory results, AI, and I think everybody here has probably heard this before is only as good as what information is being fed into the tool.

[00:15:54] Michael Pupil: And so. Understanding where the information or the data points are coming from [00:16:00] is incredibly important. I would go out on a limb to say not everything on the World Wide Web is true. And so if AI is consuming everything on the World Wide Web there is definitely going to be some data in there that is skewed, that is biased.

[00:16:19] Michael Pupil: discriminatory, a whole bunch of, you know, things that we don't want to necessarily be feeding the tool. So who and what is going to be in control of those data sets is incredibly important. We did that exercise earlier about those ones and zeros and weight factors. Perhaps time isn't as important to some people as it is to lose weight.

[00:16:40] Michael Pupil: And you can get a very, very different outcome. Whereas mine was a seven for somebody else, it would be. Absolutely not. Can't do the carbs. I need, you know, to go on vacation pretty soon. And the baiting suit is a little bit tight. So I, I'm going to go below the seven threshold. Accuracy is where you know, I referenced before as one of [00:17:00] the most important pieces.

[00:17:02] Michael Pupil: Item three for the recommendation from CFPB was data privacy and security. Mentioned before that this is an incredibly important piece, not just for AI in general, but for all of us that are the stewards of information of PII to ensure that we're protecting that data and that it's not misused or mishandled.

[00:17:21] Michael Pupil: And I think that continues to be a cornerstone of all of us that are on the call today. And then lastly, which I thought was One of the funniest things that could have happened is human oversight and ensure that there is human oversight of AI models that decisions are made by AI subject to review and can be appealed by humans.

[00:17:40] Michael Pupil: Well, the whole idea behind artificial intelligence is to, you know, be able to give those tasks over where the system is going to mimic human behavior. Eliminating the need for human behavior. This is where some of the fear comes in, where we're all going to lose our jobs to AI. But funny enough, [00:18:00] CFPB had said, well, even though you're going to be using this, there needs to be human oversight over an artificial intelligence tool.

[00:18:07] Michael Pupil: And it does make a lot of sense for today because I don't think that the tool is perfect. And I'll give you guys a couple of examples before or later on in this piece. But It's definitely a little bit of a tricky piece here, where if you want to modernize consistently across the board, you still have to have that human oversight to it.

[00:18:26] Michael Pupil: So a little bit of a, of a mix there, mixed signal. Well, instead of just picking on AI, let's talk about the digital tools and communication, because I think that's where we're going to be using it most on the collection side, certainly used on the lending side in terms of pulling data in, but not too long ago, there were.

[00:18:46] Michael Pupil: A lot of advertisements and news articles coming in about credit card debt, reaching over a trillion dollars. And you know, Lexop does a lot of work in, you know, kind of probing past due customers in terms of their [00:19:00] mindset, their methodology, the why. As to where people are going when it comes to pass due bills, but there are other groups that do this as well, McKinsey being one of them that recently they had conducted a survey to find out of those credit card customers, you know, where do they prefer to get their communication and they divided it into four categories using FICO scores of, you know, low balance, low risk, all the way to high balance, high risk.

[00:19:26] Michael Pupil: And where were those channels of communication preferred? And you can see that email and text messaging actually almost leading the category across the board for every single one of these, and at least a significant portion of this, in terms of the back and forth interaction. And so that's telling us that the digital channels tend to be leading you know, some of the, the older, more traditional channels of having to physically go into this or even having conversations across the board.

[00:19:56] Michael Pupil: Now. Phone and live and text messages and letters, [00:20:00] those still have a place. And it's, you know, kind of the tools in the tool belt. If we think back to the horse railing that that was there, not all tools are meant for every single job, but definitely communicating with customers and the method that they prefer has a higher result.

[00:20:15] Michael Pupil: And, you know, there's a lot of studies, not just McKinsey or Lexap that are, that are coming out with these, but a lot of organizations would probably support this. to leverage on top of what it is that we're looking at. They also took a look at digital channels for partial payment or for full payment.

[00:20:33] Michael Pupil: And in the 0 to 9, 10 to 29 and 30 days past due, what you'll see in every single circumstance, the digital channels outweigh the traditional channels for full payment, as well as partial payment going across the board pretty consistently as well, ultimately resulting in digital channels effectively getting More of those past due bills than not than the traditional channels.

