# Generative AI Use Cases and Adoption Patterns in Banking

- **Show:** Banking Transformed with Jim Marous
- **Published:** 19 Mar 2024
- **Source:** https://thefinancialbrand.com/banking-podcasts/
- **Transcript:** https://podcasts.opennash.com/shows/banking-transformed/generative-ai-use-cases-and-adoption-patterns-in-banking-7d1b44/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Banks should start using generative AI now, in small practical steps, instead of waiting for a perfect plan. Daragh Morrissey of Microsoft argues that early use cases such as developer tools, contact centers and knowledge bases build the skills, data and customer trust that later, more ambitious uses depend on.

The tension is between speed and caution. The technology makes mistakes, so banks first use it internally with humans overseeing it and safety guardrails. Meanwhile, delay carries its own cost, because employees, customers and rivals are already adopting it. Morrissey warns against over-building, endless use-case lists and hiring data scientists without a clear purpose.

## Key insights

1. **Start with developers, then Office tools, then custom uses** Morrissey recommends developers first because it is the most mature use, with 30% to 40% productivity gains. Out-of-the-box Teams and Office assistants come next, and custom projects such as contact centers follow.
2. **Contact centers can become continuously learning** Generative AI can summarize calls, coach agents live and check compliance. Feeding every conversation into a searchable knowledge base means the bank learns from each interaction, and agents can move to higher-value work.
3. **Banks test safely inside before going customer-facing** The first wave of use cases was internal productivity. Banks are learning to deploy safely internally, and Microsoft's guardrails limit what a chatbot will discuss, as in the Mercedes in-car assistant.
4. **Start small instead of perfecting use-case lists** Morrissey says some banks stall on lists of 50 use cases that they try to cost out, even though business cases rarely deliver what they promise. He advises starting with the smallest model and using real use to learn.
5. **Reskill staff rather than hiring expensive data scientists** Morrissey says one Chinese bank hired 30 data scientists without knowing what to do with them. He argues that people who know the bank's products are better placed, and generative AI is easier to build than traditional predictive models.

## Questions this episode answers

### What is a copilot in the context of generative AI?

Morrissey describes a copilot as an AI assistant that helps you start or do a task while a human oversees it. He says this matters in banking because the technology doesn't always get things right.

### Where should a bank begin with generative AI?

Morrissey suggests starting with software developers, who see 30% to 40% productivity gains. Next come the built-in Office and Teams tools, then custom projects such as contact centers and knowledge bases.

### Is generative AI safe enough for a bank to put in front of customers?

Morrissey says banks first learn to use it safely inside the organization. Guardrails can then limit what a customer-facing chatbot will discuss, though he admits the technology isn't always accurate.

### Can small banks compete with big banks on generative AI?

Morrissey believes it levels the playing field because it is easier to build than traditional predictive AI and doesn't need a large data science team. He says banks that start earlier will build more skill, though.

### How can generative AI help with old banking software?

Morrissey says it can explain what legacy code such as COBOL does, document it and convert it to a modern language. A large Wall Street bank has done this for about two years.

## Facts

- **Daragh Morrissey** [0:02:39]: "We have one big Wall Street bank doing that for about the last two years now and the reason why I like talking about it is that the benefits of it are very well understood now"
- **Daragh Morrissey** [0:03:33]: "A lot of that was written in C++, and we're converting it to a new language that's going to be a bit more secure and will reduce the number of patches that you get on Patch Tuesday."
- **Daragh Morrissey** [0:03:08]: "I studied COBOL in college. That's how old I am. And there's a lot of COBOL code there."
- **Daragh Morrissey** [0:07:40]: "Sometimes it can be every three, every six months, they lose about 30%. It's because it's a tough job."
- **Daragh Morrissey** [0:04:28]: "I have one Canadian bank that just wanted to do that and play it back to the customer"
- **Daragh Morrissey** [0:05:01]: "We have one insurance company where we did first notice a loss where it started to pull out the key points in that conversation."
- **Daragh Morrissey** [0:12:29]: "companies like Morningstar, they've put in kind of co-pilots in front of all of their data. And again, it just makes it easier to consume that market research."
- **Daragh Morrissey** [0:13:02]: "we put the power of chat GPT into the cars and this was all built on Azure OpenAI"
- **Daragh Morrissey** [0:13:46]: "We actually built them with OpenAI directly, and then we built them into our own services, these guardrails."
- **Daragh Morrissey** [0:12:29]: "They sort of answer two or three questions questions, and then they will bounce you out to a contact center or a website."
- **Daragh Morrissey** [0:16:08]: "We work with a company called Cell Machines. they have a great um have this really human like avatar"
- **Daragh Morrissey** [0:24:25]: "I did see one bank in China. I know they went out and they hired 30 data scientists, but they didn't really know what to do."
- **Daragh Morrissey** [0:28:28]: "We sort of see 30 % to 40 % improvements in developer productivity. So that's a really quick win."
- **Daragh Morrissey** [0:31:51]: "I think one use case that McKinsey had recently was actually training wealth advisors, new wealth advisors with simulated conversations."
- **Daragh Morrissey** [0:33:17]: "Jensen, the CEO of NVIDIA, he was on a panel recently and he said, you know, domain expertise is actually going to be more valuable."
- **Daragh Morrissey** [0:36:01]: "we have these things called plugins so we can reach out and say what is the latest stock price of Microsoft and you get back an answer"
- **Daragh Morrissey** [0:36:42]: "At the moment, it's quite expensive to build your own foundational model, a large foundational model. Now, the price is coming down all the time."
- **Daragh Morrissey** [0:37:10]: "I had one bank that built a browser recently"
- **Daragh Morrissey** [0:37:42]: "some banks kind of getting to a list of 50 use cases and then somebody comes in and tries to cost them out and assess them"
- **Jim Marous** [0:38:57]: "I use the analogy of Bank of America's Erica platform, the voice banking platform."
- **Daragh Morrissey** [0:40:59]: "Sam Altman talks about this thing called artificial general intelligence."
- **Daragh Morrissey** [0:42:06]: "we have this other wave of tech coming, which is quantum. And that's going to break encryption."
- **Daragh Morrissey** [0:22:40]: "The first thing they gave us was Teams recording, And then it was a month later, I saw it show up in PowerPoint."
- **Daragh Morrissey** [0:13:02]: "you can ask it, give me some tips for an Irish roast dinner but you can't ask it, you can't ask it, I want to rob a bank,"

