Transcription:
Daniel Wolfe (00:08):
All right, so why don't I get started with just setting the stage here. So this is a different sort of format from what we're doing with most panels, and you'll see that I have only one question for the panelists. And the reason for that is that I want to involve all of you in the discussion at your tables. So before we actually even get into that, I want Michael Moeser from Arizent Research, a colleague of mine to set the stage. If you were here yesterday, you saw his full presentation on some research that we're publishing in May. This is just one slide of it, but I figured Michael could tell us a bit about industry-wide, what we're seeing with AI.
Michael Moeser (00:43):
Thank you, Daniel. So this is part of the research that was presented yesterday in terms of the emerging emerging payments tech. And so we arise in research sponsored by I2C, went out and surveyed approximately 125 bankers across the United States. 83% were banks, 17% were credit unions. And we asked the question, in what areas is your institution using or planning to use generative AI based tools to support your payments business? And what we found is that 83% almost all were using or planning to use gen AI in some form, yet we did find a small percentage about one fifth, almost 17% had no plans and currently were not using gen AI based tools and the most common areas. And I think one of the reasons Daniel liked this slide, and it really gives that food for thought in terms of where are people using it.
(01:48):
You can see that fraud risk management is the number one customer service being number two almost evenly. And part of that may have to do with the fact that predictive AI and other forms of AI are used in those functional areas. But then as you work your way down in terms of things like underwriting product development, I think somebody mentioned here being very low on the totem pole, there are certainly factors that may lead to that being either higher or lower at your organization. But I think the thought process here is these are very common areas that are either utilizing or organizations are planning to utilize gen AI based tools.
Daniel Wolfe (02:33):
Thank you, Michael. So the one question, again, this totally different format, a panel discussion with only one question. The one question I have for each of you other than please introduce yourselves is how have you either in your organization or something you've witnessed that stuck with you used AI to solve a problem? Jody, you don't get away from being first by sitting farthest away from me, so please introduce yourself.
Jodie Kelley (03:01):
Hi, I'm Jodie Kelley. I'm the CEO of ETA. I hope you all know of this. We are the trade association that represents the payments industry. We actually did an AI focused event recently, and I have to say that the piece of it that captured my interests the most was related to small business and how generative AI is used to support small business. One of our member companies has a small business tool that incorporates payments obviously, but also incorporates things like a marketing component to help small businesses run. And the marketing uses generative AI to draft emails, marketing emails for small businesses to generate images for small businesses. You just say to it, draft an email to my customers advertising this event and it does. And the feedback was extraordinary from small businesses who are spending all day every day just trying to run that business and make a living and don't have time or expertise in an area like marketing. And so I thought that was a simple yet elegant example of the way that gen AI can be used in our industry.
Katie Whalen (04:12):
Great. My name is Katie Whalen. I lead the North America issuing business for Fiserv. And we've actually been looking at a number of, have been using ai, a number and machine learning in a number of different areas of our business over the course of the last few years. We started with leveraging the Mave or Watson's program into our IVR program. We also have deployed AI and machine learning into our fraud and analytics tools for fraud detection on transactions to adjust strategies based off of different data inputs. But more recently we've pushed harder into the customer service area and we work with some of the largest banks in the United States and are an extension of their business as we process for their accounts that we host on file in our data centers. And so we are very much of high touch white glove engagement service with regards to customer service.
(05:09):
How we engage with clients, we get a question and then have to go look at documentation to be able to answer that question. And so what we've done within our team is really loaded a lot of our doc, if you will, or documentation for how we do things on our platform, technical documentation, servicing documentation. We have thousands of pages of this doc that we use for various different kind of areas of managing portfolios for cards. And we've loaded that into the database and we're now applying gen AI on top of that in order to kind of mine that data for kind of information to be able to present it accordingly. And so we ran a pilot at the end of last year in the fourth quarter where we loaded about 10,000 pages of documentation that's kind of complicated, different set of settings, how kind of different calculations work with regards to credit card portfolios and started mining that and received as part of our pilot, about 85% of the answers came back that were in line with what we would be expecting and 36% came back with exact perfection from an accuracy perspective.
(06:17):
And so as we start to think about the models that we're leveraging and how we can continue to refine the models for how we can mine for that information to then make our servicing even more faster, more efficient, and then also then the training, the models on the type of tone and language that I want our teams to respond by to kind of increase that engagement with our clients in a more positive way has been really where we've been looking at. And then there's a number of different things that you can offshoots from that experience that we can also look at from there as well. But that's where we've been really focusing to enhance the customer experience and client satisfaction of how our day-to-day interaction works within our teams.