[00:20:59] Michael Pupil: So don't [00:21:00] take this as I'm advocating for eliminating every, you know, traditional channel that's out there. I think where organizations have been looking specifically in the last couple of years has been to modernize their, their channel. And by the way. That is not a new phenomenon. I've been working, you know, in technology for quite a while.

[00:21:21] Michael Pupil: There's a lot of credit unions and banks that are on this channel today. And so I'll ask and I'll give a couple of analogies as to what I've lived through and where I see collections taking place. If any of you had been working for a credit union or a bank back in the days when www insert, you know, credit union.

[00:21:42] Michael Pupil: com was the primary digital channel of choice when mobile applications came out and I was, you know, in the, the mobile application security space, the number one comment that we got as an organization was we don't need a mobile app. We have a website and our members [00:22:00] are going to go to the website. Well, fast forward the clock.

[00:22:03] Michael Pupil: Every single organization on this webinar today has a mobile app. If you're in lending, you absolutely have it. And so that changed pretty quickly. Fast forward the clock a little bit, and even before pandemic, electronic signature was discussed quite a bit and electronic signature for things like lending for mortgages, for car loans, for regular loans, even credit cards and things like this.

[00:22:29] Michael Pupil: The number one comment that I got was. we value the relationship with our members. We absolutely do not want to have an electronic signature. We want them to come into the branch so that we can have a conversation and develop a better relationship with our members. Password the clock and COVID aside, electronic signature is very widely used today.

[00:22:50] Michael Pupil: In fact, it strengthened the relationship with a lot of those members and those customers, because you made it easier for them to be able to acquire the product and service that [00:23:00] you were looking at extending. And then leaving those conversations up to the members to be having for, you know, other, other pieces to this.

[00:23:08] Michael Pupil: And I see the same thing with collections it's you know many have already sent emails and many send text messages, but it goes more than that. And I think that the digital. Journey for a lot of organizations are still continuing to this day. And so I'll leave those analogies with you guys. You can call me out if I'm incorrect on those, but definitely a number of conversations and maybe a little PTSD from some of those those conversations, but I see history repeating itself.

[00:23:37] Michael Pupil: So it leads us into what's next. And this is my favorite slide because it says a whole lot of everything and nothing at the same time. Establish the problem, ask the right questions, lean on industry experts. Well, that could be said for every single, you know, scenario that we're looking at. You're always trying to figure out what the problem is, ask the questions as to why, and [00:24:00] then figure out, you know, who's leading the way.

[00:24:02] Michael Pupil: Now, this next slide, I'm going to preface for any of the credit unions. Don't get upset about this, but I have a whole bunch of bank logos that are on here. And while we don't necessarily compare one to the other, there is a lot of benefit for organizations that jump into the pool first to kind of work out the kinks.

[00:24:20] Michael Pupil: And you know, kudos to those organizations that try to lead you know, for digitization pieces. And there's no knock to saying, why don't you test it out first before I go in, you know, next. But there's a lot of organizations out there today that are using forms of ai and this isn't all of them, of course.

[00:24:39] Michael Pupil: And there's a lot of credit unions and a lot of telcos, a lot of utilities that are also using AI tools to figure out what data sets. are necessary to input into the system to give us those ones and zeros and one to five weighted averages in order to make sure that the decision making of either [00:25:00] provisioning something or how to communicate with a member on a past due that comes into place.

[00:25:07] Michael Pupil: And so it's incredibly important as to An excellent transition to meet the past due customers of 2023. I'm going to leave this slide up for about five seconds and give you a little bit of context as to what this QR code is. Lexop over the last couple of years conducts a study about the mentality and the mind frame of past due customers.

[00:25:28] Michael Pupil: So our customers. Customers who fall into a past due state, and we probe it in terms of the why, the context as to how they feel, how they resolve themselves, why are they late, all anonymous, of course, in order for us to kind of get a sense as to how the demographics and the world is changing, because everybody has different values for the questions that we ask, everybody has different ones and zeros.

[00:25:51] Michael Pupil: And weight factors to the questions that we have, and that becomes incredibly important, especially on the collection side. And so what we found [00:26:00] was that we call it a prioritization of bills makes a lot of sense. Even if we don't necessarily use the language, we definitely you know, have that innate ability into us as to 10 bills or six bills over the course of the month, which one gets paid first.

