Among the things you'll learn:
- Understanding the innovation trajectory: Gain insights into the technologies that transformed banking over the last five years and what trends will shape the industry's future.
- Real-world applications of emerging technologies: Discover practical examples of how banks leverage AI for fraud prevention, distributed ledger technologies, and next-gen architectures to address today's challenges and anticipate tomorrow's needs.
Transcription:
Chana Schoenberger (00:09):
Okay. Just to start off, let's talk about Gen AI, right? When you say technology and payments right now, this is pretty much the thing that everyone's thinking about. How and when are you using it at the moment?
Melissa Tuozzolo (00:23):
I think, and first of all, thank you for having me on the panel. Thank you for coming. Again, this very wide ranging topic, but for us in the Gen AI space, we've actually been using AI for quite a while in the payment space, and I think a lot of our institutions have been the same, where things like auto repairs, things that you can do within the payments flow to try to make things go a bit more seamlessly. Again, that's something that we've all been using for quite a while Machine learning, et cetera. The introduction of Gen AI is quite interesting, and we're looking at it from a few different lenses, right? One from the products themselves and what we can actually use to use Gen AI to enhance what we're offering. So things like being able to predict whether or not a payment will be converted if it goes into a certain market, a certain bank account, if it's got certain identifiers on it, which is quite interesting.
(01:20):
But again, that's kind of closer to that older machine learning use case. Where I find it super interesting being in the service world, so I run our service and account management team globally, is that we're using it in the way that those of you here who have subscribed to ChatGPT, like I do, disclaimer, I don't use it for work, but in my personal life, actually thinking about ways that you can use basic AI capabilities to augment what your people can do and have them interact, have them be much more efficient, build additional quality into what they do, et cetera. So we've been on a technology journey over the last couple of years where we've built the foundation of getting our people all on the same platform. So from a query management's perspective, everybody around the world is on the same platform. And now we're slowly introducing Gen AI into that platform.
(02:12):
And again, it seems relatively simplistic, but things like draft an email, ask a question, summarize a case, summarize a complicated string of emails, things that used to take people maybe like 10 to 15 minutes of their day is now instant. And we're already seeing a huge amount of uplift in efficiency. So even with our baseline pilot users, et cetera, we're seeing about 10% uplift in efficiency and how quickly they can get through things. We're also seeing an uplift in quality. So again, that's just the very beginning. We're also looking into things like agentic, ai, et cetera, which I'm happy to talk about more later. But for us, we really see it as the foundation of the future of how our service organization is going to operate, and then beyond that into other areas of the organization as well.
Chana Schoenberger (03:02):
I had a friend tell me who works as a bank executive that she finds it very useful for performance reviews because there's nothing worse in the world than writing performance reviews for your people. So she'll have a bulleted list of the person's accomplishments, which she gets from their self review and her own notes on their performance, and then she'll just say, write me a performance review using these bullets.
Paul Margarites (03:22):
My wife literally just did this with her performance reviews, not even joking. I mean, we've done it with things like mission statements and general content for decks. It saves time, it saves a lot of time,
Chana Schoenberger (03:35):
And you're supposed to write it to that sort of mediocre tone that ChatGPT has anyway, so it's not like it matters.
Paul Margarites (03:45):
I mean, look, everyone's kind of saying this, but it allows us to really focus on the valuable efforts. Writing an intro to an email is not a valuable effort. Making a decision that's hard and interesting and thoughtful, that's going to impact your customers and your colleagues, that's where we need to be focused. And I'm just going to plus one, everything Melissa said, because I mean, that's the story. I mean, one of the really great things about it just to start before we even get to how we're using it, is it's pushing the conversation, right? It's pushing the conversation on what are predictive analytics, what is machine learning?
(04:18):
What is Gen AI? Because everyone's got some incredible ideas of what we could be doing at the bank. They bring it to the innovation teams and it's Gen AI teams say, can we do this with Gen AI? And they say, this is great. It's really analytics, it's really predictive analytics. What does that mean? Why am I thinking about this way? So it's one-upping all of our skill sets in terms of what is ai, what are the types of AI, et cetera. And then the other great thing that I find about it before we even get to the use cases where're using it for is it's driving creativity. Some of the ideas in terms of how we could use it in the bank that I've heard from folks across the bank are incredible. We've got our normal use cases, drafting paragraphs and things like that, supporting our contact center, which we're using. Those are great, but there's some really interesting use cases around code triage and other pieces that are coming into play that are really bringing people together to rally around these use cases and actually start testing them. And because of the infrastructure, which took some time, you can do that kind of quickly. It doesn't take that long.
