How SMBs can Use GenAI to Unlock Business Opportunities

A high-level bank executive will discuss how generative AI (GenAI) will empower small business bankers as strategic advisors, giving them the ability to automate repetitive tasks, allowing them to instead focus on providing personalized customer insights, streamlining loan applications, analyzing market trends to identify potential opportunities tied to a specific customer's market and needs, and offering customer-centric loan products and services to help small businesses prosper.


Transcription:

Penny Crosman (00:13):

Welcome to the second session of our event today on how banks can use generative AI to support small businesses. I'm Penny Crosman, Technology Editor at American Banker, and I'm here with Brianna Elsass, who is Head of Digital Experience and Technology for BMO's retail and small business segments. Thank you for your patience. You have an interesting role in that you are creating the technology that's used to interact with both consumers and small businesses. How does that work? Are you able to take some of the things you're building for consumers and apply it to small businesses or the other way around, or can you build one platform for both segments sometimes?

Brianna Elsass (01:00):

Yeah, and thanks for having me. It's an interesting question, and actually it's one that we've built in a grouped fashion so that we are actually able to get a little bit more economy of scale for our small business customers. So we do have one app that we leverage for both our small businesses as well as our retail customers. We have a separate instance when it comes to the higher commercial segment customers, so they actually have a holistically different experience. But what that does allow us is when we are building technology, we have the ability to scale in some of the features that we're able to build. We could build it for the consumer first and then scale to small business, or in some fashions, we build for small business first and then we scale it back to the consumer customer when it makes sense from a use case perspective.

(01:58):

So as an example, Zelle is an interesting one. The use case, the business case for that made more sense when you look at the broader retail customer segment and the general consumer leveraging Zelle. But we were able to take that once it was implemented and then continue to expand it so that our small business customers were able to leverage the functionality as well. And when you think about the world post-COVID, you have so many folks that have a bit of a blended life themselves between their side gig, whatever they do in a gig economy, that they're blending their personal finances and their small business finances. Making sure that we're keeping, from an experience perspective, that holistic customer need together and how they work together is, we think, a pretty critical way of looking at how these small businesses need their services.

Penny Crosman (02:58):

So just before we get deeper into this, how do you guys define small businesses? I know banks tend to define the category differently.

Brianna Elsass (03:09):

We tend to look more around the revenue and the lending side, looking at when you need a line of business credit up to a million dollars or $5 million in net revenue. That encompasses our small business segment. Above and beyond that, you start to go into the mid-market and the commercial space that actually is more centralized up in the commercial space for us from a servicing side.

Penny Crosman (03:39):

So when you think about the small businesses you work with, what are some of the ways that they are using generative AI either to run their business or in their personal lives? As you mentioned, it's so intertwined. Are you seeing some interesting examples out there?

Brianna Elsass (03:56):

A lot of it is very dependent on the small businesses themselves, and I think this is a bit in the same iterative learning fashion that we, as individuals, are starting to play around and get comfortable with it. So where there are general tasks that they can start to leverage AI to automate a bit better, where they're looking at enhancing their marketing capabilities, their overall sales funnel, where they're able to prompt with maybe better descriptions for their products or, if I'm a real estate investor, the property that I'm listing, how do I pull in more renters, more customers? How am I marketing this? The influencers? Where am I putting my content out there for customers? AI is starting to weigh in very heavily with that and where I'm able to push my content out to customers in a broader fashion or even automate the tasks that were very manually heavy before and take that and streamline it.

(05:02):

I think there is a large portion still that as our small business customers themselves are getting used to is where am I still the one that holds all the knowledge, all the value, all the insights on how I run my business, and where am I able to get this tool to help me do my job more efficiently? And I think that's actually where in a lot of places banks play as well. We become that tool to help them be more efficient, be more effective, help run their business better. And so they're starting to evaluate AI in a similar fashion to how they've historically been expected to evaluate sometimes their finances and how it's helping them accelerate their growth and what they need to do on a day-by-day basis.