[00:26:16] Michael Pupil: and which one gets paid last. And if there isn't enough money to go around, which of the six or the 10, you know, which one of those bills gets punted to the next month and gets pushed out. Now, if you are a provider of that service or that loan, Wouldn't you like to know that you are number one or number two on that list and not necessarily the last one that has the risk of being pushed out 30, 45, 50 days before some of those longer term workflows end up going down the road of threatening of repossession, which is not a good experience for, for anybody to go through.

[00:26:51] Michael Pupil: So it becomes very, very important that QR code will be on the presentation. Again, this is recorded. We'll share it with everybody. If you didn't have a chance to scan it, there [00:27:00] will be an opportunity for you guys to get it afterwards as well. So let's talk about decision making because that's a really important one.

[00:27:07] Michael Pupil: I am going to preface this with, I apologize if you've heard this joke before. I've said it a couple of times, I still think it's hilarious and for those that have not heard this, I think this is a perfect ability for us to talk about decision making, and how we get for a conclusion as to, is it a one, or is it a zero, and what's the weight to it, and so, there was a chat about it.

[00:27:28] Michael Pupil: GPT conversation that was, that was put in place, and I don't know who the person was, but basically asked Chad, you know, how much is two plus five? And of course, it accurately answered that it was seven. But the individual said that my wife says it's eight. And chat replies back two plus five is actually seven, not eight.

[00:27:47] Michael Pupil: It could be possible that your wife made a mistake or misunderstood the problem. And so accurately to be argued, the husband had said my wife is always right. And chat GPT had [00:28:00] said, I apologize. I must've made an error. My training data only goes up to 2021. I may not have the most current information.

[00:28:06] Michael Pupil: If your wife says it's eight. It must be eight. So I think we've learned a couple of things from this you know, this example here. One is different data sets are not equal across the board to everybody. Sometimes two plus five will equal eight. And the other part is that ChatGBT is more successful than any relationship that I've ever been in.

[00:28:27] Michael Pupil: And maybe there's something for me to learn about not necessarily digging my heels into every single. scenario possible. So a little bit of a fun piece to that. Apologies if you've heard this joke before, but I love it. I think it is absolutely crucial, but also very telling of what data sets are we pulling in in order to make these decisions.

[00:28:49] Michael Pupil: Where this becomes incredibly important is when a past due customer lands in that state, there is a whole bunch of emotion. That that individual [00:29:00] undertakes and over 50 percent of that is not a positive one. It's things like feeling frustrated, embarrassed, anxious and overwhelmed. Those are not good feelings to have.

[00:29:09] Michael Pupil: And when 50 percent of the population is in that state, communication with empathy. Is absolutely required and so this is where training the tool training AI to make sure that we're communicating in an empathetic manner is able to shift those feelings into a positive one rather than just a transactional one.

[00:29:31] Michael Pupil: And the reason for that is that well, the data also tells us that if those individuals. fall in a state of past due. If there is a negative experience with communication in resolving that problem, there is a significant percentage of those individuals that would switch their provider. So retention is at risk.

[00:29:53] Michael Pupil: So it's not just about the touchy feely that the happy go lucky things. This does make a lot of business [00:30:00] sense to make sure that something like empathy and understanding is absolutely trained and employed at every single one of our organizations that when we communicate with our member, with our customer, that it is resolving the feeling of frustration, embarrassment, anxiousness, or even being overwhelmed into a feeling of being highly valued.

[00:30:22] Michael Pupil: I would love nothing more 2024, that that 8% Ends up being a much, much higher statistic across the board. And I would think anybody that you know, is on this call, but also try to strive for that first part of the graph to be as high as possible. So if I start talking about data is key you know, you know, is it.

[00:30:46] Michael Pupil: You know, what kind of data are we, are we putting through, there's a saying that that I recently came across working with a credit union where the individual who heads up that department said bad things happen to good people. [00:31:00] And sometimes that, that is a true statement. The last couple of years under COVID, I think had thrown a lot of organizations into a loop.

[00:31:09] Michael Pupil: And what happened at this one particular credit union was that they had offered a program that allowed their members to either defer, delay, or simply just not pay you know, some of the obligations that they had as, you know, businesses were closing down. There was a lot of fog in front of us as to what was happening, but they continued this practice for roughly about a year, maybe a year and a half, or somewhere around that vicinity, which was a significantly long time.