Melissa Tuozzolo (05:25):
Yeah, no, just again, plus one, everything that you said. I think the thing that's really driven us to be very prescriptive about what is machine learning versus an algorithm versus Gen AI is our risk profile around things. We have a very strict risk infrastructure and Gen AI is new as well. We should, AI is new, right? There are concerns about hallucination. We have a very strict protocol that we actually don't expose anything that is truly Gen AI directly to our clients. And this is also part of just the ethos of who we are as an organization. We are a people-focused organization. So from the very beginning we were saying this isn't to replace people, this is to your point, to augment what they can do and have them spend more time on the things that really matter for our clients. One of the other use cases we have is creating service reviews for our clients.
(06:20):
And we have clients that bank with us in 40 plus countries around the world, lots of different payment types, lots of different market infrastructures, et cetera. You spend the money to build that foundation to get everything into the same place, and then you can automate the creation of that deck that used to take somebody 40 hours to do so that instead of that 40 hours of pulling together data, which by the way they could make a mistake because it's manual, they can spend that time actually thinking, what can I talk to my client about that's actually going to make a difference in this conversation rather than just presenting them data. So it's huge. But again, I did want to underscore that point. I think there's a lot of AI washing out there as well for sure, where people are like, I use AI for this and that. It's like it's machine learning. It's stuff we've been doing for 15 years.
Chana Schoenberger (07:05):
We had a say yesterday that the easiest way to get a company funded right now is put AI as the end of the.com
Paul Margarites (07:16):
Eight years ago was blockchain, right?
(07:19):
It's the question of it's a great solution. These technologies like Gen AI is a great solution, but is it the right solution is the first question. And I think, and probably it sure it's the same in HSBC, but you have some sort of center of excellence asking those questions. Someone's got a great idea and it's the expertise to say, okay, do we solve this with open banking? Do we solve this with better analytics? But if it's Gen AI, let's unpack that. And then the infrastructure, it's not just the technology infrastructure, it's the people infrastructure to make sure that there's approval, there's an assessment, there's a business case, there's actually people who are going to use it. And then you do the use case and it's great and it's exciting, but you can't leave it there. You got to actually use it.
(08:01):
You got to get people using it out on a daily basis.
Melissa Tuozzolo (08:03):
So true.
Chana Schoenberger (08:04):
Okay. We touched on this slightly, being a people organization, how do you deal with the human capital issues that are involved in Gen AI? I know that one thing that many people have said is that it really makes it difficult to train a younger generation because all of the rote scut work tasks that we all did in our first three years working are now done by computers, but people still have to learn how to do them. You can't wake up one morning and be a vice president. You have to have done analyst and associate work first to understand how do you deal with that?
Melissa Tuozzolo (08:40):
Yeah, so there's two elements of it. So I think one is how do you still create a pathway for people to get trained within your organization and make them almost be able to do that oversight of the AI, which is really what the role is in some of these instances. And then also have people not afraid of it. Because I'll tell you, so from my organization, I've got a lot of junior people around the world that work in our service team, especially in offshore locations. And when I started talking about this, I mean, the first questions I get is the obvious one, am I going to lose my job? Why are you coming? I'm not going to use this thing. I don't want to use my job. I don't want to train it. I don't want to touch it.
(09:21):
But we've really changed that mindset by, to your point, it's not just about creating, it's commercializing it, getting it into the hands of our people and showing them it's not scary. This is to make your job more interesting. This is taking away the basic things that you need to do. Drafting that email. If you're a junior person sitting in India responding to a large multinational corporate in the US and you're worried about your standard of English, you could just have it draft your email as a baseline, then you can add what you want to. And to your point, it's that very standard ChatGPT speak, but that's kind of what you want, right?
Chana Schoenberger (10:02):
Yeah.