## Transcript

**[0:00:11] Jim Marous:** Hello, and welcome to Banking Transformed, the top podcast in retail banking. I'm your host, Jim Marous, owner and CEO of the Digital Banking Report and co-publisher of the financial brand. In today's episode, we're thrilled to have Daragh Morrissey, the Director of AI at Microsoft Worldwide Financial Services, join us today to discuss the exciting world of generative AI and its potential applications in banking. We delve into various use cases and adoption patterns, including internal organizational applications, enhancing customer experiences, and the development of new product offerings made possible by this new technology.

**[0:00:51] Jim Marous:** Dara provides a roadmap for banks looking to adopt generative AI, emphasizing the importance of defining a clear strategy, choosing the right co-pilot for specific use cases, and fostering a culture of continuous learning and reskilling. This episode is a must-listen to for anyone interested in understanding how generative AI is reshaping the banking industry and unlocking new opportunities for innovation and growth. Not a day goes by that the opportunities and challenges of generative AI aren't the focus of both internal and external conversations in the banking world.

**[0:01:30] Jim Marous:** The good news is that generative AI case studies are emerging that can help pave the path for progress. How can generative AI assist in areas like contact centers, fraud detection, market research, and code modernization, as well as how can it enable more personalized and engaging customer interactions. So let's dig into what's happening. So welcome to the show, Dara. Before we start, can you share with our audience a little bit about your role at Microsoft, as well as a short background around how Microsoft is partnering with financial institutions on the deployment of new generative AI solutions?

**[0:02:09] Daragh Morrissey:** Sure thing, Jim, and it's great to be here. So a bit about myself. So I'm from Ireland, as you might be able to tell from the accent. I work in the worldwide financial services team and I'm really lucky in my job. I get to see all of the innovation taking place across all my customers across the world. I'm based out of Redmond, which is a suburb just outside Seattle where Microsoft headquarters are. And I'm amazed by the wave of interest in this technology and what banks are doing with it.

**[0:02:39] Daragh Morrissey:** It's really incredible. and I think the use cases that you mentioned, all of the banks are starting with things like code modernization. We have one big Wall Street bank doing that for about the last two years now and the reason why I like talking about it is that the benefits of it are very well understood now that it's been in market for two years and it just started out like a kind of autocomplete type capability but now there's so much more you can do with it.

**[0:03:08] Daragh Morrissey:** You can do things like automate your testing and what I find really exciting when I talk to banks about it is all the legacy code that banks have. I studied COBOL in college. That's how old I am. And there's a lot of COBOL code there. And we can do things now where you can ask a question about how this code was written. What is it doing? You can even do things like document the code in line.

**[0:03:33] Daragh Morrissey:** And then you can even do things like convert it to a more modern language. And we're actually doing that in Microsoft, you know, Windows. Windows is the product I'm using right now. A lot of that was written in C++, and we're converting it to a new language that's going to be a bit more secure and will reduce the number of patches that you get on Patch Tuesday.

**[0:03:52] Jim Marous:** Well, it's interesting because it's really a bridge from the old to the new, which is really exciting. And it provides a lot of opportunity because bankers who are legacy bankers, been around for a while can actually envision how this can move the organization forward in a much simpler way than we've ever done before. So when you look at the use cases around contact centers, fraud detection, market research, what are the most exciting use cases that you've seen in the marketplace in financial services?

**[0:04:26] Jim Marous:** Yeah, I think the ones that really excite me,

**[0:04:28] Daragh Morrissey:** I love the contact center one, you know, and within that use case, there's a couple of different stages you can take you can do things like just summarize the conversation between you and the customer and i have one canadian bank that just wanted to do that and play it back to the customer because you know anytime i ring a contact center i never remember what i said i never remember what we agreed and sometimes i don't remember what the follow-ups are so it just shows that you know the ceo of that bank told me i just want the the customer to know that I'm listening to them, you know, and that's kind of a basic one.

**[0:05:01] Daragh Morrissey:** We have one insurance company where we did first notice a loss where it started to pull out the key points in that conversation. You know, Jim drove into me in his Porsche and wrecked my Mini or something like that. It would actually pull those points out. And then you can actually, the advanced stage of this now is where you can do things like agent assist, where you're actually coaching the contact center agent in real time. And you can actually do it to do things like track compliance.

