AI, Data and the Future of Banking

Speaker A: Welcome to the Power of Data, the podcast by Dun and Bradstreet. Data is everywhere and there is more created every second of every day. Join us to hear from leaders unlocking the value of data.

Speaker B: Welcome to the Power of Data podcast. I'm Nick White, head of sales for the Worldwide Network. Today we're joined by Michael Conway, UK banking and fintech industry leader at IBM. Michael, you've held several leadership roles at IBM driving transformation, growth and AI innovation. And now you're leading the banking and fintech sector. Huh. So before we dive into our conversation today, would you mind sharing a little bit about your journey and what are some of the most exciting things and the opportunities that lie ahead for you in your role?

Speaker C: Sure. Firstly, thank you so much for having me. It's really, really brilliant to be here in your beautiful offices. Um, so what's excited me over the, over my career so far, I think being uh, ah, an entrepreneur in an organization the size of IBM is always fun because frankly you can have startup ideas but you know you're going to get paid at the end of the month. Um, and I think it's that sort of culture which has really spurred me on in my career, trying to build business on behalf of IBM, um, and build teams around that to make sure that we're, you know, we've got an army of folks behind us to deliver the change that's going to transform our clients, uh, businesses of course that's always powered by technology. Um, and I tend to try and be at the forefront of technology. I often talk to our clients, also to our teams that ah, if it's on the frontier we're interested, if it's, you know, the running of that technology, we should then hand back. We never want to bake ourselves into the problem, um, or indeed the solution. Um, we always want to be on the cutting edge with our clients but really it's the people, right. Whatever uh, technology we use. I started in data, moved into AI about a decade ago. It's the people that I've surrounded myself with and that IBM, uh, has nurtured in that time that make the job as fun as it is, both clients and my team members.

Speaker B: Fantastic. Michael, uh, you mentioned you've been in AI for nearly a decade. What was AI like 10 years ago?

Speaker C: Uh, the Wild West. I think it's probably got wilder if anything. Um, but I think what's really interesting is a lot of the world woke up to this two years ago, about two and a half years ago actually. But really we've been plowing that field for a long, long time. If you think back, even just in IBM, you think back to, you know, when the first use of the term artificial intelligence was.

Speaker B: No, tell me.

Speaker C: M. 1956. Um, so it's an old field. An old field at least the idea of it is. Um, in that time, I think two major things have happened. Compute, the amount of compute and grunt that computers can, uh, muster and better mathematical approaches. Right. And what we've witnessed over the period is those two things coming together and then major leaps forward. Uh, IBM Deep Blue, which is probably one of the more famous ones in 97 with Garry Kasparov playing the uh, Grandmaster at chess at his own game and beating him M Famously. So Jeopardy. In 2010, I think that was where that was the real start of linguistic AI. So if you're familiar with Jeopardy, it's about really difficult language games around. I'm going to give you the answer. You need to work out the question really obtuse facts as well. Um, but then in 2017, what Google did with the transfer former with their paper, all you need is attention. That's what brought us the T of GPT. Um, so that architecturally and mathematically those leaps and bounds together with the compute over that period have caused the explosion that we have today. I think 10 years ago we were still doing massive amounts of valuable work, right? Centered mostly around the contact center, things like virtual agents. When delivered, right, they were phenomenally, uh, powerful for both customers and um, uh, the organization that they served. But unfortunately they got a bit of a reputation for being terrible because there was too many terrible ones. So the good ones really suffered.

Speaker B: Great. Thank you, Michael. Um, we're getting into the meat of the conversation now. Um, we're in an era of transformation where business, technology and data are more interconnected than ever. How do you see this shape in the banking and fintech industry and what specific innovations or trends stand out for you right now?

