AI in practice: building on data integrity to implement the proper tools

Michael Ruttledge, Chief Information Officer at Citizens Bank, and Bal Shukla, Head of Business Transformation for U.S., Financial Services, at Infosys, know that customer primacy, deposits and fraud are top of mind for banks and credit unions. Only through data can those priorities be effectively tracked. As these seasoned technology experts tell BAI CMO Holly Hughes, AI is the engine to make sense of that data. And AI has the capabilities to transform a bank’s tools for uses across departments, creating a cost-saving ecosystem.

Don’t miss the other insights, methods and tools that can help your financial institution shape and evolve its AI strategy in the BAI Deep Dive: A practical approach to AI.

You might also be interested in:

• To prepare for gen AI tomorrow, financial institutions should prioritize data organization today
• The future of banking is AI
• 6 ways financial institutions can harness the power of AI

TRANSCRIPT:

Holly Hughes, CMO at BAI:  Hi, everyone. My name is Holly Hughes. It’s my pleasure to host a conversation today focused on what financial services leaders should consider when implementing AI within their organization. I’ll be speaking with Bal Shukla at Infosys and Michael Rutledge at Citizens Bank. During our conversation, we’ll gain their perspectives on the challenges to overcome when deciding to build or invest in AI capabilities. Bal and Michael, welcome so much to the conversation. Really looking forward to chatting with you today. Michael, I’ll go ahead and start with you. Certainly, traditional AI has shaped financial services for a number of years now and generative AI is accelerating the conversation for sure. Would love to know from your perspective, does this mean that banks in your view are beyond the experimentation phase with AI and starting to think AI first in shaping technology priorities?

Michael Ruttledge, Chief Information Officer, Citizens Bank: Yeah. That’s a really interesting question, Holly. I think the reality is AI has been around for a very long time, and I think banks such as Citizens have embraced that in terms of if you think we’ve been using machine learning models in our credit underwriting, in our fraud systems, in our marketing systems for many, many years, and I think so as the financial services industry. So I think what’s changed is obviously with gen AI and large language models, it’s actually caught the imagination of everybody, which is a good thing, because I think it certainly brought AI to the forefront of the conversations with our board members, with the CEO, with the CFO, with the most senior levels in the company, and there’s definitely a lot of interest and I think that will be reflected in the level of investment and the amount of mind share that we give with AI.

And the advances in gen AI are real. There’s a lot of hype around it, but the reality is there’s some game-changing technology that’s out there and we’re looking forward to really adopting it and using it in a safe and secure manner. But I think we already had over 90 use cases just from a brainstorming session alone at the bank. And now we’re not pursuing all 90 of them at once. We’ve got a very careful, structured approach use case by use case. But yeah, I’m excited about what’s there.

Holly: That’s great. Bal, question for you to build off of that, talking a little bit more about that balance, traditional AI versus gen AI, do banks have to decide between investing in traditional versus gen AI? Or do you think it’s a natural progression? Can organizations successfully scale both? What are you seeing with all the different clients that you work with?

Bal Shukla, Head of Business Transformation for U.S., Financial Services, Infosys: Excellent question. Holly, as Michael mentioned, I mean, organizations have already been delving into AI for past few years and there have been multiple use cases there. How I seen the industry today is gen AI is more an extension of the classical AI use cases and to solve the same business problem. For example, Michael mentioned a lot around credit underwriting, fraud, marketing, these use cases and all. Look at the time it takes to get a marketing campaign. It takes couple of days to do that piece because you have to create creatives, you have to create multiple things, which goes to multiple approval process and all. Now gen AI can help us create those things ahead of time, create those things much faster now. So it almost augments the same business problem in moving forward.

Similarly, you look at contact center use case and all, wherein you improve the productivity of the reps to provide the answers to how we do that. So you almost look at is AI has been there for a while and which is what we call classical AI, but there are certain forms that we could not cover over there. That’s why gen AI comes to make it help and that’s why it can almost augment it to expedite the same business value that we get from end-customer perspective.