**[0:05:29] Daragh Morrissey:** So you only recommend a certain credit card at a certain time based on what they are earning and things like that. So it's very exciting. And what you can do with that use case too is... The other use case that's really popular is this knowledge... Concept of a knowledge base that's much more powerful than traditional search. So you can... A lot of banks have done this with their product documentation. And you can put a knowledge base around it where you can ask any kind of question.

**[0:05:56] Daragh Morrissey:** And what you can do is actually combine those two use cases. So you can actually learn from every conversation into the bank. You can push it into your knowledge base. And this way you get a continually learning bank, and it's learning from every conversation. And I think that's really exciting. And, you know, we had intelligent banking on marketing materials for years, but I actually believe this is a really intelligent bank, if you like.

**[0:06:20] Jim Marous:** Yeah, it's interesting because we interviewed a gentleman by the name of Brian Romley that mentioned this kind of – I'll call it a digital twin, which is really not a good analogy for it. But where you can build this insight into the customer over time through the question and answers that are going on as well as the products, tweets, and everything else. But it's all captured. And so what your service recommendations would be would be different than mine.

**[0:06:48] Jim Marous:** but on a one-to-one basis as opposed to a segment or something that's artificially developed or based on where I am in the life cycle or in my life finances. And you look at that and you start saying, you know, people are going to have to buy into it. But once they do, they're going to see the power of their information in a secure setting to be used to improve their financial wellness, which is really kind of exciting because it's a dialogue as opposed to AI by itself, the traditional AI that could build models.

**[0:07:23] Jim Marous:** This is really capturing insight on a real-time basis. And as you said, even sharing it with the human aspect of a financial institution in a way that can coach them into how to have conversations, which integrates multiple customer service sectors, which is really exciting.

**[0:07:40] Daragh Morrissey:** Yeah, I think the other side benefit we're seeing in a lot of the projects is one of the challenges that banks have is contact center staff are churning out at an alarming rate. Sometimes it can be every three, every six months, they lose about 30%. It's because it's a tough job. You're sort of seeing people at their worst. Sometimes they're hanging there for a while to get through to somebody and then they can't find the answer.

**[0:08:07] Daragh Morrissey:** So that is a very tough job. And what we're seeing now is improved job satisfaction in contact center staff. You know, the agents can actually start to do what we'd love them to do, which is sort of start to cross sell and upsell or, you know, become much more and do much more kind of higher value stuff. I'm not saying what they do currently is low value at all, but it's just it's it's just I think that is a tough and underappreciated job inside a bank.

**[0:08:33] Jim Marous:** You know, one thing that I really didn't understand is as generative AI has evolved, but I'm starting to understand a little bit more now, is the whole concept of co-pilots and co-pilot stacks and how you can choose co-pilots for a specific use case. Can you explain a little bit about the whole concept of a co-pilot?

**[0:08:51] Daragh Morrissey:** So again, if you think about the concept of somebody helping you to do your job or somebody helping you get started on a task and you're sort of overseeing what they do, I think it's an ingenious concept that Microsoft invented. And it's because there's a number of really important reasons for it, especially in banking. You know, we know this technology doesn't get things right all the time. And a lot of the use cases that banks use, they're sort of, they're with a tolerance of margin of error, if you like.

**[0:09:25] Daragh Morrissey:** And what you need for generative AI, if it does something for you, you do need a human to oversee what it's doing, make sure it's getting it right. And the thing that I'm really excited about is this concept of a co-pilot that can actually do stuff as well. So when we brought it into Office and Word and PowerPoint, it will actually not just help you find an answer, it will actually go and do something, which is really, really exciting to me.

**[0:09:53] Daragh Morrissey:** The fact that you can create a PowerPoint from a Word document and go back in the other direction is just mind-blowing. I mean, I use it all the time to record my meetings with account teams. and I meet account teams every week to get prepped to meet their customers. And in the past, I would sort of try and write down this into OneNote, type it in myself. I don't have an admin. And it's just great.

**[0:10:16] Daragh Morrissey:** It's like having a small team around you. And where it's going is really interesting is that you can have co-pilots now that help you understand data and you can have a conversation with data. You can model data in different ways. I think a lot of people haven't realized that power is just amazing if you start using it.

**[0:10:36] Jim Marous:** With this concept, with this capability, you can actually democratize data sets much easier than you have in the past because you can help the people that get the information actually use the information as opposed to simply just another dump of data, correct?

**[0:10:53] Daragh Morrissey:** Yeah. And even what we have is we see generative AI being used with the outputs of predictive AI. So, you know, a couple of the partners I work with, they have anti-money laundering solutions. And what they have is this concept of a co-pilot that is looking at all of the fraud incidents. And it's kind of helping you narrow those down to real ones faster by just talking to the data. And you can even then partially generate the next steps, you know, what documenting an incident looks like.

**[0:11:25] Daragh Morrissey:** It could actually sort of go and do that. So I think that's where it's really exciting. It's sort of, it's kind of at the end of a value chain around some of these processes. You can actually sort of make the inputs or the outputs even easier to digest and use.

**[0:11:43] Jim Marous:** So, if banks are looking to infuse Gen of AI into products and customer experience solutions, what have you seen that financial institutions really have to keep in mind as they're developing these things to make it so that directionally they're doing the right thing?