Penny Crosman (05:52):

So when you think about serving those customers, how do you think about incorporating generative AI into that, for instance, in the way that you interact with them?

Brianna Elsass (06:04):

There are a lot of ways that we're evaluating AI and how we service our customers. We're having to, as a lot of banks are having to do right now, balance out how much we actually have customer-facing and interacting with customers, the compliance, the privacy, all of the regulatory things that we're balancing in that mix. So we're lightly stepping in when it makes sense to put something out in front of customers so that they're interacting with it. But oftentimes it's on our side as well. We're evaluating the tasks that we can do faster and more efficiently on behalf of our small businesses, the customer questions, and making sure that we're empowering our service agents when they're having the conversations around loans or best-fit products, processing those loans or those items, and looking at the tools, the tasks that we can automate and streamline as much as possible, all the way down to as we look at even our ability to accelerate development and what we're able to put into production on a faster basis so that we are giving the customers that next best feature that they want and need in order to serve their financial needs.

Penny Crosman (07:35):

So are you still in the piloting and testing and idea generating phase of this, or have you put some of this in place already?

Brianna Elsass (07:45):

A lot of it is still in the piloting and the testing phase for us as we just get more comfortable with making sure that we have the right privacy guardrails in place, that we have the right balance. We don't want the hallucinations on the AI front around some of the information, so doing a lot of validation that what we're able to put forward, the insights we're able to share with customers, the ability to serve up their personal information to the agent who's having the conversation with them in the right fashion, makes sense that it's all being checked and balanced. That's a lot of what we're doing now. When we are able to get to a point, we have a lot of ideas that we're looking to kick around in terms of actually enabling the customer to do the interaction themselves and hopefully streamline a lot of their activities.

Penny Crosman (08:43):

I remember you came to our digital banking conference, and one of the things you talked about was providing chatbots to customers, not just small business but consumer and small business customers, and yet trying to make those interactions feel natural and human and not making it feel like just a static robot kind of thing. Have you figured out how to do that?

Brianna Elsass (09:10):

On some levels, AI is helping with that as well because it's getting savvier in terms of the tone and the interaction. So it's taking what you would see or feel right now with a lot of the branches in the tree-type flow with chat deflection and really looking at is it doing the answering of the conversation itself and where do we make that leap even to executing the feature, the behavior, the task on behalf of the customer? And so that's one area where we're exploring just how deep down that rabbit hole we want to go and how quickly. So I think the chatbot, the deflection, is probably the baseline that a lot of folks have started to dip their toe in around, looking at really that next iteration as we get the rigor in place, the privacy controls in place, all of that, and where we're able to continue to dive in there.

Penny Crosman (10:20):

When you say deflection, can you explain what you mean in case people don't know?

Brianna Elsass (10:24):

Yeah, the deflection, when I say that, it's the engagement prior to getting an agent on board. So is there a set of really easy FAQs that are out on the public sites that I can leverage AI to go out and grab and populate in real time based off of the question that the customer's asking? Is there a way for me to help them self-service within the chat component itself without actually getting a human interacting with them at this point?

Penny Crosman (11:01):

That makes sense. Sometimes it feels like some of these chatbots, it's like the new generation of interactive voice response: press one for

Brianna Elsass (11:11):

English.

Penny Crosman (11:11):

Press two, that kind of thing. And I know a lot of times when banks are deploying generative AI in their call center, it's kind of like how many calls can we turn into not calls and save the time of people who answer those calls? Well, on the flip side of trying to make these chatbots feel natural, there's sometimes a danger in making them too human. I don't know if you or anyone here remembers when Google demonstrated its Google Assistant in 2018, and they had it make an appointment at a hair salon and make a reservation at a restaurant, and it sounded really human and it said things like "uh" and little filler words. I remember people clapped because they were impressed at how real it seemed, but then there was a lot of backlash afterward about how creepy that seemed that the person answering the phone thought it was a human when it wasn't. Do you ever worry about that, that a customer might be interacting with some kind of bot and thinking it's a person when it's not?