Daniel Wolfe (06:59):
Cathy.
Cathy Beardsley (07:03):
Good morning everyone. My name's Cathy Beardsley. I'm the president of Seg Pay. If you don't know who we are, we're a payment facilitator. We're providing payment services for e-commerce sites. So we're in the baby stages of AI in our organization. We're using machine learning today as part of our risk management tool set to look for patterns we might not be able to pick out. But where we really see some benefits from AI is adding it into the customer service team. So one of the features we provide for our merchants is we are their first line of defense for end user questions and today we get oodles of tickets and we have three ways for those consumers to reach us. It's either by phone, email or chat. And the majority of consumers go into that chat area and wait for an agent to come back and help them through AI.
(07:55):
We could actually be addressing their concerns much quicker, not waiting for an agent. It would allow the customer service reps to step up and help more complicated, help them more complicated issues and provide better level of support there. The other area that I think would be interesting is really in our sales team, we get oodles of leads in that they have to comb through. Some of them are junk, some we just can't take. So if there was a way for us to kind of weed out those sales leads and either give a yes or no real quick, that would make my sales team much more efficient. The other area which I thought the team would be most excited about was risk management. So they are interested in going past machine learning, but there was a fear on the risk management side is that could the sink take a life of its own and start declining good transactions? And so I think they're going to take it a little slower than some of the other departments within the company.
Michael Moeser (08:54):
Alright, Daniel, could I ask a question here? Since Katie talked a lot about what they're doing in terms of their utilization of IBM Watson and it sounds like you're a little further along the path than other people and very curious to see what kinds of obstacles that you either have faced in your implementation of Gen AI and maybe the hesitancy and how you overcame them. Because I think a lot of the people here in the audience, they're probably early on in their journey as a lot of banks and credit unions are, and sort of interested to hear what's been your experience in terms of that, either pushback and how you've overcome it. And I realize you probably, there's still pushback going on, but what have you faced and how are you dealing with it?
Katie Whalen (09:44):
Can I answer a second question? No, go ahead. Go ahead. So it's a great question, Michael. I would break your question probably into two parts. The first is kind pushback from the teams, the adoption within, and then the second is around what are we doing to just drive it within the business. On the first part, we talked a little bit about this in our panel yesterday, but the way we've done this as part of a pilot is really kind of identified individuals that can be evangelists within our organization and be part of that pilot. So it's a group of about six people that we've identified that have the right attitude are change oriented minded, want to be kind of part of this and I think and kind of serve as leaders within the team. And we've identified them as part of that pilot and as we've been going through, they in turn see the benefit in their day-to-day, and it's really about change, behavioral change management.
(10:41):
And so they're adopting it into their day-to-day practices. And then coming out of this pilot as we widen the use cases and adoption within the team, they then are the ones that are teaching their teammates to kind of adopt the product and can be really evangelists for the benefits that they're seeing in their day-to-day. So that's the approach that we've taken and that's really quite amazing when we start to actually show in real examples of this is an example of somebody inputting into gen AI, this is the outcome, the sparks in people's brains start to go off and the gears start turning over like, oh, I could use this for X or I could use this for Y. And it becomes actually a really great dialogue because it's not just that one use case, but people start to think about all the different applications that they can have in their day-to-day, which I find to be really encouraging and something that we've seen to be really impressive.
Michael Moeser (11:35):
Super. That's really helpful.
Katie Whalen (11:37):
That's helpful.
Michael Moeser (11:38):
I don't know if it looked like we had a couple other thoughts from Cathy and or Jodie.
Jodie Kelley (11:48):
We are going to add, but we're all looking to you to see if we're allowed because you had said one question.
Daniel Wolfe (11:54):
I mean this is Michael's show now.
Jodie Kelley (11:58):
I just add one quick thing. When we look across the breadth of our membership, what we see is a lot of experimentation with gen AI, a lot of interest in ways in which it could be used to drive efficiencies, but proceeding with great caution. Ours is a very heavily regulated industry. There are a lot of questions around gen AI when you get to something like decisioning, like big questions, right? So driving efficiencies, coding, customer service for sure. And some of our member companies have hundreds literally of experimentation paths running. But I would say moving cautiously and finding that, although it can do a lot, it's still in the early stages and there's some surprises that come up along the way. So lots of opportunity there, but I think it's going to take a while to really fully realize.