Speaker C: So I think some of this is absent of industry. It's just, uh, in the wider economy, things are going to be changing quite a bit around AI, um, especially with what's going on at the moment as well. The need for more homegrown manufacture, all that sort of stuff is an interesting challenge to start applying AI to as well. I think in terms of banking, I think we've done quite a bit of, uh, research recently and something like 60% of banking CEOs appreciate they're going to have to take on a level of additional risk to exploit these AI, um, revolution. So I think organizations Risk appetite is going to have to change a little bit. I'm not saying just throw everything to the wall and hope for the best. You have to do this in a responsible and confident way. But you do need to tweak those boundaries a little bit. Um, we worked with one bank where if they followed the risk procedure as written, the uh, AI that we'd written with them, um, the amount of checks that that AI would need for one work, one week's worth of work was 19 man years. So you've reduced a week's worth of work and you've introduced 19 years of effort, uh, to check that you've done the work. Right. So obviously that's not going to fly. Right. So risk appetite has to change, it has to modernize or move with technology. Um, but equally I think customers expectations is climbing. You think about what happened with the fintechs, the neobanks in the last six or seven years, maybe a bit longer. They brought with them an expectation of how a user interface should look for a banking app. Right. And then you've got people that are born into that world that are only ever going to use the monzos, the starlings, the revoluts, and then the more heritage banks have to then keep up with them and then hopefully overtake them because they've got bigger pockets, right. And they've got a much bigger customer base. But what they did for the user experience I think AI is going to use for, is going to do for what people are going to come to expect around intelligence. So how can you advise me in a way that's going to help me meet my financial goals? Right, and how is that going to be really, really low friction, really super, super personalized to me, but free or at least low cost. Um, so I think this wealth advice, money management, financial health is going to be an area of real uh, interest for both banks to explore, but also customers will come to expect.

Speaker B: Great, that sounds really fascinating. Got a question later on where I'm going to go a little bit deeper into some of those exciting things you're working on. Um, many companies want to harness AI but are struggling with implementation. What are the key factors that separate those leaders from the ones that are being left behind in your opinion?

Speaker C: Ah, so again in our studies what remarkable numbers stood out to me that only 8% have got a strategic um, goal at ah, a board level around how to implement AI.

Speaker B: 8%, 8%, wow.

Speaker C: So 92%, even though I could do that maths in my head haven't, um, so effectively what they're letting happen is what I call a thousand flowers of bloom. They've just throw out the season, hope for the best, right. And federate it out, see what happens. A bunch of experimentation and nothing happens, um, of any meaning or materiality. I think what the. There's a very strong correlation and I suspect causation between organizations that have central control and central strategy that says we want to achieve these things and we are going to figure it out in these chunks. They experiment with that, then they are able to execute on it at scale and then they federate versus the ones that federate. Hopefully throw the seeds out, hope for the best and then expect to pull the flowers out the field. It doesn't work. Um, so I think the centralization is a key thing. I was thinking about this for my team actually last week and I've come up with a bit of a two by two because a word that we don't often use in business but I think we should is wonder. Right. This stuff, when it's delivered brilliantly, creates a space where wow, like it's a proper wow moment. You don't often get that sitting in an office.

Speaker B: No.

Speaker C: Um, but I think there is a. The two by two is you can have like I've got wonder and I've got grit and I'll unpack what grit means, um, on uh, the horizontal. Because a POC that creates magic or wonder that doesn't scale is full of wonder, but it doesn't transform anything. Right. There's a business value chasm that it hasn't crossed. Well, how do you then cross the business value chasm into something which is full of wonder but then moves the needle on a business. And grit is a bit of an acronym. It's a lot of grit and hard work. Um, but the acronym is governance. Back to the risk point. You need to do this in a governed, responsible way, but you need to readjust your risk appetite and also those checks and balances which might be out of kilter for the new type of technology. R is rigorous. You need to be very systematic and understand how you're applying the technology, how you're applying people to that technology and what outcomes you're looking for. Intelligence, both artificial and human. Good artificial intelligence is sort of table stakes, but good human intelligence is an underbate skill. Knowing how to apply this stuff, having the experiences of previous implementations, all that sort of stuff is deeply important and a bit like grit. There's tenacity, right. So you need to not always take no for an answer and push forward when there's massive amounts of resistance. So what I want and what my teams work on is creating this sense of wonder, but executed in the real world, moving the real business KPIs because just creating the wonder. We've seen so much of that. I see it so much on LinkedIn. Uh oh, we did this wonderful POC. Fantastic. Now scale it and they can't. So crossing that business value chasm is where the haves are going to be separated from the have nots.