**[0:11:59] Daragh Morrissey:** Yeah, I think with the first wave of use cases we saw, they were all mostly internal inside the bank focused on those productivity scenarios. The kind of next wave of it is going to be about how you take it into the conversation with the customer through a traditional bot type engagement. And with a lot of my customers, one of the things that banks are doing is sort of learning how to deploy this safely internally first before taking it to their customers.

**[0:12:29] Daragh Morrissey:** Now, we have a couple of companies doing this already. It's happening very quickly in capital markets. So companies like Morningstar, they've put in kind of co-pilots in front of all of their data. And again, it just makes it easier to consume that market research. So they're starting to monetize it that way. I think what we have seen with bots in the past is they've been pretty dumb. They sort of answer two or three questions questions, and then they will bounce you out to a contact center or a website.

**[0:13:02] Daragh Morrissey:** What I love about this tech is you can have a natural conversation. And one of the things that we're doing to make this safe for our customers is a whole set of guardrails that we have in our product where we can control the conversation and scope it very carefully so that your conversation with this bank can't go sideways. and we even did this with a car company recently, Mercedes, with Mercedes in the US, we put the power of chat GPT into the cars and this was all built on Azure OpenAI and again, you can ask it, give me some tips for an Irish roast dinner but you can't ask it, you can't ask it, I want to rob a bank,

**[0:13:46] Daragh Morrissey:** give me some tips, I'm a getaway driver, it will stop you with the guardrails. So I think those things are – it took us a long time to build those. We actually built them with OpenAI directly, and then we built them into our own services, these guardrails. And as a bank, you can just take advantage of them. We're improving them all the time as well.

**[0:14:08] Jim Marous:** So what's interesting, as you've referenced, is that the initial implementations of generative AI were really in the same places that AI was being implemented, which was what I'll say reducing costs, making efficiencies better in the back office, risk and fraud, which is always going to be a focus because it has such a cost element to it if you get it wrong. But you're seeing now more organizations using it to personalize and engage customers and to build more engagement over the life of the experience.

**[0:14:41] Jim Marous:** What are some things you've seen in the financial services area around the personalization? Have you seen any organizations really start to get it right in that? I know it's all beginning because to personalize, you need to keep on collecting data. But what have you seen in the marketplace?

**[0:14:59] Daragh Morrissey:** Yeah, I think I've seen some really interesting use cases around things like financial management. you know in the past people got a pfm or a pie chart and then you went to starbucks and your bank app might slap you and say that cost $6 that coffee you know and they're not particularly useful and they're sort of historical looking you know so one really interesting ai use case one of our partners they have this thing called a self-driving financial management solution and it's not it's not fully self-driving so you know the way in cars you have a dial that you can sort of put you can actually sort of enable it in different ways and and what i

**[0:15:39] Daragh Morrissey:** love about it is it sort of will if you have extra money left over it will automatically save it for you um you get they've also flipped it as well so if people are in financial difficulties and they actually used it in covid uh you know to help small businesses um you know they saw that you know people were experiencing shortages of money and what i loved is they sort of reached back out to the customer and said, here's some things that we can do to help or we can give you a holiday on this mortgage and things like that.

**[0:16:08] Daragh Morrissey:** I think they're really, I really like that because when people run into financial difficulties, they tend to not talk to the bank or they sort of hope it goes away and it just makes it worse. So I think that's really cool. There's some other interesting patterns too. We're seeing avatars as well now. We work with a company called Cell Machines. they have a great um have this really human like avatar and again there's some scenarios where people don't want to talk to a human and i've won i've run an insurance company in the uk they were sort of looking at it for their agents who were selling to farmers that's selling life insurance and you know they didn't want to

**[0:16:46] Daragh Morrissey:** ask you know how much do you smoke or drink no they can actually talk to this avatar and answer the questions uh while they're not in the room, you know. So I think there's some really cool stuff emerging. I think where it's going is, I think you'll be able to have a conversation with the bank in a new way when this matures. I think being able to sort of just talk to your bank in a natural way is going to be game-changing.

**[0:17:12] Jim Marous:** Well, it's interesting too. I've talked about this in the podcast before. I had an experience with Delta Airlines where I had a first world problem with my seat as I came back from Amsterdam to Detroit. And I wanted to kind of solve it while I was on the plane. It was just one of these missions I had. And I talked to the bot. I talked to the customer service agent without the agent being there. I talked to the website.

**[0:17:36] Jim Marous:** I even went to social media and didn't get to the solution point I wanted. But I forgot about when I landed. And when I got back to my house about a week later, a human called me and said, by the way, I see that, you know, you've had some challenges and that you've used virtually every tool we have to get your problem solved, but it hasn't necessarily been solved because we haven't seen an end point of this.

**[0:17:59] Jim Marous:** I'm calling now to let you know, number one, we apologize. We know that we're working very heavily on getting all of our channels to ask the right questions to get the right answers, she goes. But the challenge right now is we have a problem with the different channels talking to each other where they don't repeat some of the things that they should learn along the way. She goes, but we're using generative AI to build solutions that will actually be captured and then distributed to all channels immediately so you won't go through this repetitive process.