Brianna Elsass (12:27):

Yeah, I do, both personally and professionally, if I'm being honest. I think there is an element, and that's really where I've probably used the words compliance and privacy quite a bit in my conversation, and I think it is part of banks, right or wrong, whether or not customers always give us this feedback. We are that tenant of trust and safety. We are responsible for quite a bit in their livelihood, in their financial wellbeing for these customers, for these small businesses. So where we're able to show up for them when it comes to the safety and the soundness of where we're injecting AI and how that's being integrated and the quality of the information that we're providing through that tool is going to be critical. It is part of, by nature of offering it, it's going to be part of our brand, it's going to be part of our reputation from a reputational risk perspective.

(13:31):

And so really diving in and making sure that the right disclosures are out there when it's, "yes, you are interacting with AI," or where we have dropped something into the experience that allows them to interact hopefully in the positive, right way, but making sure that we're being very clear with this population. As a whole, it's getting more comfortable in different ways. We used to, if we all remember back, there was a time where, for example, personally I love and live in the Google world, so I like it when my Waze is connected to my calendar, and it knows that the next time I get into my car, that's the appointment or the meeting that I am going to, and helps by proactively bringing up the directions. That used to be creepy and scary. As we get more accustomed and as we adapt to how that effectiveness is, how much more efficient that makes us, where we start to expect it, it just becomes more ingrained and more natural for us. We adapt with the technology. So I think there is an element of riding through that change curve with the customers and working with them, getting them through that change curve with us, and being transparent will help. I think there's going to be a lot of interesting things we can do on behalf of customers and small businesses that is really kind of exciting technology to put out there, but we're going to have to work with them in getting them through that pipeline.

Penny Crosman (15:19):

What's on your wish list? What do you wish you could provide that maybe it doesn't seem that easy or possible today, but would, I don't know, help you attract more small businesses, help you serve them better, help you retain them better, whatever it is. What would you really like to be able to do?

Brianna Elsass (15:42):

Well, if we're blue-skying, I think having a level of deep knowledge around what the small business needs are so that I can look at and leverage line of sight to, "Hey, we know seasonality is coming up, this is going to be a busy time. In an ideal world, you would have different lines of credit, you would have a different set of cash flow waiting. Let's proactively seed you for payroll that's coming up in two months that you're going to be short on, and let's extend you a line of credit because ideally around this time for the busy season, around the holidays, and we know that the tariffs are going to be coming in." So it's able to pull in the business seasonality, the trends that you have, the cash flow, as well as the overall market indicators that say these are the right set of things we're going to preset you up with without you having to come and ask, because I know you because I'm your partner. And then this is where I think as a partner I can show up best to serve you and have you then have a level of understanding and expectation that I can do that for you and service you in time and make it more of a delightful proactive relationship as opposed to reactive.

Penny Crosman (17:12):

So really understanding the customer, what they might anticipate, what they might need at a certain point, offer them a quick help, a quick loan, quick help of some kind just when they need it. That totally makes sense. What are some of the things that get in the way of that? I know a lot of banks have legacy infrastructure, they have data silos. It's hard to get the right data at the right time in real-time to make all that happen.

Brianna Elsass (17:43):

I think those are all really great roadblocks, and I don't know that without the banks going through and restructuring things, you're going to have to figure out a way to bridge the silos without restructuring. That's very costly, and it's a lot of time, and you're going to lose momentum with where you're trying to achieve. And so is there a middle layer data lake? Is there a way that you can build a bridge across channels so that you're able to establish a way to leverage the AI or to leverage the information across multiple different silos to pull it together in one unified way that you can offer them up regardless of the channel that they're interacting with you? In theory, you should be able to do that without having to completely overhaul your backend, but it will take some dedicated delivery structure of data, structure of how you're building your technology that will take a little coordination, but it is much easier and much more achievable than restructuring your entire backend.

Penny Crosman (19:00):

That's something a lot of banks have wanted to do. So are you talking with different data management vendors? Is that the kind of route you need to take, and then you can layer the generative AI on top of this middleware situation?