Speaker B: I'm looking forward to using the GRIT acronym in my next meeting.

Speaker C: My team will be delighted.

Speaker B: Yeah, absolutely. I hope you've got that patent pended. Um, so the next question is changing the dynamic a little bit and it's about the workforce and upskilling being such a key factor in industry being able to um, take full advantage of AI in the future. Um, what strategies have been most effective in equipping employees to have the right skills and how are IBM playing a role in that in transformation?

Speaker C: So skills is massively important as we've spoken about. I think to fully execute on your AI strategy, you need a human strategy as well. How is your team going to A, not be resistant but B, really thrive and flourish in this new world? There's a bunch of stuff that I can unpack here, but I think and some of it's generational and I don't mean that to be sound too grandiose, but personal story. I've got a five, a five and two year old and we were looking for schools for my five year old, she was then four, um, and we went to one particular school and their IT lesson was on physical notebooks, pages and pencils.

Speaker B: Wow.

Speaker C: So she's not going here. Like we owe it to the next generation to understand the world in which they're going to come into. Because the world that I'm not professing to know what the workforce is going to look like when my 5 and 2 year old hit the workforce in 20 or so years time. But I know it's not going to look like today and it's certainly not going to look like where we're sitting around with pieces of paper and pencils. Right. So there is a generational thing where schools need to get better at actually how do we use AI? A bit like, I guess what my parents learned with the calculator. There was a resistance to the calculator. Well, guess what? If you put tools and use them in the right way, the problems that you can solve are much bigger. I think technology firms like Ourself has a responsibility to lower the barrier to entry to using this type of technology. It shouldn't just be the very best data scientist that can get the value out of AI. And actually I think with this latest uh, wave of the last couple of years, I think that has gone away. Right when my dad is using ChatGPT to book his next trip to Cyprus or the itinerary to Cyprus, I think that's when it's really broken through. Right. I think uh, the barrier to entry is nearly non existent. You have got to have a digital device and you've got to have access to Internet. If you have those two things and you have an idea, then the barrier to entry is disintegrating, which is great in the business sense. Um, I think education of the team is massively important. So is hands on experimentation. At IBM we ran a massive uh, hackathon or jam across our whole global organization last summer. And that is everyone from the deepest technical person to the secretaries to the whoevers. Um, because this is a game for everyone to play. Um, and it can't just be the IT function that benefits from it. Um, and we frankly, we run loads of programs for senior executives, senior leadership of our clients to then bring them on the journey Again. Back to my causation correlation point around centralizing. It's centralizing as well as having senior uh, sponsorship of those central functions or central ideas. They're the organizations that really work. So actually you need to start with the board level, then the rest of the C suite and then the minus ones and then you go down the organization. This needs to be very senior, senior sponsored, um, and then uh, it needs to show how everyone, how their job's going to get better.

Speaker B: Okay, great. I'd be really interested. Again it's not here but um, when you think about the skill sets for somebody to have in the future to still have value uh, in the workforce, what are those human skills that you think are going to be most valuable to the workforce in the future?