**[0:18:30] Jim Marous:** I thought that's a key element that AI by itself wasn't able to do. But generative AI, because of the dialogue aspect from my perception is, you don't have to keep on the same dialogue or you can advance the dialogue based on what the last conversation was. I also see it as general AI being something that is a composable solution that you can make it as small or as big as you want based on what your needs are.

**[0:18:57] Jim Marous:** But does it level the playing field between your small and your large financial institutions or does it have the opportunity to? Because the technology is not overwhelmingly expensive. There are costs involved. But if you focus on a specific problem area, small finance institutions can develop tools that are as effective or more effective than large ones, correct?

**[0:19:20] Daragh Morrissey:** Yeah, I think the thing, for me, the biggest thing about it when I talk to my customers is I wouldn't overthink this. I think you're going to have to do it. I'll tell you a story. One of my kids, she got into trouble. I tried to manage her screen time on her phone and she hacked it. you know, and she can hang out. I love that. Generated a monster. So I bought this Nokia Flip phone from Amazon for 30 bucks and I made her go into school with it and she was mortified.

**[0:19:57] Daragh Morrissey:** You know, what it was. The funniest thing was it happened around this time last year. And I was sort of thinking to myself that day, is this thing real? Like it seems to be huge this thing and uh we're sitting down with my wife we're having a glass of wine and uh she sent this apology to both of us and the apology was about this long on the phone it was like and it was using all these words i hadn't heard her say you know it was like i undoubtedly have areas for personal growth and things like this and my wife is like that's lovely and i said no it's all i can see in this message are a ton

**[0:20:30] Daragh Morrissey:** of nvidia gpus uh spinning around and generating this technology from you with AI, you know. So she, now, and then I realized actually, you know, so she is going to be going into the workforce in a couple of years and she's using it now. She's using it now as a tutor. She was using it to do homework. We thought we called that. But so she's using it every day. So as a bank, you know, you're still going to have to hire people in four or five years and you're going to be, she'll have a choice.

**[0:21:00] Daragh Morrissey:** go to a bank that has AI and one that doesn't. It would be like, Jim, you starting in Microsoft next week and we wouldn't give you a PC. I think it's going to be like that. And I think, so I don't really think there's an option here. It sort of already happened. And so I think, and in terms of leveling the playing field, I do think it does level the playing field in that everyone is learning together.

**[0:21:24] Daragh Morrissey:** The thing I do like about generative AI only too is it's easier to build than traditional predictive AI, which is where the big banks had really good capability. There's a lot, though, to building a traditional machine learning model. You have to get the data into shape, you have to label it, and then you have to build a model, and then you have to do A-B testing, and then you have to bring more data to it over time.

**[0:21:49] Daragh Morrissey:** And with generative AI, we've even had some POCs that were built by our sales teams. I've never seen that before. So we didn't need to get data scientists in. we just sort of brought the data to the capability. So it's very accessible. And then, of course, we're going to be giving every bank sort of out of the box AI with Office. So even the tools that you use, there'll be co-pilots there. So I do think it's going to...

**[0:22:15] Daragh Morrissey:** I think what could be differentiating though is banks that start earlier will build up the capability inside the organization to actually use it. So even at Microsoft, we're getting a lot of training now on co-pilots, how to use, how to prompt them to get the right stuff back out. I think that's going to take a little bit of time. I think what we are seeing with banks is you don't need to do all in one go.

**[0:22:40] Daragh Morrissey:** You can layer these out very slowly or phase them out slowly. So even at Microsoft, we didn't get Word, Excel, and PowerPoint all at once. The first thing they gave us was Teams recording, And then it was a month later, I saw it show up in PowerPoint. And then, you know, so you can make it digestible from a user's point of view.

**[0:22:59] Jim Marous:** Well, it's interesting, too, because from Microsoft's point of view, right now it's all open field. There's a whole lot of things going on. But very quickly, you're going to have use cases you can deploy to financial institutions. Let's just take the financial institution set and say, you know, we may have a basis set. Very much like we did with cloud, where initially everybody's just using it the way they're using it. But then we realized there were so many elements to the deployment of cloud technology that were reputable, that were exactly the same from institution to institution.

**[0:23:31] Jim Marous:** So Microsoft can really play the role of a consultant saying, here's all the basic information you have to have, and we're going to get you there. And then where you want to fine tune it, where you want to make it specific to your organization, that's where you can take it on your own. But that's going to make that the learning curve is extraordinarily faster than it was with AI, which was still – it's still – yeah, we know how people have used it.

**[0:23:56] Jim Marous:** But actually helping somebody get to the starting line or beyond the starting line is difficult. And when you work with financial institutions, what roles and skills do you see as essential for success? And how are organizations right now answering that need? Are they developing from within? Are they using partners like Microsoft to get them there? Or are they all just fighting in the battlefield to get talent? Or a little bit of both?