[00:31:40] Michael Pupil: Now, by doing so, by not reporting, you know, that, that information, what actually ended up happening was there was an increase in the credit bureau score because there were no late payments and it artificially inflated the credit score that had come through. And so, now that we have, you know, exited through that [00:32:00] tunnel, now those members that are applying for credit have an inflated score.

[00:32:05] Michael Pupil: And so, the data is actually skewed. In what was the best of intentions actually result in somebody acquiring too much money or alone being too high and when combined with other data sets, like how much they earn or what are the other obligations has actually turned out to be a problem because now they've they've borrowed too much.

[00:32:28] Michael Pupil: And so data is key, but you've also got to take a look at the data. Now that was a real world scenario. Maybe some of you also had some of those programs. Maybe some of you have also experienced, you know, what it is that I'm saying, but here's a great example as to demographics. And of course I'm exaggerating a little bit, but let's, let's just play along for a second here.

[00:32:49] Michael Pupil: And so we'll take information from member A. And member A, some of the demographic information that we have, well, we know that they identify as a male. We know that they have told us that [00:33:00] they were born in 1948, so we know the age group and the demographic that they are in. We know that they were raised in the UK, so we can make some cultural inferences in terms of, you know, who they are and what they are.

[00:33:12] Michael Pupil: They've been married twice. They live in a castle. So I'm getting very, very, very specific with these individuals and that we can see, you know, where it is that they come from. And of course, if you're living in a castle, they are wealthy and famous. Now making this very minute on purpose in order to prove a point.

[00:33:30] Michael Pupil: Now we'll take member B and we'll take some demographics to see if there's some comparisons. They also are a male, born in 1948, raised in the United Kingdom, married twice. Also live in a castle, wealthy and famous. I'll give 30 seconds and I know we don't interact too much back and forth here, but you can probably start to guess a couple of the characters that are going to come up with.

[00:33:54] Michael Pupil: And if you thought of Prince Charles as member A, You were correct male born in [00:34:00] 1948 and all the demographics that are across the board and we can imagine if we were going to be speaking to, you know, the king, I called him Prince Charles, but now King Charles, if we were to be speaking to the king, we probably have a certain demographic now member be Aussie Osborne with the exact same demographics, probably not the same way that we would be speaking from one to the other.

[00:34:20] Michael Pupil: So while personas and demographics, Are are definitely a thing that help us identify, you know how we should be, you know, making decisions. It's not the only thing that, you know, kind of makes sense. And so there's lots of different things that are out there. And so sometimes the human touch I would advocate is still something that we need to work on, despite all the amazing progress that AI has been doing.

[00:34:46] Michael Pupil: And who knows? Maybe I will be able to differentiate in some other way. You know, King Charles and Ozzy Osbourne. So with that in conclusion, and I'll wrap things up here before we move into Q and a [00:35:00] you know, it really does become a balance between the enthusiasm of having a new tool and using a new tool and using it correctly and being able to, to pull out the right type of information and being able to trust.

[00:35:15] Michael Pupil: The integrity of the decision making, and I think that that is that the essence, not only just of credit unions, but of any lender that is out there or anybody providing a product and service to individuals across the board. And so that's what I'll leave you with today. I'm going to turn over to the Q and a, I think we've got a couple of things that come through here.

[00:35:37] Michael Pupil: And so I'm just going to flip to a screen. So apologies for not looking directly at this. First question of the group how will the integration of AI impact the customer experience? Well I think I'm going to reference. Three slides. The good, the bad, the ugly. I think that the impact the customer experience [00:36:00] could be one of those three.

[00:36:01] Michael Pupil: And the good is I travel quite a bit. I'm in a hotel room here right now. And oftentimes when I have to interact with any of the providers that I have. It is oftentimes not during business hours. It's either very late at night, very early in the morning, in between flights. You know, it could be erratic on.

[00:36:20] Michael Pupil: It's a little bit all over the place. So working with providers that have a very, very flexible schedule for me to communicate and interact with is an important feature for me specifically. And I'm just talking about that type of demographic. There are certainly Others that you know, would, would probably differ, but that customer experience or that member experience for me is an important one.

[00:36:42] Michael Pupil: Now, obviously with AI. depending on what information it is feeding me. We have all gone through the experience of, you know, making a phone call and then instead of having to route or listen, we just start repeating operator or agent or help or person. [00:37:00] And, you know, we're not even listening to what the prompts are.