Speaker C: There is a bit of a dichotomy, um, and this is my opinion, um, but there is a bit of a dichotomy coming and I can see it happening where on the face of it, it looks like we're able to automate pretty junior jobs. You think about developers, my wife's a lawyer, you think about the grunt of a law, uh, a junior lawyer, of doing the effective paperwork for filing and photocopying, etc. Generationally that's how they've learned how to be a good lawyer or a good developer. That is largely an automatable problem. Now, however, what's not an automatable problem is expertise, because you need to know how the AI is going right and how the AI is going wrong. And you only get that with expertise. But how do you get that expertise if you've automated the track to get there? Right? So there is a real dichotomy. And I'm not saying I have an, uh, answer for this, but I think the forming opinion or hypothesis that I'm forming is we need to teach those, those junior jobs at averticommas, we need to teach them how to work with AI. So it's not just how do you develop, it's how do you develop with AI and what are the things that you need to look out for so that you are the expert guiding the AI? Because I think that's a big problem that we're going to face in not many years if we automate the bottom rung of the ladder. Uh, you can't jump, jump, jump to the top, right? Um, yeah, I get you.

Speaker B: I get you. It's those foundational skills you learn, uh, in any role you do within a company that kind of shapes. They're the experiences that shape your more experience and seniority in the future.

Speaker C: And then you mix that with COVID and what Covid did to our work environment, the learning by osmosis of being physically present with each other. If you take that away, or at least dilute that, ah, plus AI. Ah, there's lots of things we need to think about. Future workforce strategy.

Speaker B: Uh, yeah, for another day, I think. But, uh, yeah, it is a worry. It is a worry for sure. Um, again, I've gone off piece a little bit, but, uh, a friend of mine is a teacher and he says he runs a program. Um, he uses AI to mark homework and check to see if homework has been generated using a Gen I, A program. So he's marking homework generated by another computer.

Speaker C: And in the business world, we're seeing that with things like complaints. So complaints are being, uh, generated en masse by AI because it's easier to get them to do and to try and get some money back from an organization by complaining en masse. Because it's such little effort to write a prompt, say, write 15 different variations of this.

Speaker B: Yeah.

Speaker C: Um, so, yeah, we're seeing a lot of that in the enterprise world as well.

Speaker B: Changing topic, uh, a bit. But moving on to something that I'm sure the, uh, viewers and listeners will hear about is some of the more Exciting AI driven projects that IBM are working on without sharing too much of the trade secrets. What are some of the things that you're most excited about in the future?

Speaker C: So I think what's really wonderful about AI and I don't think it can be said of many types of technologies is you can radically reduce the cost to serve, but you can also massively improve the customer experience or colleague experience. Right. Typically it's one or other you either make something way better and it costs way more or you make it cheaper and it becomes worse. What we found is the magic between those two things. Make it cheaper but better. Right. And that is a remarkable outcome when you think about you abstract yourself to any business outcome. So some of the cooler ones, you know, in my world, virgin Money, um, with uh, their virtual agent ready, um, everything has to be read with them. Um, but it's the most popular channel in their organization, including humans in the brand, which is somewhat amazing, right?

Speaker B: Yeah, that is amazing.

Speaker C: Um, and you know you can pretty much do all your banking needs and journeys through that, that channel. Um, with massively high net promoter score, working with Nationwide, we've created their center of excellence or a center of expertise around how to do this at a really scaled and governed um, organizational um, way back to my point around centralization and things like things that were unimaginable a few years ago where when you're testing outcomes for customer calls, typically banks have to test outcomes that things are getting done in the way that the customer wanted or the customer needed. From a regulatory perspective also just from a good sort of housekeeping perspective, making sure customers getting what they want and understanding what agents need, extra training, all that sort of stuff. So typically, you know, banks would dip test, right? Because that's the whole point you stick uh, a dip in of X percent and it's pretty small percent to make sure that agents and customers, uh, agents are doing what they should and customers are getting what they should. Well imagine a world where we don't have to dip test. You can do 100% sampling. Then again in massive aggregate, you understand that the control plane or what you can see in your organization is 100% of interactions, not this really tiny amount that I'm checking. And then you can see, okay, well this agent is consistently struggling on this type of journey or this customer really needs some additional help doing xyz. Well actually if you can see the world in aggregate, I think that's massively powerful. So that's the sorts of things we're doing with Nationwide as well. Um, we're doing some phenomenal work uh, with complaints management across a few clients. Uh, complaints is a topic, uh, close to my heart as you can probably tell. But actually it's a really brilliant problem for AIs. To understand how to index, how to understand what a customer is complaining about is a really complex thing. And actually frankly from my observations across many organizations, humans are really bad at it. They're really bad at understanding the core component of what a complaint is and what they need uh, to be able to fix that. And we've got like uh, over a thousand percent efficiency gains in some of these processes um, at scale. Again not in poc. If it's a poc. I'm not interested at scale, uh, working across organizations whole customer base. So we're really proud of those sorts of things. They're full of wonderful, but they're full of grit.