Melissa Tuozzolo (10:03):
So it's perfect. So there's that element of it. Now, to your point about how do you train people, what we've found is that the new generation that's coming in is so incredibly tech savvy to begin with. I mean, they grew up with the internet, they grew up with smartphones, they grew up with all of this stuff. They a digital first generation. So they're not afraid of the new technology in the way that perhaps some other people might be afraid to touch and feel these things. I mean, I have people on my team now that not only are they kind of embracing this and like, oh my God, this is so much better than what we had previously. They're actually learning how to write the prompts themselves, which is really cool. They're learning how to use the new technology, and they're starting to build things. Now, again, that creates a whole another host of risk and control environment and making sure people don't go too far off the pathway. But I think when people really start to touch and feel this, it makes it super real for them. They feel like they're part of it, and that's really, really important. And then on the training side, it's thinking about people were worried about giving calculators to kids because they wouldn't learn how to do math.
(11:22):
And we all know there was that time that we had to write things out long form and all of that. But I kind of equate it to that there was a time before Excel, there was a time before all of these other things, and then look at the leaps in productivity that it's allowed. I think if this is all harnessed in the right way, it's going to have the same outcome.
Paul Margarites (11:38):
Look, the skillset isn't for an analyst and associate isn't being able to create a really great PowerPoint deck and we've done the work. I can create a really great PowerPoint deck. It's telling a story, and they're still going to learn those capabilities. How do I tell an effective story to make a decision? How do I advise on a recommendation? It's not how pretty the graphics you were made in the PowerPoint and look, that's necessary because that's our medium right now, but we're going to be working in a new medium. And it's more about what is the right prompt? Just like when Google and Yahoo came out, you could actually be good at Googling things. So it's creating the right setup and then it's the human in the loop, and everything we're doing is human in the loop right now. So it's okay, what did it spit out?
(12:23):
Does it make sense? What do I need to adjust? Those are skills. Those are the same skills you develop as an analyst associate today. You'll keep developing them in the future. You're just going to develop 'em a little bit differently, just like we did in the past as well and generation before us did as well. And I do want to touch on the comment around job security. It does come up, I think in our industry, there's something that's been happening for probably the last decade or so, particularly as the technology industry has changed, the war on talent, excuse me. But there is a war on talent that goes on. It is kind of gone beyond just banking and into technology. I've seen roles where talent has moved from a large bank to a large technology firm.
(13:09):
20 years ago that didn't really happen unless it was a vendor of the institution. It happens a lot more now. Amazon hires a lot from banks, all the big tech industry does. So we need talent. So being able to develop solutions that empower our talent, make them more efficient, that's a good thing. It's not going to lead to less people working at the bank. It's going to lead to the people at the bank being able to do more with what they have.
Melissa Tuozzolo (13:39):
And I think just, sorry, just to add to that, one of the coolest things that we've been able to do with this is run hackathons for people. So when I first heard like, oh, you could run a hackathon for your team with this, I was like, my team's not technical. They don't know how to code. They can't do this. They're like, no, you don't need to code. All you have to do is write in. You write in prose about what you want it to do. So going back to the whole agentic AI thing, which is the new space we're going into, if you create an AI agent that can do a specific task, again, you can write that all in prose. So one of the things we're looking at is I want an agent to oversee all of the email traffic that goes to my service team and identify whether or not it's a complaint. Because today, right, there's regulation on what is not a complaint, but we really leave it up to our people to say, is this a complaint? Is this not a complaint? And let's be honest, some people are going to be like, I don't want to do extra work. This is on the verge. It's not a complaint. It's fine. So something like that, again, you can create that agent to do that all writing prose, which is just a completely new thing for people.
Paul Margarites (14:43):
We did the same thing, and that idea and that concept came from the interns. So I am not worried about the talent for the next generation.
Chana Schoenberger (14:52):
Yeah, that's great. Yeah. Wow. Next they're going to be making a TikTok about your service complaints.
Melissa Tuozzolo (14:59):
Well, that's a whole other set of concerns.
Paul Margarites (15:01):
I'm not right for that one.