[00:37:02] Michael Pupil: We just want to force our way to get to, you know, another human being to have that sigh of relief so that I can take care of the problem. That's not to say that. Had I have listened to the prompts, I probably could have resolved my problem, but I just felt too frustrated in that moment that I just need a human being and felt that, that, that cover.

[00:37:22] Michael Pupil: And so you know, there, there could be the, the bad part of it. And the ugly part is. We don't really know what the answer is going to be until we start seeing the data that comes through. And so what if this was an application for a loan where I wanted a mortgage and it just tells me, no, you've been declined without any context, without any education, without any, you know, sympathy or empathy as to, as to what's going on.

[00:37:45] Michael Pupil: And what if it does the opposite? What if it gives me access to something that I probably shouldn't have access to and that is going to hurt? And so there is a. There is a double edged sword to this. And so I think the customer experience is managed with [00:38:00] the analogy of the good, the bad, the ugly, depending on what it is that you are trying to attribute this to and how is the answer going to be coming through?

[00:38:08] Michael Pupil: And that's probably why we spend so much money training our teams that are customer facing or member facing to employ all of these things. Great question. How are organizations training AI systems? With client information. Well, that's a, that's a good one. Again, we, we talked about how AI is only going to be as good as the data that you were feeding it.

[00:38:33] Michael Pupil: This is going to be, I don't think that there's a one size fits all. So apologies to the person asking this question that I'm not going to be able to give you a black and white answer because I think it is completely dependent upon each organization as to what are the data sets. That they are feeding their AI tool in order to come out with their conclusion.

[00:38:52] Michael Pupil: On that ugly slide, we talked about CFPB making some recommendations about being transparent as to how you are using AI when it is [00:39:00] being used. And of course, what are the data sets that are coming into that? I think each organization measures their risk analysis as to what data points they are using.

[00:39:10] Michael Pupil: Again, if you are using the entire worldwide web to just crawl and take. Data, I would probably say that there's going to be a lot of misinformation that is out there and we can get into a debate about misinformation and what that kind of triggers all of us to believe. But not all things are true. And that is a really tough statement to make because.

[00:39:32] Michael Pupil: people have different points of view on the same item. So it is, in my opinion, still one of those gray area matters that I think are only decided on each organization, which is why this, this topic is so incredibly important. And I can't wait to see how Hollywood, you know, kind of flips the script you know, so to speak, pun intended as to what that would look like now post negotiations and what their agreement is going [00:40:00] to look like.

[00:40:00] Michael Pupil: And while Hollywood writers are not necessarily what it is that we're looking at today. I think there's going to be some lessons learned in terms of what the fears, risks, and resolutions are to a lot of those clauses. Third question here, what do we need to consider before implementing AI? Everything that we're talking about here.

[00:40:21] Michael Pupil: And I would I think there's one thing that I truly value in the credit union space, and that is that credit unions are very, very open to sharing information with other credit unions, more so in that space than I think in many others. There's lots of, you know, industry conferences for banking and lots of industry conferences for telecom and utilities, for sure.

[00:40:42] Michael Pupil: And everybody does share, but I think that there is an inherent Bond of credit unions of sharing their experiences. You know, both good and bad which I would highly recommend that if this is coming from a credit union you know, seek [00:41:00] out associations or events that are that are taking place in or around your area that you can get to and ask the questions.

[00:41:09] Michael Pupil: You know, of course, this was not meant to be, you know, necessarily a commercial on Lexa, but more of a, a high level overview of you know, AI machine learning and what we're seeing in terms of trends. So if you do want to we'll flip over, you know, some communication afterwards that we can, we can chat and maybe we can get you in contact with a couple of people that we know that are, are doing some funky stuff.

[00:41:33] Michael Pupil: All right, well, we're coming up towards the 45 minute mark here and I wanted to kind of conclude this over, but if there are some some questions out there, we'll probably be putting out some answers or you can reach out directly to me. I'm pretty easy to find that QR code is back on that right hand side of the screen that if you missed it the first time you're able to get to there.

[00:41:58] Michael Pupil: It was a pleasure to to speak [00:42:00] with or to speak with all of you today. Hopefully we've all learned a couple of things and look forward to speaking with you guys again soon. Thanks so much everyone. Bye bye.