Brianna Elsass (19:17):

Yeah, you can definitely go that way and look at data management vendors. You could look at is there ways that you can even start to restructure components of it in your own internal system? You can create a temporary lake or a temporary repository just around specific components that you want to start to build some of these agents around to be able to service specific needs so you don't have to boil the ocean all at once. You could start with a set of tasks and start to nuance it that way.

Penny Crosman (19:54):

You mentioned a minute ago these agents, an agent going from your calendar to Waze and different apps that you have and so forth. I'm actually working on a story right now on what are the dangers of over-reliance on AI for banks. One of them that a couple of people brought up to me is the idea that using agents could compound errors. One agent has an error and then hands off to another agent, which introduces that error somewhere else and so on. You have this compounding effect. Is that something you worry about?

Brianna Elsass (20:32):

Oh yeah, for sure. That's where I think the checks and balances come into play, making sure that you're doing that due diligence around what's the subset that you're able to pull. Are you able to validate it against the data? Are you able to compare it to the expected results? And doing that level of ongoing evaluation is going to be critical. We're going to get a level of familiarity, a level of reliance, and you could very easily start to propagate even bad data. If you're leveraging data based off of the agents, and you're starting to propagate bad data, you're going to have bad results coming out of it. So that is going to have to be something where in the next iteration into the future, data quality, data remediation, agent remediation, agent quality, all of that is going to be a very, very critical component of the overall success on how we leverage AI holistically.

Penny Crosman (21:46):

I mean, I think for some of those reasons, some banks have been kind of hesitant to use agents at all, like use an autonomous bot that's interacting with the generative AI model because of the uncertainties, potential uncertainties around that. Where do you guys stand at BMO? Are you kind of ready to go with deploying agents throughout your bank, or are you a little more like, "Let's keep testing this for a while and make sure this checks every compliance box before we start unleashing it"?

Brianna Elsass (22:21):

We're not ready to go yet. We are working towards getting the right agent at the right moment with the right scenario in a spot that we're comfortable with and doing the validation, because you don't want to—it goes back to that reputational risk. There needs to be a healthy balance between what level of pushing forward with innovation and then making sure that you are taking the right safety, security, privacy, all of those really good practice nuances into consideration and making sure that they're in line with where you want to go from an innovation perspective. But we don't want to fall to the point of we're so afraid of the innovation that you don't actually start to become familiar with it, that you're not actually aware of it, that you're not testing into it because that's detrimental as well, right?

Penny Crosman (23:21):

For sure, for sure. It's a balance. You mentioning reputation risk reminds me of the Air Canada case, which I'm sure you know about, where

(23:30):

Somebody wanted to go to his grandmother's funeral. He asked a bot, "Will the bereavement discount apply if I apply for it after my flight?" And the bot said, "Sure, you have nine months or something to file for the discount." And then when he tried to apply for the discount, he was told, "No, that's not our policy." So the bot said one thing and the actual policy was something else. It seems to me that that is a risk for anybody where you've got a policy that gets updated or a policy that's different in different places. I feel like that's the kind of thing that could happen really easily. What do you think?

Brianna Elsass (24:17):

Yeah, I completely agree. And that goes a little bit back to the bad data in and bad data out. Is there ways to start to contain or make sure that you're curating the data that it's really being fed so that you're able to provide the right outputs on the specific nuances? I don't know from the Air Canada perspective where it was gathering data if there was an old policy that it reached out to versus the current one that was in place. But if you can contain this specific agent to "it needs to only be referencing the FAQs and the information that's out on the public site with this set of the repository. This is the only repository that you can go and leverage for data that you can culminate and pull together in an FAQ back out to the customers," that starts to help those guide rails, right? Then yes, you still need to do the due diligence around, "is the information still up to date? Is it still the actual, the right information?" But then at least you can start to try to structure where all of that is coming in and what it's being pulled from. That makes sense. That's a lot of those caveats out there: "It may not be the most accurate and you should always do your double checks and all of that kind of thing," once we start to leverage it.