Daniel Wolfe (12:55):
Did you want to, I'm good. Okay, go for it. Also to what Katie said about being an evangelist, I will say that that does help humanize AI to a lot of people who see only the headlines and Terminator movies and stuff like that and are afraid that it's out for their jobs. In our own organization, we had a session on AI and journalism and the most skeptical cutthroat reporter that I know was sitting in the front row and was just making sure that the person presenting was just taking no liberties, had no wiggle room, had to just answer every single one of her questions. I'm thinking this poor person presenting has just been undermined in every way. But by the end of it, she actually was on board with it and was asking me, I wasn't, the presenter was asking me how I could show her how to use AI to do things on her own.
(13:50):
So I think just even having the ability to just be in the room with that sort of thing and with somebody who really believes in it and understands it is incredibly helpful to getting it going in your own organization. Before we move on to the next one, just quickly, I had an example I wanted to share from our own reporting, which is about Commonwealth Bank. We wrote a story about them last year in Australia. Commonwealth Bank of Australia used AI when it noticed that a lot of folks using its digital payments system were using the messaging field, the memo field, to send abusive language. And this is something that they found was not really the sort of thing you could find with keywords. It wasn't just people peppering profanity in there. It was a lot more subtle. You could really tell only that this was financial abuse, elder abuse, domestic abuse by seeing the pattern.
(14:41):
And so that's what AI is good for is looking at a large set of data and seeing the pattern. And they've found a large number of instances where they saw evidence of abuse when they were able to apply machine learning, large language models, other forms of AI to that. And they opened it up as well that you can go online and find the tool that they built with their partner, which was H two.ai, to apply that yourselves if this is an issue that you're having within your own organization. So all these different examples, all these creative examples. And what I want to do now is just kind of metaphorically turn off our mics up here and I want to ask you all at your tables, and if you're sitting by yourselves at a table, you may want to join another table. Or if you're sitting only with people, ask yourselves the same question, ask each other the same question, we'll take about 10 minutes and we'll join you as well to say, Hey, how have we seen a really impressive use of AI to solve a problem?
(15:47):
Or maybe I have a problem, maybe there's a problem that was discussed here that I would solve a different way. What are your reactions? And then in about 10 minutes we'll ask you to raise your hands and share what you've learned. And after that you can just keep networking, exchange information, maybe solve some real life problems after the event. So it's over to you all about 10 minutes. And I hate to interrupt everybody's conversations, but I'm hoping you can share now we can intermingle some of our, you're not even stopping talking. Wow, this is amazing. We're definitely doing something good here. All right. I'm going to hand it back to, all right, share with the room. Share with the room what you've learned.
Audience Member Mike (16:37):
Okay. Introduce yourself. My name, whoa. My name is Mike. I'm from Premier America Credit Union in LA and Ventura County. And so we are talking about the customer service aspects to it. We are looking at adding it to our frontline team members, call center and branches, adding in all of our procedures, making it so that when someone has a question on how to do something, how to open a cd, how to basically do anything where they would normally have to dig through a procedure on our intranet, they could ask the AI. The AI will be able to know exactly how to do that based on our procedures and that it makes the team member more knowledgeable and able to give out quick information. We're also looking at adding it as a sales tool so that as the call is in play, the AI can suggest based on that individual customer or member what products or services they'd most likely be interested in.
(17:43):
So it speeds up that time to be able to make the cross-sell and to be able to add different products. So the main thing that we're looking at though is making sure that we stay within the walls of the credit union and that information doesn't escape and go into the larger LLM so that our procedures aren't out there. One of the things we also spoke about, which was really interesting, is being careful to use AI to make decisions for lending. You don't want to have the AI make decisions that are going to cause your institution to suddenly be discriminatory. And that's because you can't control. It may make decisions where it's just going straight off of what evidence it has in the system. And then if you're not paying attention, suddenly you're not making those decisions and you're discriminatory and outside of the regs. So some really interesting things.
Daniel Wolfe (18:45):
Cool, thank you. Who else is brave? Any brave ones here? Sure. Alright. I'm in the brave corner of the room. Introduce yourself please.