Chana Schoenberger (15:02):
Yeah, yeah. No, I love it when especially every summer, all of the banks have those intern classes and they end up doing a recruiting TikTok or something, which gets roundly roasted. Nothing funnier than middle aged managers in a TikTok, but that's what's adorable about it.
Paul Margarites (15:22):
And cringey,
Chana Schoenberger (15:23):
Yeah, it's super cringey, but you got to give them something to laugh at because if you don't know what they're laughing at, they're laughing at you.
Paul Margarites (15:30):
Look, I did the internship program. I remember back then too, we used to do videos of end of year videos with the whole organization and whole team of sketches and comedy. The hookyness has always been there. Now it's just again, taking on a new medium.
Chana Schoenberger (15:46):
Yes. To switch gears, this is a very short panel and I want to get some analyst questions. Let's talk about real-time payments. Okay. So one of the things you brought up when we spoke about this was headless APIs, globalization of payments. Where in the world is this a particular growth trend? What's going on here?
Paul Margarites (16:05):
So I think this is a customer experience story, and I'm really excited about it. So us as a bank, we've entered kind of a period of time where all of the solutions and everything we work with, we work with effectively in this giant ecosystem has moved into an area where everyone has kind of enabled their own capabilities with their APIs. So it's almost like we can see the guts of every institution we work with laid out there and saying, how do we work with them? Here are the microservices to do it. And what we're able to start doing and the industry is effectively doing, is starting to stitch those together, stitch the ecosystem together through the bank to being able to provide a series of capabilities to our customers in the way that they want to take it. So that's a lot of buzzwords. What did I mean by that?
(16:52):
What I mean in there is that we're able to take parts of solutions that exist in the market and put them into a single customer experience. And real-time payments is a great example because it's kind of a born API first product. It's an API first product. You send your messages via API, you receive information via API right back. And what we're able to do is connect that to other things like invoicing solutions and other pieces to provide kind of an end-to-end solution for our customers around payments and information associated with those payments and the experience that they want to get.
Melissa Tuozzolo (17:28):
Yeah, it's pretty incredible to see what our clients have done. So going back to the ingenuity of people off the back of real-time payments, APIs, these headless APIs around the world, especially in markets where real-time payments are now the norm. So you go to India and they do 140 billion realtime payments in a year. Wow. I mean, it's like a UPI, absolutely massive. You almost can't use credit cards in certain places because they're just so used to people using their phones and using real, I mean, China's the same way. It is completely reshaped the entirety of their payments ecosystem in the markets. And with that now what we're seeing, the combination of the infrastructure, the API capabilities is our clients are starting to do some really cool stuff. So in Asia, we have a client that does food delivery, but instead of using, again, credit cards for food delivery, they do real-time payments. They pay the restaurant, they pay the driver, they pay in real time, they settle the whole thing. And it's amazing. Now, as a bank, it's also quite stressful because imagine that your payment system goes down.
(18:41):
While somebody is trying to get their order of ramen. It's a big problem. It happens at That's a crisis. Yeah, it happens at two in the morning. It's a big issue. So you have to really think about your resiliency, your infrastructure. You cannot go down. We also have a client in India who's put this all together and they actually allow their employees to draw down their payroll whenever they want. They normally pay them once every two weeks or once every month, but they have the ability to actually, as they're working, they accrue their funds and they have it in a little wallet for them, and they can say, you know what? I want, whatever. I've accrued a couple of days early this year, so I want to get paid right now. And they press a button, they get paid instantly at two in the morning, at five in the morning whenever they want to get paid. Originally, when they came to me with this idea, I was like, this is a terrible idea. You're going to have all these people pushing to get paid at midnight as they're out at the bars. Haven't really seen that. People are more responsible than I was giving them credit for. But again, this gets woven into the infrastructure of how people are managing their finances.
Paul Margarites (19:49):
I mean, to that point, it's challenging why things are the way they are in certain cases. Why do we get paid every two weeks? We get paid every two weeks because payroll processes are really painful, really complex, and you need to make sure it works appropriately. Realtime payments are available. If a person earns their wage, they have every right to that wage for them. That might be the difference between a late fee and a credit card, which is far more expensive. Exactly. I think it's a great point, and it's just one use case of many though, with a real-time payment. For instance. And Melissa, to your point, half the innovation comes from our clients. It's amazing to see what our customers are doing, particularly in the business space. So there's so much innovation going out there. We dedicate so much time and resources to innovation. But again, a lot of it comes from our customers and their behaviors.