Penny Crosman (25:43):

If you're limiting it more to FAQs and like a standard set of scripts, then aren't you losing the advantage of generative AI because you're basically doing a rules-based pull of existing libraries of information? So it seems like you don't really need to generate anything. I don't know.

Brianna Elsass (26:03):

No, you're

(26:04):

Right. I would say it's the difference between where we all aspire to get to versus the training wheels approach that we may have to step into first around some of these. I would say it's slightly easier. There is probably an element of life and death when it comes to the airlines, but if it's the help on the front end, it may not be as detrimental than if we give you bad information as a bank around your finances, or if we tell you that, "No, you've got enough credit, you can go out and you can purchase whatever," and that's not actually the case. That is a moment in time or a potentially detrimental item of a far bigger significance than, unfortunately, the gentleman in his bereavement discount for the airline ticket. But I think that's the balance of figuring out the right places to tiptoe into the experiences so that we can test and learn, get it out there, get more comfortable with it, actually help teach and train and coach with not only internally our policies, our procedures, our evaluation, but our customers and their familiarity, and where they trust AI to be and where they don't trust AI to be, and really garnering that kind of negotiation between where we are and where we could be.

Penny Crosman (27:29):

That makes sense. And people in the audience, feel free to type in questions if you like, and they are anonymous, so don't worry that we're going to identify you or something if you just want to ask a question, and there are no dumb questions also. So to what extent, when you're thinking about this and how you're going to create the new experience and how you're going to serve clients more efficiently using generative AI, to what extent do you think about the competition both from rival banks and from fintechs, and do you feel like you constantly have to look at what other people are doing, and is that motivating? Is it threatening? How do you look at that?

Brianna Elsass (28:12):

Well, the answer on that question is going to be a bit personalized, independent. I personally love competition, so I see it as engaging and interesting, but not everybody enjoys competition quite the same level as I do. I think, for me, where I'm looking at what other banks are definitely doing, I am super curious just because it's a really good watermark around the industry's comfort level with where we are going with AI. And I think that's helpful for us to have a broader perspective on where our colleagues in the same industry have been able to get to. Where I spend a lot more time in terms of evaluating is where have the Amazons and the other companies, the other fintechs, the other bigger groups are going to get and push AI at a much faster pace than what banks are going to do. And they're going to do a lot of that pre-grooming for customers way before banks get around to it.

(29:20):

Not to pull back on the whole COVID situation, but people got far more familiar with being served up the right customized solution based off of our TV watch history with Netflix and all of those kind of things at a way faster rate. Personalization moves at a far faster speed because of COVID and because our day-to-day use and familiarity and needing that next curated piece of components. Amazon has done a tremendous amount there as well. So it's really looking beyond the banks when it comes to something like AI because they're going to be faster than us.

Penny Crosman (30:01):

Great. Let's take a question from the audience. This question is, "Given that banks have access to financial data of small businesses, do you anticipate that banks will deploy generative AI enabled banking agents providing real-time financial advice to small businesses?" I think you kind of said you are thinking about that earlier.

Brianna Elsass (30:21):

I think so. I think it will. It's the speed at which and the depth at which that I think will be a bit varied. I think it'll get there for small businesses. I honestly think it'll get there for the individual consumer as well. And it's just how far deep in does that really go? Part of it is going to be the comfort level of the bank and the comfort level of the individual receiving the information on how far they want to push it.

Penny Crosman (30:52):

Have regulators said anything about that, about using generative AI to provide real-time advice to small businesses? I haven't heard of that, but

Brianna Elsass (31:02):

I haven't heard of that as of yet. There are far more in terms of the consumer side of things, and we tend to, at least at my bank, we tend to leverage that as sort of the "is there a healthy gut check" in terms of if we're willing to do this, if we're able to do this for the consumer, does it make sense? Is it the right move to do it for small businesses? Just because there are potentially, and sometimes less regulations around small businesses doesn't necessarily always mean that it's the right way to just jump in head first. Sometimes it does make sense to sort of follow suit with the more stringent oversight as a way to go in.