Holly: Now there’s certainly lots of buzz about that, lots of buzz about AI within the organization, lots of vested parties I’m sure that you’re having to speak with and talk to. Michael, another question for you. From your experience, how does AI strategy implementation best work in terms of all those stakeholders? Is it top-down? Bottom-up? How do you get organizational buy-in beyond just the technology department who I’m sure is very excited about it?

Michael: Yeah. I think it’s a combination of both. I think the senior-most leaders in the company certainly have to have buy-in. So I do think it helps that top-down approach and certainly if these topics are being discussed at a board level, then the CEO, the CFO, et cetera, have to really know about it. And here at Citizens, they really embraced it. We’ve done multiple education sessions with our board and with our executive committee. So there’s definitely been, I would say, some top-down emphasis on it. At the same time, the ideas come from the ground up. So what we’ve done is we’ve reached out. We’ve had multiple forums to solicit ideas. We created basically a cross-functional steering committee, if you like, across the bank with representatives from risk and marketing and technologies and security to really examine the ideas that were coming forward, and some of the business teams, and really prioritize those use cases. But it was from the ground up that we were getting these ideas. So I do think it’s both. It’s a combination of the two.

Holly: Yeah. It makes a lot of sense. And another question for you, Michael. At the current stage that we’re in, there’s certainly a lot of choices for banks when it comes to AI solutions, which vendors, what technology they might opt for, when cost and time are certainly of the essence. Any high-level advice that you’d give folks listening for making sense of all the choices, all the noises out there, helping leaders really make good business decisions in this area?

Michael: Yeah, I think it’s very important to think of the business value that you’re going to get out of these use cases. You don’t want science experiments. You really got to think through at the end of the day what’s the value that this will deliver for the business. So I think that’s very important to think through. The first two use cases that we’re adopting at Citizens, as Bal mentioned, one of them is for our contact center. What we are doing is we’re using our internal knowledge management platforms, leveraging large language models to really surface up in a much more efficient manner than we’d done previously information to our customer service reps that allows them to service our customers much more efficiently.

So before they’d be surfaced up a number of… If they’d ask a question, “What’s the best product for X, this customer?” There will probably get six documents back they had to wade through to get the answer. Now with gen AI, it’s very specific on the answer and we can test that. So we have controls in place that look at the answer it gives, compare it back to the answers our legacy system would’ve done and make sure it sits within those boundaries and it’s been very successful.

Holly: And Bal, I’m sure you see a lot of use cases as well. What would you add from your experience?

Bal: Right. So Michael mentioned. So every institution has got hundreds of use cases…in the gen AI space. You talk about HR, finance, legal, security. You have enough use case going on. But it’s also about making it real, which means the way we have done in the traditional AI space and every use case will take life of its own. It took a long time to get the use case implementation at scale. In gen AI space, we have to get at a scale. One of the key pieces here is looking at building block approach. Michael mentioned about large language model, the embedding model, the generation that needs to happen. If you look at a lot of use case have similar patterns. So one of the thing that we are finding is take the use case view and because it has to go to steering [inaudible 00:09:12] for everything, that way it has to take a pattern-based approach, for each pattern solve using a block-based approach.

With the building block approach, now each of the building block can change. Today we have one LLM model that can do much better for one kind of use case. Tomorrow another LLM model will come. Today we think embedding in certain way. Now it’s getting further fine-tuned. We talk of fine-tuning. We talk of zero-short prompting. So these things will evolve in technology. As long as we keep a platform-based approach with the building blocks across board, which everything can move and evolve as you do that piece, that’s how organizations can further up and thrive in this area. That’s what I’m seeing in industry today. Take a step. Take some use case. But then take a pattern-based approach so you can scale and go further up.