Melissa Tuozzolo (20:38):
I love when a client comes to us and says, I have an idea. Let's pull out the whiteboard. Let's get the, let's see what we can do. And that's where these things come from. It's amazing.
Chana Schoenberger (20:49):
So do you guys have good examples of things the clients have brought to you?
Melissa Tuozzolo (20:52):
Well, this payout, whenever is a big one. We've had clients that have come up with real-time payment schemes that aren't really official cross-border real-time payments that we've helped them to facilitate. We facilitated the fiat for offline crypto exchange type thing, or not crypto, but blockchain type capabilities. I mean, all sorts of stuff where you first hear it and you're like, how did you think? Who came up with this in your treasury team? But then you sit down and you're like, that's actually brilliant.
Paul Margarites (21:26):
It happened years ago. Again, a lot of the larger customers out there on the business side, they started building transformation teams within their treasury group.
(21:35):
And sometimes it's one individual, but sometimes it's a team, and it's those teams that typically are saying, how can we change something? And you can't change something really without the bank involved. So then they pull the bank in, they work through an idea. We've done the same in past instances like blockchain. This is something they want to look at. Let's look at it. We've had customers come to us with crypto conversations as well. You know what? I'll be honest though. A lot of the times the conversation goes towards, there's something that's existing already. We just need to use it a little differently.
Chana Schoenberger (22:07):
So refinements,
Paul Margarites (22:08):
Almost.
Chana Schoenberger (22:09):
Okay. I want to make sure we don't run out of time for audience questions. So let me see if anyone has a question and if not, I have a lot more.
(22:15):
Yes.
Audience Member 1 (22:18):
Hi. I'm a with KeyBank, so I'm currently an associate. So from that previous discussion, a lot of AI is very useful for deck building and client pitches, et cetera. We're not using it at work, but I do use ChatGPT for my personal stuff. So I am comfortable with using it. I just haven't applied it to my job just yet. But with the topic and the trends and stuff like that, I know it is coming. So I would like to ask you, what is your advice for someone like myself, associate or analyst? How should I embrace AI to help out my team and add value?
Melissa Tuozzolo (23:08):
Yeah, I mean, I think the fact that you're using it in your personal life, you're going to be able to run circles around people when you are able to start using it in your institution. I use it my personal life too. So even before we had access to all of these great tools, I use ChatGPT to plan my future sister-in-law's bridal shower to write my speech for it. I use it all. I use it to plan my summer vacation with my family. I absolutely love it. But I think since you've had experience with it, you understand it's all down to prompt engineering. How do you ask it? The question, when you ask it the question, how do you check to make sure that what it gave you is actually the right information? Sometimes it does come up with bogus stuff. Those are skills, and those are going to be really important skills as this type of capability is built into your organization and you as an associate relatively new in your field, you're actually going to come with a really fresh perspective, but a really powerful toolkit that is going to make you incredibly valuable.
(24:14):
So I would say continue to play with it in your personal life, but think about if I had this capability at work, how would I use it? And then lobby as much as you can internally to get it available.
Audience Member 1 (24:25):
Thank you. Yeah.
Audience Member 2 (24:28):
Awesome. Hi, this is an interesting panel. I too work at KeyBank. I'm wondering both at TD and HSBC, how many employees actually have access to ChatGPT or how have you started to roll in and given some of the risk and compliance issues?
Paul Margarites (24:51):
We're rolling it out right now across, we're rolling out our co-pilot piece across mostly today. It's our actually technology team that has access to everything. So they use it, our developers use functionality within the GitHub, and then those in the contact center have started getting access to it as well. That has been our approach.