Audience Member Kelly Hobbs (18:57):
Sure. My name is Kelly Hobbs and I'm with Value Dynamics, which is a Collinson group company. And our focus as value dynamics is card LinkedIn affiliate offers, just ways to earn and burn within bank and other programs. And first and foremost, I would like to say amongst this group at this table, one of the things that I just took away in listening and sharing is that sometimes AI can be sexy and sometimes it just serves very operational and fundamental purposes. And all the governance and ways that you want to use AI can look and be very different. Our organization uses AI to provide very meaningful card LinkedIn affiliate offers to consumers. So we are connected with networks such as Visa, MasterCard, Amex, and so forth. We receive transaction data, we cleanse that data through our machine learning and algorithms and AI play a role in making sure that when you as a consumer are a member of a loyalty program that you're actually receiving cashback and other offers that means something to you. Historically, those offers have been what I affectionately call the spray and pray approach where it's a very merchant focused approach and it's like, Hey, let's just get offers out there. But we have flipped that on its head and through AI, hopefully you all as consumers in your own programs are receiving offers that really mean something to you, even maybe to the point where you're like, Ooh, this is scary. I didn't know that they knew somebody knew in the stratosphere that I needed that. So that's how that works.
Daniel Wolfe (20:47):
Thank you. I see another hand over. Oh, I see. I'll make my way to you next.
Audience Member Rahul Dave (20:51):
Hi, my name is Rahul Dave. I come from the private equity background currently working for a firm called Zodianto. So there are three use cases on which we are currently working the first, the fourth largest bank of North America where the CEO reach out to the finance team. If there is a change in the interest rate, which particular product lines are going to get impacted and I need to know my revenue projections at the real time. It is difficult for the finance team to come back with those predictions right away. And that is where this entire discussion started. How can we create in-house LLM, where my executives can go on their mobile apps and just simply ask the question, they know what is forecasting will look like? Where are the risk controls governance from the every single line of business perspective? The second use case, which we guys from the P side come across multiple times, one of the biggest custodian, they are trying to solve the problem of creating a next generation advisor experience for their broker dealers and RIS.
(22:07):
Now they build this AI component, which is a companion AI, which look at the experience of every single advisor their day-to-day, which kind of processes, what systems they're using, et cetera, and come back with a recommendation. They just launch a new platform two months back, it is by the purchasing X wolf, you can check this, which has the companion AI in place. The third and probably the final is this particular bank is ingesting a lot of data, thousands of documents every single day. So AI piece what they have built now, and it is there by the dock, LLM, you can probably Google it and the name of the bank, it's the largest bank in North America right now. They have done this entire document ingestion classification and the entity extraction. So if say Apple is there in thousands of document apple ink versus Apple Corporation, how can you identify and say it's a one single entity and then every single information from the various systems, it is able to pull and give you a single data point. So these are the few use cases which typically we guys come across.
Daniel Wolfe (23:19):
Thank you. And you'll be oh applause and you'll be the last one just because of time. But go ahead and introduce yourself.
Audience Member Virginie (23:30):
Hi there, I'm Virginie from Tolomeo. We are a bank in Puerto Rico and we've been experimenting with generative AI for the last six weeks, I would say. We've asked actually a company to help us anonymize everything that we put into the agent so that we wouldn't have client data ending up training the model. So that is the first thing that was very important to us before we allowed our people to use it in their day-to-day. And what we're finding at the moment is that because our people do not work in their native language, they use the agent to translate everything that they write to enhance their English text. So that is the first thing. They also use it to summarize text to amalgamate text data into, for example, shorter KYC profiles. And the approach that we've adopted at the moment is to let people experiment with it so that in a couple of months we determine the best use cases to increase their productivity. Because I think that like all of you, we don't know what it's good at and we're using this pilot phase to really determine what to use it for. What we're finding is that our people really like the fact that the company has put the tool that they were using in disguise on their own devices onto work devices in a confidential sort of way.
Daniel Wolfe (25:20):
Alright, well thank you everyone for participating and for sharing your insights and what you've learned. I hope you've made some new friends along as part of this process or at least colleagues and just stay in touch with each other, stay in touch with us. As I mentioned before, the research that Michael presented, that's going to be published in May. So keep an eye on American banker.com for that and give yourselves a round of applause. And thanks also to the folks who are brave enough to be on stage with me.
American Banker Interactive Workshop: Solving problems with AI
April 12, 2024 10:54 AM
25:58