Speaker B: Full of grit at the same time. And, and is it just enhancements in technology that's making these things available today that they wouldn't have been able to be done three years ago?

Speaker C: Yeah, I think it's funny because if I reflect on some of the models that we're using are four or five years old and actually this was one of the massive um, head starts my team and I had. We were using LLMs in 2019, 2020, but there was just a technique, it was a data science technique for geeks. Right. And ChatGPT happened November 22nd and it just blew up as a market. And actually some of the stuff that we're doing now because if you think about what we're doing, we're doing lots of classification type work, we're doing lots of generation, but actually the classification work uses models which are quite old and quite established, all open source. Um, so I guess in theory we could have done this years ago, but I think the big difference happened was appetite and awareness. Because when a board are aware of this stuff and have an appetite for it, it then moves the needle very much quicker. You also then pile in on, well actually the stock market values AI as well. So every board wants to say we're doing AI. All of these things, um, combine into a very ripe environment for transformation.

Speaker B: So if we fast forward into five years from now, so 2030, um, what do you think the world of AI will look like in the context of banking and fintechs?

Speaker C: So I can't tell you what it's going to look like next month. I'm an avid and voracious reader, um, both in industry and outside and I can't keep up with it. Right. So to make any predictions I have no idea. I think all I can talk about is I guess previous trends have a decent um, have a decent ability to tell us where we're going. I think open source is going to become more and more prevalent. I think open source labs have basically caught up with proprietary. Proprietary then jumps but then open source catches. So I think that the march to open source is going to become more and more aggressive. I think models are going to become smaller and more specialized so we won't have the. Well we will have. I often say we'll live in an and world rather than or that's all about optionality but you have the big frontier proprietary models doing really exciting frontier things but then I think you'll have really small specialist modules or models which chain together for a workflow because they'll become more manageable, more expert, easily governed and I want to do my end of month accounts in maths. I don't want to do it in the spirit of Shakespeare. It doesn't like. I don't need that. Um, one thing, one cooler thing that I've started to think about is if you think about probably early 2000s with the rise of BPO business process outsourcing and ito with it basically the move to shift to the east we moved a lot of work to India typically but also the Philippines et cetera um because of a, the huge amount of people but be the skill that they were building in those people um and you know the elastic ability to just onboard lots of people to do a certain business or IT process. I think where we'll end up and this is I hate predicting because I hate being wrong. Um, but where we could that was sort of resource arbitrage, that's what it was called. We could get into where we're moving towards intelligence arbitrage. So I've got this human intelligence over here that might be in India, it might be in London but I'll um, resource arbitrage between that but only once I've made the decision that it needs to be human intelligence actually I might have intelligence that I'm arbitraging with the machine. So I've got AI over here and this AI which is domiciled in this country with these regulations is a far better place to run that type of process where I've got a different type of process where actually it's closer to the compute which might be on the west coast of the States and M I can't afford the latency so I have to put it over there. So you're going to arbitrate between AIs and humans. And I can see this sort of world of chess forming where you'll move your IT and business processes to where it's best consumed, whether that's by a human doing the work or by an AI.

Speaker B: AI is only as powerful as the data it learns from. How critical is high quality, well managed data in building an effective AI solution? Um, what challenges do organizations face in harnessing their data for AI and how can they overcome that?