**[0:24:25] Daragh Morrissey:** Yeah, there was. I think by predictive AI, there was a huge war for talent. You know, I did see one bank in China. I know they went out and they hired 30 data scientists, but they didn't really know what to do. I actually don't think that's sustainable or a good thing for a bank to do. you know what I think is more sustainable is people giving people the opportunity to reskill and do these roles you know because if you think about it you know nobody knows the bank's products better than the people inside the bank and their customers you know and when you if you hire a load of data scientists they don't they don't they don't know what you

**[0:25:04] Daragh Morrissey:** need to use it for they just want to be told what to build and it's it's a that was a big challenge I met a data scientist a couple of years ago who had a get an NFL sports agent he was getting paid that much and this was at the really high end of AI it was like sort of self-driving car a self-driving cars AI so I don't think I don't think that's really achievable and now what we what we've done in Microsoft is we have a whole set of tooling to help you build what I think is a data-driven bank and the really what you need to do is sort of take the really amazing data

**[0:25:36] Daragh Morrissey:** scientists and split out what they do across different people. So the really amazing data scientists can take data, build a model and tell the story about the data. And that doesn't need to be all one person. We can get people that can work with data, build a generative AI application now, and you can also build a testing or test it yourself. And so it's just becoming easier. One of the things we did as well is create this concept of a citizen developer or citizen data scientist where, Jim, you could just take a set of data.

**[0:26:10] Daragh Morrissey:** We can actually prompt you to say, it looks like you need one of these models that could be sentiment analysis or something. And it will actually build the model with the data. So I think over time, it's going to become even more democratized. You know, I'd say you copied it in Excel. You'll be able to do incredible things with AI in a couple of years. So what I do see banks doing though is they do want to become AI and data-driven.

**[0:26:33] Daragh Morrissey:** banks and I have seen CEOs say this to their people you know I don't want data to be I don't want this to be something in the corner by an analytics team I want everyone in the bank part of this value chain and if you don't want to do this maybe go somewhere else you know it was very blunt actually what they said and some of the banks that are doing this are um you know what they see is I see amazing impact you know they they actually make decisions based on their data.

**[0:27:01] Daragh Morrissey:** A lot of people don't trust their data as well, which is another thing. But that's, I think,

**[0:27:05] Jim Marous:** where it's going. I think it's interesting because it makes employees feel more valuable if they're involved in the deployment of solutions based on the data. And, you know, I've been banking my whole career. And so much of this time, I look back and say, you had to knock on the door, you had to beg, you had to plead, you had to barter to get reports, which were simply reports. They weren't solutions. They simply told you where you were on a specific day.

**[0:27:31] Jim Marous:** And, you know, organizations build entire banks around those data points, but it was never shared with anybody because there's a trust factor. There's a, you know, it's another thing to go from if it's a data sheet versus a solution sheet and going to the next steps. And it never went towards a uniform strategy or roadmap. And you did a presentation recently at bank director. And one of the things you really spend a lot of time on it looked like was talking about how do you start this process, the key steps in defining your AI strategy and the roadmap.

**[0:28:05] Jim Marous:** Because, again, just like any solution, if you don't have the roadmap and strategy set out for what you want to achieve, you'll never get there. It's just like a GPS system in that car. If you say, I want to go north. North is a whole lot of directions. There's different variations of north, not due north. And so you need to set that roadmap. How do organizations get started?

**[0:28:28] Daragh Morrissey:** Yeah, I think the first use case I would start with actually is your developers. It's the most mature generative AI scenario. And as you kind of build new applications for this, why not build them with generative AI itself? So I think your developers as well, banks struggle to attract developers. So it's really about getting more output from your existing developers. We sort of see 30 % to 40 % improvements in developer productivity. So that's a really quick win.

**[0:29:04] Daragh Morrissey:** And then I would think about the out-of-the-box AI, gen AI that you're going to get from us. If you start to introduce it to Teams and Office, you're going to hit a ton of use cases there that are sort of horizontal across the whole business. And then what you're left with then are a set of custom use cases. These could be things like, I would start, you could actually start with contact center, just enhancing what you currently have in your contact center.

**[0:29:31] Daragh Morrissey:** You don't have to rip out your contact center either. It's just about sort of adding the capabilities on top. And I think building a knowledge base too is a great way of learning how to use this inside the organization. Your product documentation isn't super sensitive. So it's not the end of the world if something happens there. And it's a great, this is where a lot of banks have started. I think the other really interesting use case I didn't mention is how your relationship roles inside the organization can use this.

**[0:30:01] Daragh Morrissey:** So one really interesting use case around advisory is how you might use it to prep an advisor before the meeting. So typically advisors, they log into lots of different screens. But if you could use generative AI to give them a pack just before, here's the client, here's Jim, here's his portfolio. Some of this might be generative and predictive AI, but it would kind of produce this for them. Summarize what happened in the previous meeting. And then in the meeting itself, you could also even start to have a conversation with the data.

**[0:30:34] Daragh Morrissey:** You know, you could say, if I add 10 % to my 401k contribution, what does that look like in 20 years? You can even start going there and then post the meeting, you know, summarize what happened, send back the actions to the customer. I think I'm really excited by those. And even when I meet, I have a financial advisor and he writes down stuff in pencil on an iPad. He's like me. My handwriting's really bad. I think those things, those productivity improvements are really cool.

**[0:31:07] Jim Marous:** Well, it's interesting because it can be generated so quickly. I mean, you can give it basic information. It can come up with really strong recommendations, solutions, insights that would take years to find because there's so much out there. You know, one thing we found is in recent podcasts, we've discussed how generative AI can assist in the learning and reskilling process, both formally and informally. How do you envision generative AI as a teaching tool as opposed to a learning tool?

**[0:31:41] Jim Marous:** You know, how do you see it as being a teaching tool to move entire sets of employees forward in their careers, but also just in their personal lives as well.