Penny Crosman (31:48):

And what are, you've mentioned compliance challenges like data privacy and security. Are there other compliance elements that you think about as you're building these things? Do you worry about the potential for bias to creep into an interaction because the large language model picked up some clues from somebody and identified they were a single mother or they were a member of a protected class and a deadbeat or something like that? Do you worry about any of that, like that sort of fair

Brianna Elsass (32:34):

Yes, fair lending type of lending? Yeah, for sure. That's where when we look at where it comes into play, where we leverage AI, that'll be a slower level of adoption in my opinion than on the less possible areas for bias. Anything can have a level of bias. I'm not naive to that, but trying to find the areas that it is a bit more safe, a bit more measured, a bit more guarded so that we are not doing that. We will have to keep very much in mind and do the regular checks and recalibration around the AI agents to make sure that it's coming in with the right level of information, nuances, perspectives, all of that.

Penny Crosman (33:35):

That makes sense. All right. Let's take another audience question. This one is, excuse me, "Are you looking at ways to use generative AI to upskill your small business bankers or branch staff to better service customers?"

Brianna Elsass (33:51):

Yeah.

Penny Crosman (33:52):

Yeah.

Brianna Elsass (33:55):

Upskill is maybe not necessarily the word I would use, more enable potentially. I definitely think that there is a use case around if I'm calling in and Penny, you're helping me with a servicing issue. In an ideal world, if I'm behind the scenes, if I'm running the technology that you're servicing me with, it would be one of those where maybe as part of the verification, it pulls up a summary: "Hey, Brianna, in her last couple of interactions, she was asking about these kind of questions. In the last 30 days, she made this type of transaction history. We project in the future, she's probably going to run into a shortfall because usually something happens and she's ramping up, but she's got outstanding invoices." So here's the high-level summary: "This is Brianna as our customer, as a small business. She's got these pain points, she's got this history, she's got this behavior pattern." And Penny, you now need to establish the connection on what it is that I actually am calling about and needing. How are you best able to service me? How best are you able to solve the need that I'm going with? And so it's a bit of enabling and empowering the agents, or the customer service representatives, to be better suited or skilled in having the conversations that they're already really good at. So it's less the upskilling and more the enabling.

Penny Crosman (35:37):

That makes sense. And do you think to do that you need to aggregate data from other banks and credit card providers and such so that you can get that full picture of not just the accounts that somebody has at BMO, but their accounts at Chase and Wells Fargo and Venmo and whatnot?

Brianna Elsass (36:00):

Now we're dipping into open banking. I think there is a possibility, and I think the quality of the insights that we could provide would improve. When we talk about a customer level of comfort around how much information am I exposing to you, how much do you know about me without me really opening up the curtains before I get creepy, that's going to be a whole other level of learning and working with our customers to get comfortable with. So I think open banking has a tremendous amount of value that we, as banks and institutions, could provide to our customers. If we know the holistic view, we are going to need our customers, those small businesses, to be able to provide us that, to be comfortable with us looking beyond just our own doors. I think that's where there's a tremendous amount of insight, and it'll be really interesting to watch where the fintechs play. You've got different groups like Rocket Money and other areas that are looking through and coming through subscriptions right now, regardless of the FI that they end up paying those from. But you could look at that and take that to the next level even, right? And so it's where we're going to tiptoe into some of these things. I do think there's a tremendous amount we could offer.

(37:34):

I just don't know how quickly they'll invite us into it.

Penny Crosman (37:37):

And if you'll end up having to pay for it, it might change the equation as well. So we'll see what happens with that.

Brianna Elsass (37:44):

We'll see when Jamie goes with that. Yes.

Penny Crosman (37:46):

Exactly. So we were talking about upskilling a minute ago. A lot of banks are kind of pushing generative AI out to the entire workplace and saying, "You should use this for doing rough drafts of your emails and starting your presentations for you and preparing for client meetings and summarizing reports and so forth." They're doing training and trying to get people to be generative AI savvy. Has there been a push like that at BMO?