Michael: Yeah. I think that’s a really good analogy, Bal, because I do think this is… It’s definitely building blocks. So step one is to get your data altogether, to get… The quality of the data is so important to this. So we followed this data platform approach where frankly over the course of the last few years, we’ve consolidated all of our data into a data lake in the cloud and we’ve come off of appliance solutions like [inaudible 00:10:31] and Teradata and IBM Big Insights, et cetera, et cetera, and we’ve consolidated all that in one place. And that then we can run the modeling on those. So we have tools like SageMaker and H2O that our developers are using, but we’ve had some of these tools for a while. Now that we’re able to do the same leveraging some of these tools with some of the generative AI technology, we’re building and it’s an ecosystem. We’re building a lot of this on the existing platforms that we built in the cloud and that’s what’s allowing us to move at speed.

Holly: Yeah. Let’s talk a little bit more about that technology piece because you bring up some great points. And Bal, I’ll point this one to you, but Michael, feel free to chime in. A lot of times when there’s new technology, the obstacle that comes up is legacy systems or, “How will I get this to all work?” It can be really difficult to integrate into a tech stack. How have you seen AI plug into existing systems effectively? And Michael raised a couple great examples. Any real-world examples that you could share with the folks listening today?

Bal: Absolutely. Absolutely. I’ll come with some of the top ones that is hitting bank today. And as you look at right now, we are running into a phase which we call interest rate risk and liquidity risk. What it means is how can banks bring more deposit into the bank? But more than that, how can you build stickiness into the customers through that piece? What it means is the core banking of every bank is mostly on legacy platform and banks are transforming, but mostly on main from legacy platform. But if you don’t know the pattern of transaction that’s coming in and connect back to the data point, Michael mentioned about having an entire data lake on cloud, if you don’t bring the entire pattern together, know the insights and be proactive in identifying some of those patterns and take quick measures on fraud as an example, [inaudible 00:12:30] continue to bleed on fraud. Fraud is a big thing for today.

So the whole idea is the level of maturity of an organization to able to integrate from the core banking, which is where your transactions are there, connect it back to proactive monitoring into a space called fraud, an example, and then give the golden signals back to your decisioning platform before it becomes a fraud. That’s very important use case and example to call out. It also goes back to the point is how we reduce what they call positives, that we get [inaudible 00:13:02] comes out from there. How can we reduce those signals which are false positives that comes out? This again comes to same point is because the more you have false positives, the more people are then the operation side to go and look at all of those false positives now. So again, it tends leads an impact to the bank from figuring out having more people in operations.

So if you look at this one example to call out because everybody’s focused on this, I’ll give another example as well. Same is based on deposited, every looking at pricing and what we do for relationship-based pricing for organizations. So you’re now creating capabilities to ensure is how can we do a better pricing strategy based on relationship of customers to do a better rate to customer who deposit, because you can’t have one size fits all as an example for deposits. So now you want to look at relationship-based pricing for a customer based on total customer value or [inaudible 00:13:56] in the memory of the customers what customers can do. So again, same area to go on.

And then when we have deposit, it goes to interest-bearing deposit and non-interest-bearing deposits. So how do you look at entire piece, again, bank to maintain liquidity. So there’s several examples. I gave few of them, cash flow forecasting, another example, commercial space to help businesses build your cash flow forecasting. Again, everything that you talk, you go back to your core, go back to your core legacy platform, core lending platform, core deposit platform, connect the dots across the two to come forward. So bank setting advantage of it and that’s how it’s all coming together today to move forward.

Holly: Well, we’ve covered a lot of great ground. You guys have shared a lot of great information. I’ve got one more question for each of you and a little different tone. I’m going to ask you to reflect a little bit and then one of you to maybe look out into the future. And Michael, we’ll start with the reflection if you will. I mean, you’ve certainly gained a lot of knowledge over the years in the area of AI. As you reflect on your experience, what’s the one thing you would tell your younger self when he was just venturing into this space when it was kind of the shiny new penny?