**[0:31:51] Daragh Morrissey:** Yeah, I think it is a great tool. I think one use case that McKinsey had recently was actually training wealth advisors, new wealth advisors with simulated conversations. You can actually simulate, the Gen AI can act like a customer and then actually can score, can kind of score the conversation or coach the advisor, you know, which I think is very cool. So I think the other thing is it makes – it's like – I think it's like you're promoted as well.

**[0:32:21] Daragh Morrissey:** So I even pay for ChatGPT myself, and I just – it's just – and I have –

**[0:32:28] Jim Marous:** You work for Microsoft, so that's a reason for ChatGPT. I'm paying multiple subscriptions already on multiple platforms because each platform is somewhat different in the way they come up with answers. And some are better in certain categories. Some are better in others. but I'm not second guessing my decision on those at all. No, no.

**[0:32:49] Daragh Morrissey:** But again, I even use it sometimes if I'm meeting a certain, let's say I'm meeting a certain persona in a bank. Now I'm not super deep in capital markets, for example. So sometimes I will ask it, tell me about this person's job or what are good questions to ask them. Now I don't trust the outputs of it 100%, but it sort of gives me enough to get started. And I think it's actually, you know, we have kids using it.

**[0:33:17] Daragh Morrissey:** Kids in Africa are kind of using it as a tutoring tool as well. So I think it is an amazing tool. It's sort of, I think one other interesting thing was Jensen, the CEO of NVIDIA, he was on a panel recently and he said, you know, domain expertise is actually going to be more valuable. So say your knowledge about banking is not in any of these GPTs yet, you know, or it is there, but it's sort of, you know what's right and what's wrong, you know.

**[0:33:49] Daragh Morrissey:** So I think he said domain expertise will become more valuable. He said like he wouldn't, but he wouldn't send his kid now to learn to be a coder. Now, I don't believe coders are going to disappear overnight. But he said, I think, you know, what I would focus on is building your own domain expertise in the future.

**[0:34:07] Jim Marous:** You know, that's that's so interesting because as I got used to using it at the very beginning, I mean, I started November of 2021 or whatever it was, 2022. And I found very quickly that the output was as strong as the input and or as weak. And a lot of people said, you know, I don't believe in it. It's not working really well. It doesn't give me anything new. And I quickly come back and go, what do your questions look like?

**[0:34:35] Jim Marous:** And you just discussed the fact that as we get better at working with this technology, the technology is going to get better at providing us capabilities we didn't have before. Yes, some of the things I'm doing is simply a time-saving device. In other ways, it's a way to avoid writer's block. If I'm building content, I will ask the question of an AI, a generative AI tool, and it will help me get over the hump on things I wasn't aware I might have been missing.

**[0:35:05] Jim Marous:** We also heard, and I've mentioned this a couple times on the podcast, we had a person that wrote a book that had multiple authors as part of the book, and he would take each chapter and go to the AI tool. if you were to give this a one out of five stars, what was wrong with what was written from your perspective? As opposed to asking us to take it to a higher level, he actually said, where's my blind spot here?

**[0:35:30] Jim Marous:** Because you can get somewhat lazy at times and think that everything it's going to generate is right. Well, it isn't. And you need to know, okay, what could be wrong to know how to go forward? So just kind of exciting. And we look at these opportunities, these potentials, but there's also challenges. What challenges do you see on the horizon with regard to the development and the deployment of AI solutions, generative AI solutions, in the financial services area?

**[0:36:01] Daragh Morrissey:** Yeah, I think one of the challenges is going to be, I think, getting some more precision on the answers out of it, I think, is one. Now, we are closing that gap in a couple of ways at Microsoft. so one of the things now we can do is get real-time data into a conversation we have these things called plugins so we can reach out and say what is the latest stock price of Microsoft and you get back an answer and you can see that you get traceability on the answer you can see that it is actually what it is and that's great for discrete answers that have a very defined kind of they're either wrong or right And then

**[0:36:42] Daragh Morrissey:** I think where we're going, though, is we're going to get better at helping you build your own models. It's going to become cheaper to do. At the moment, it's quite expensive to build your own foundational model, a large foundational model. Now, the price is coming down all the time. One challenge I do see, though, is banks love building stuff, you know, and stuff that they don't necessarily need to or will make any difference to their customer.

**[0:37:10] Daragh Morrissey:** so I had one bank that built a browser recently um again it was kind of my thing was why why would you do that you know but if you build um a lot of banks are trying to build a foundational model and then add on a use case and I don't think that's the right way I would start with the use case work back to what you need and then pick uh pick a model you know start with the smallest one and as as you improve as if you need more accuracy go to a larger one and that way you're sort of it's how you build technology in the past you know you never picked your database and then built

**[0:37:42] Daragh Morrissey:** an application it was it was always been working back from what you need to do so so yeah i do see i do see that as a challenge though i think um i think the other challenges are um are going to be around you know how you how we integrate it into into more applications i think i think we're we're doing a ton there in our platform to make that easier um but yeah i think and the other thing is analysis paralysis is another point uh pattern i'm seeing i'm seeing some banks kind of getting to a list of 50 use cases and then somebody comes in and tries to cost them out and assess them and i just think

**[0:38:21] Daragh Morrissey:** that's it's sort of not a waste of time but it's uh you know business cases never they never deliver what they said they were going to deliver right either in either direction you know so i think it's a waste of time i i think you should You really get started and learn. You don't learn to drive by reading a book. You actually get into a car. You go out onto the road. You don't go onto the highway first.