Speaker C: Nice small question, three parter. So I often talk about data, uh, in two parts and it's around IP in particular. Um, and these are my terms, they're not industry terms, I'd love for them to become industry terms. But I talk about inbound and outbound ip. So a problem that was pretty well understood pretty quickly but still was a problem was what I call outbound ip, I E I as an organization give a model, my data that then goes into the model as training data and then my competition down the road can then access that. Right. So that's outbound ip. That is somewhat of a solved problem because uh, your models will typically be behind a firewall that you understand and you can turn off the training side of things and all that sort of stuff. So in an enterprise world that's largely a solved problem. The less solved problem is inbound ip. If I as an organization consume model from model provider X, I won't actually know, let's say Y because X is a thing. Model provider, yes. And model provider Y is not uh, transparent where the data from that for the training of their model came from. I'm presented with a risk of if they're litigated against that litigation might flow down to me because I might be in breach of copyright or whatever, um, in the event that they're found in breach of copyright. So that's a less well understood problem. Now it's largely running its course through the Supreme Court. So there's a big test case. New York Times vs OpenAI is a big one in Japan. Interestingly they ruled last year, I think that models, data in models is an IP interesting stance. So that would exonerate or make it's open season for the model providers. But I think in an enterprise one needs to understand the risk whilst this is working its way through the legal system. For IBM's part, we've uh, deliberately gone down a path of very open and transparent data harvesting. So our models, our family of models is called granite and the Granite model series is we publish what data goes into them as training data. And either there's open source with a very permissive license, we own the data, or we own the copyright. So we can tell you with absolute confidence that however these Supreme Court things are going to get sorted, you're using trusted data. And I think that's really, really important. Um, back to my point around and, or, or we'll still live in an and world. There'll be place for all these different types of um, models and approaches to training. One thing I would add though is there is sort of in the round of 99% of the world's data that is the Internet is in a model or another, we reckon Less than 1% of organization enterprise proprietary data is in a foundation model. So actually when you think about, okay, we've sucked up all the world's open data and I say open it inverted commas because it's not. It doesn't mean it's, you know, dun and Bradstreet data might be on the Internet, doesn't mean you don't own it. M. But if we class that as open with a lowercase o. The world's Internet data is in a model. Well, actually, what power could we get if we then put proprietary data that only you have, only an organization has, and only they know about their customers. A bank like Lloyds or RBS or NatWest, they're like hundreds of years old, right? You think about all that data that they have, you then pair that with the world's open Internet data. Amazing outcomes can come. Um, so, yeah, long answer to a long question, but hopefully useful.

Speaker B: Yeah, very, very useful. And I could see the world where that creates the wonder argument, the answer that we look for.

Speaker C: Exactly right.

Speaker B: Thank you, Michael. Um, my final question for you today for any business leaders that are still hesitating about adopting AI or what's your key message to them?

Speaker C: So I always have the same answer and it sounds massively trite and disrespectful, but I always said the best place to start is to start. Um, there is the biggest resistance that I see at senior leadership level is, God, this train's moving so fast. I'm just going to, I can't, I can't get away. I'm going to wait. I'm going to wait. I'm going to wait. It's never going to slow down. If anything is speeding up, just jump on. And I think the biggest trick is to look at use cases that in extremis, even if it's going to be throwaway in 12 months time. Does it pay back? If it pays back it doesn't matter if you throw it away because actually what you're going to learn is hugely important. There has to be an ROI payback to it. I'm not saying just throw money at a problem that is a black hole but be really responsible in how you allocate capital. But start, just start experiment, see what works, see what doesn't. Um because if you wait for this train to slow down it's not slowing down.

Speaker B: Yeah. If you wait for a point where you see this start to plateau off.

Speaker C: Yeah. And frankly you're too late.

Speaker B: Too late. Thank you Michael. Thanks very much for your time. My pleasure, absolute pleasure. Thank you.

Speaker C: Thank you for having me.

Speaker D: Find out more about how Dun and Bradstreet can help your business be better. Contact us@marketinguknb.com and remember to subscribe on Apple podcasts, Spotify and Google podcasts.

Speaker A: It.