Brianna Elsass (38:19):

Yes, there has. Again, we're evaluating the right guardrails, making sure that the information that the AI has access to, what is it able to grab, what is it able to get, where are we able to push back some of the conversations as well? So there's an element of that that we are looking at. We are pushing it out so that broadly more of our associates can get exposure, get experience with it. So it is definitely one that we're asking folks to get comfortable with. It'll be over time. It'll be one of those moments where folks will have to look at the,

Penny Crosman (39:14):

I think Brianna

Brianna Elsass (39:15):

The licenses internally, a small cost in the grand scheme of things. Can you hear me?

Penny Crosman (39:21):

Sorry. Yeah, we lost you for just a second, but you're back.

Brianna Elsass (39:23):

Yes, I'm back. Sorry about that. There is the overall economic impact that we will have to take into consideration broadly speaking companies. The cost for the licenses is one thing, but the cost of running AI holistically, the impact on the environment, the economy, all of those other things that we will as a broader society need to take into consideration where we want to push for AI. And I'm sure AI will help us even get more efficient in all of that process as well at some point. So it'll be the trajectory and the growth of this will be an interesting one to participate in.

Penny Crosman (40:08):

Well, yeah, you brought up the cost of licensing some of this technology, and I can imagine it adds up if you're talking about giving it to everybody and even using it within the retail group as you are doing. How quickly do you expect a return on AI, and do you think that using generative AI could make it eventually less expensive to serve small businesses? I know bankers always say it's hard to make a small business loan; it's not big enough to support all of the data gathering and underwriting work you have to do to make that loan. Could this change the equation?

Brianna Elsass (40:52):

I think so. I think so, because a lot of the data gathering, we should be able to streamline the input of the forms and fields. That goes back to sometimes you're leveraging and requiring a dependency on legacy systems that you don't have the interest or the cost of funds to be able to do. If you have a tool then that will be able to accelerate that or make it easier for you, that just helps in terms of the overall cost to serve. So I do think it makes it easier and more effective to be able to service those. So I think it'll start to change the equation. Now, relatively speaking, it's also going to decrease the cost to serve the retail customer and the commercial customer and the wealth customer. So you're going to have, from a proportion perspective, right, rising tide raises all ships, you're going to get an element of that as well. Does it over skew and help in terms of making it easier to service the small business ones? Probably disproportionately compared to the other ones, but there will be an element that it helps all of them overall.

Penny Crosman (42:12):

That makes sense. And sometimes people talk about the concentration risk of all the banks really, especially the bigger banks, using the same set of cloud providers and the same set of AI model foundation model providers. And that there's the risk of one of these companies, these hyperscalers or foundation model providers, experiencing a data breach or an outage and having a domino effect on everybody. But it seems to me there's also a risk of a foundation model getting something fundamentally wrong and then everybody picking up the same wrong information or wrong approach to something. Do you think about that, that idea that everyone's using the same stuff?

Brianna Elsass (43:10):

I'm less worried about that, to be honest. Maybe that's a bit of a naive piece on my part, but it's going to depend on how much is bought versus built. There may be a set of foundations that people leverage to start, but my guess is you're going to have a different level of AI adoption and integration at the bigger national banks versus the smaller banks in terms of their ability to build, their ability to customize. So if it was a foundation model and having a set of bias or a data breach at that level, yeah, you're going to get components that it then becomes exposed and you're going to have an impact, but it won't be like an industry-wide thing. You won't have everybody. Banks are going to want to differentiate and differentiate their services, and therefore their models over time and their use of AI will also then therefore start to become customized and nuanced specific to their implementation.

Penny Crosman (44:29):

All right, well that makes sense. Well, Brianna, thank you for giving us all this time today, and I found this very interesting. So thank you so much. And all of you, thank you for joining us for this session, and I hope you enjoy the next one with the moderator, John Adams. Take care.