**[0:38:43] Daragh Morrissey:** You start small. And I think that's what most banks are doing. But I do see some banks that haven't done anything yet. And they're kind of trying to come up with a perfect use case. And I just don't think that's a good use case.

**[0:38:57] Jim Marous:** Boy, that is so key because I think that we see in the industry that it's going so fast, catching up is going to be so hard. I use the analogy of Bank of America's Erica platform, the voice banking platform. And a lot of organizations say, yes, we don't need that. I look back now and say, look how much information, how much dialogue they've captured that's going to give them a head start on generative AI deployment because they've already used this interaction.

**[0:39:27] Jim Marous:** And they've already built tools that feed data into the model. And I think just as importantly, consumers bought into it, consumers started testing it, and they started feeling better about it. And I think that's one part we're missing is just because we can build it doesn't mean they'll come, which means that, you know, even if I create this great interactive tool from a customer insight perspective, our consumers can be afraid of what they're giving to their bank more though, more than they do with Amazon, who's already built this trust.

**[0:40:02] Jim Marous:** And, and I think you're right. Those organizations that hesitate and say, I want to wait for the perfect solution, the perfect use case, or I want to wait for, to see if this plays out financially, that, that catch up is very, very costly. And, and, you know, the, just from the knowledge standpoint, being able to have all your employees feel comfortable with not only using these tools, but getting data and insights from these tools that they can deploy to their customers.

**[0:40:31] Jim Marous:** So, you know, finally, you know, what's interesting is there's certainly a continuous flow of new iterations of ChatGPT and many of the competing platforms. Each of them are introducing amazing expansions of capabilities. There was one this week. There's one almost every week from one of the platforms saying now we have this and it can go here. Where do you see this technology going in just the next one to three years? Yeah, it's really tricky.

**[0:40:59] Daragh Morrissey:** I actually have a slide that I presented of Frank Director and I had the three waves of deployment inside, bringing it outside. And then on the third box, I kind of say, I actually have no idea. And I'm saying that from a position of it's happening so quick and it's happened. Like nobody saw this coming. So, you know, I think, I personally think where it's going to go is as these models get more powerful. You know, Sam Altman talks about this thing called artificial general intelligence.

**[0:41:29] Daragh Morrissey:** We're not there yet at all, but I think when we get there, there's going to be some really interesting use cases there, I think, where you might have AI sort of monitoring the bank. you know you might have an AI that would look at the core systems and do self-healing if something goes wrong or it might look it might be able to see coordinated fraud attempts on the bank in a different way that banks can currently do and again this is just my humble opinion I have no idea where this is going to go it could be brand new financial products that are built from the ground up on AI you know I think it's exciting to speculate and

**[0:42:06] Daragh Morrissey:** I think the other thing that's going to happen too is we have this other wave of tech coming, which is quantum. And that's going to break encryption. We're working on that with all our partners to get ready for that. But I think that is also going to be interesting. What will quantum do when you're building these models? And then what will happen there? And banks need to be in the cloud, I think, to be ready for these things.

**[0:42:33] Daragh Morrissey:** I'm not saying that as a Microsoft shareholder. I'm just saying you won't be able to do much of this on premise. So you need to be sort of well on your way in the cloud to be ready for these things because they're going to happen faster than we think, I think.

**[0:42:48] Jim Marous:** Well, it's interesting too, because I can see a time when maybe the banks become the source of all this insight, but they give the consumer themselves the ability to use the tools for their own use, especially when you're talking about financial wellness and predictive ability to run models and all that. Generative AI makes this whole process easier overall. And I think if the bank becomes a custodian, but the consumer is giving flexibility on making it good for themselves, that's a product.

**[0:43:22] Jim Marous:** I mean, it's a generative AI product. So, Dara, thank you so much for being on the show. It's amazing to see. We've had multiple conversations with people that work at Microsoft because you're well ahead of the curve with regard to generative AI. But it's interesting because we've gone from tell us what it is to tell us how it can be used to now tell us how it is being used. And we're talking about a total of 17 months.

**[0:43:53] Jim Marous:** This is happening very quickly. And I think, you know, the key element to know here is you need to get on board. And even if it's just on your personal base, we talked about before the podcast how we both use these tools on an ongoing basis to make us smarter, to make us more aware, to test our ability to use it. And I think it's exciting because it does open the door to all new opportunities, some challenges as well, but ones that we can manage.

**[0:44:23] Jim Marous:** So thank you so much for being on the show.

**[0:44:25] Daragh Morrissey:** Thanks for having me, Jim. It was great to chat to you today.

**[0:44:28] Jim Marous:** Loved it. Thanks for listening to Banking Transformed, winner of three international awards for podcast excellence. If you enjoy what we're doing, please show some love in the form of a review. It helps us continue to get great guests like today. Finally, be sure to catch my recent articles on the financial brand and check out the research we're doing for the Digital Bank Report. This has been a production of Evergreen Podcast. A special thank you to our senior producer, Leah Hasid, and your auto engineer and video producer, Will Pritz.

**[0:45:00] Jim Marous:** I'm your host, Jim Marous. Until next time, remember, new technologies are coming at us faster than ever before. The question becomes, will you and your organization be ready when they do?