Melissa Tuozzolo (25:12):
So same. So we've got our tech team on, I actually dunno how many employees we have on our tech team that are live in my team right now. We're piloting with 160 users in the UK and in Singapore. By this summer we'll have 1300 users live. So we're rolling out pretty quickly now. That's on copilot. We also have our own internal GPT-4 AI tool that we call Lydia, which anybody can get access to in the organization. You just have to fill out a form, take a training. So I don't even know how many people have access to it, but I love it because again, it's kind of like a ChatGPT, but for our internal team. So I could use it, say I want to send a holiday note to my team, right? Like, oh, I want to send out a note for the holidays. What should I write? And it will draft it for me. And it's phenomenal. So it takes away like 70% of that work of drafting that note. So yeah, so our people are getting very used to these new tools.
Paul Margarites (26:09):
Our Lydia, is that what you call it?
Melissa Tuozzolo (26:11):
We named it Lydia. I don't know where came from.
Paul Margarites (26:14):
Lydia. Lydia. It is coming really soon. I'm actually literally chomping at the bit for it. I emailed the Gen AI team literally every week. Are we there yet? Are we there? What are you going to call it? I don't know. I'm on the spot now. I don't know. I can't think of a good name.
Melissa Tuozzolo (26:27):
We have another one called Stit, and that's if anybody's ever seen the HSPC offices. We have two lions and it's Stit and Stevens. We named one Stit. So maybe you can come up with, I don't know if you guys have any mascots, but it's a good way.
Paul Margarites (26:43):
I think we do. I have to check.
Chana Schoenberger (26:45):
Yeah, BNY has one named Eliza because Eliza Hamilton was their first client. They're very proud of that. Oh, that's great. Yes. Awesome. Okay, we can have one more question. I see a question back there.
Audience Member Alph (27:01):
Hello? Yeah, my name is Alph and I'm a regulator, so stick in my mud. But just a quick question one, this is an interesting panel, but working with ChatGPT and any AI, Generative AI, how for an institution like you guys, how do you implement it without losing the specialness of the uniqueness? Because ChatGPT can be very generic, and yes, it's important to be able to, it's easy to get the message up and running, but there's a level of uniqueness in the interactions that we have as human beings that makes things special. So how do you not lose that while also reaping the benefits of those efficiencies?
Paul Margarites (27:44):
So first of all, you're not a stick in the mud. You're a part of our industry that adds security trust with our customers. So I just want to call that out to the other piece. I think both our institutions, we have human in the loop. We're not issuing anything out directly to clients without that human in the loop. And I'll tell you, our service and contact center teams are so incredible. There's a certain kind of culture that exists within TD Bank that is so customer centric, and I know we're all customer centric, but there's a sincerity to these teams and they're the interface between the customer and the bank. So they're using the tools today to actually ask our version of ChatGPT questions about products and services and issues our customers might be facing. They're translating it into their conversation with the customer. So they're not losing the flavor of this is who I am and unique and authentic and how I engage with customers. It's just like a quick lookup. It's a quick tool that makes them faster at getting an answer.
Melissa Tuozzolo (28:47):
And just echoing that, I mean for us, and George has talked about this, our new CEO, how he sees client experience and client engagement as one of our superpowers as an organization and the way we see ourselves. So I mentioned earlier, we're already seeing with the users that are on these tools right now, about 10% uplift in efficiency, and we're not seeing that as a 10% uplift. Efficiency means that a 10% reduction in FTE, we're calling it capacity creation. Because really what we want to do is use that 10%, hopefully 20, 25, 30% in the future. We think there's so much more we can do with this to actually have people spend more time again, doing the thinking. So most of our payment queries are, where's my payment trace payment? Where's my payment, et cetera. We don't need a super personal email back on, where's my payment?
(29:40):
All client wants to know is where's my payment? Something like that. Very efficient to push through, very efficient to even in the future, think of ways, can we take the human out of the loop of that one? Again, something very, very specific, but things like, I have a really complicated, I've just bought another company. I've done an m and a transaction. I'm merging the companies together. Help me figure out what to do with my cash pooling structure. That's what I want my people to be spending their time on with the clients. Or I've had a fraud case, help me retrieve the funds. Right? That's what I want them to spend the time on. Really those moments that matter. Not the bureaucracy, not the mindless work.
Paul Margarites (30:21):
Yeah,
Chana Schoenberger (30:23):
Great point. Okay, we are completely out of time here. So thank you very much to my distinguished panelists and we're going to bring up the next panel now.