Michael: Yeah. I would say take the time to learn, so really use this as an opportunity to learn about the new technology. It’s something that I’ve always done all my career. I’ve always been fascinated in new technology and hands-on getting to learn it. We didn’t have the opportunity to talk today, but the other use case that I think we’re working on that I think is really going to take off is code generation. So we’re using gen AI to develop code. And certainly if I could wind back the clock and start programming again, would I love that opportunity for a code generator to be generating my code for me and, better than that, doing all the documentation that I hated to do and all the testing that I hated to do. So I just think that really experimentation is really, really important and taking the time.

We’ve got a real passion for learning at Citizens. We’ve introduced academy programs. Over 400 of our colleagues have taken academies on the cloud, on software engineering. We’re developing one on AI machine learning, generative AI. And certainly we’ve also developed badging programs on data engineering, all the new database, plethora of database technology. When I started there was two databases. You were either on Oracle or you’re on Db2. Now, it’s Snowflake and MongoDB and Cassandra and SQL Server and… I can go on and on and on. So there’s so much out there to learn, but it’s fun. It’s exciting. And I really would encourage people to embrace this and just make sure that they’re setting time aside to roll up their sleeves and experiment with some of this new technology in a safe and secure manner.

Holly: Sure. Absolutely. Great advice, Michael. And Bal, we’ll have you take a look at the crystal ball. If the three of us were together in a year or two having the same conversation, what do we think we’d be talking about? Where do you think AI will be in a year or two? And what do you think the hot themes will be?

Bal: Absolutely. So I would look at AI is not by itself moving. There are other parallel streams moving at the same time. So I would look at what is regulators going to put in front of us in two, three years down the line. So I would say much more robust regulations that is coming up from where it’s today. That’s number one thing, on regulators perspective coming up. Where would technology be moving by that time? So we only heard in the GTC of NVIDIA, they have now launched Blackwell, another one which is a giant chip. So more compute, much faster compute is going to happen by that time.

And Michael mentioned earlier about moving to cloud. So every bank would be probably embracing cloud in some shape and fashion, which it basically gives a scalable capacity. And I’m sure every institution would’ve built a better data platform, which means you have got a capability to have better data platforms, harmonized data, which is a critical success by the way, harmonized data available to you. You know the customer directly now and data is available to you. If those things have come together, now comes the advantage of product innovation, how fast can you create products. The risk management space would be more commertized by that time because now you have got all those things available which can help you get now all the triggers very, very quickly.

So it’s all going to be now how banks can connect across each other with their counterparts, with the fintechs and all, but the intent would be how can they create better capabilities for the customers and customers. So one of the things would be there if we meet two, three years down the line, first thing is operations will be much more harmonized. We’re starting with human and loop. Hopefully, that’ll get further reduced because whole operations can be more and more automated with AI. We have better data. But more importantly, there’ll be more capability coming out [inaudible 00:19:20] which are much more sound for end customers. We talk of personal finance management, wealth management. We talk of integrating commercial and consumer customers together. So more and more products would be there in the market. And I think bank would be the real [inaudible 00:19:35] across the ecosystem that you look at, across healthcare, across manufacturing. That’s industry which is going to propel, because now you have insights,…networks available. So that’s an amazing story to go down, which is solve the basic problems that can go further.

Holly: Well, a lot of exciting stuff I think in our future. And hopefully we will connect and compare notes and see how everything is going. But Bal and Michael, thank you so much for sharing your insights with us today, helping really leaders understand some of these use cases, how you’re going about that work. I think you’ve given them really a lot of great things to think about and take away. And really want to thank all of you who are watching. And be sure to check out all the great content in this BAI deep dive series on AI. And you can head to bai.org for more insightful thought leadership on the key issues that financial services leaders are facing. Thanks again, and have a great day.

Bal: Thank you.

Michael: Thank you.