Workday Podcast: AI for Financial Services: Beyond the Comfort Zone

Financial services has been using AI for years, but what’s the next step with this transformative technology? Nicole Carrillo, managing director of financial services at Workday, talks with PwC’s Vikas Agarwal about everything AI in this installment of our podcast series, Shift: Moving Financial Services Forward.

Audio also available on Apple Podcasts and Spotify.

While other industries are starting to explore the potential of AI, financial services has been using this technology for years. In many ways, it’s ahead of the curve, since game-changing applications for AI and machine learning (ML) in fraud detection and BSA/AML are already broadly in use.

In this installment of the podcast series Shift: Moving Financial Services Forward, we ask what’s next for AI in financial services, what’s possible, and how we get ahead of the curve again.

Vikas Agarwal, financial services risk and regulatory partner at PwC, and Nicole Carrillo, managing director for financial services at Workday, discuss how firms can apply AI across the organization, address the trust gap, and drive revenue streams.

Here are a few highlights from the episode, edited for clarity. Be sure to follow us wherever you listen to your favorite podcasts, and remember you can find our entire podcast catalog here.

  • “Whether it’s M&A, investment banking, wealth management, or even retail banking, many parts of operations rely on a lot of words and data to make decisions. And using AI can help you make better decisions. It can help you gather and analyze more information. And that’s key in financial services.”—Vikas Agarwal

  • “Our business is managing risk and taking risk-based decisions, and why wouldn’t we want to have the power of so much artificial intelligence behind it to help us make those risk-based decisions? Not just for our customers, but for our shareholders and the businesses as well to continue to thrive.”—Nicole Carrillo

  • “One of the things we talk a lot about is this control of humans in the loop, and to view your data as imperfect, but also view it as something that can accelerate helping you both clean up the data and also get answers.”—Vikas Agarwal

About this series: Shift: Moving Financial Services Forward

This series explores how you can shift the culture, skills, and technology within your organization to make the leap forward that’s required to thrive in the changing world of financial services.

Join us in-person or digitally at Workday Rising, September 16-19, 2024. Connect with industry peers and thought leaders, and learn how we’re taking work forever forward with a single, unified platform. Register Now.

Nicole Carrillo: Newsflash, AI is not news. We've been using it in financial services for years. They say our sector is risk averse, but actually, we've been way ahead of the curve, with game changing applications for AI and machine learning and fraud detection and BSA AML already broadly in use for years. So we already know there's nothing artificial about AI.

It's already transformed our industry. In this podcast, we're excited to be asking, What's next for us and AI? What's possible? What's stopping us? And how do we get ahead of the curve again? We'll talk about the trust gap. Trust is one of our biggest assets in financial services. So is it making us overly cautious about AI?

And if so, how do we get past that? We'll talk about what intelligence means and where it lives. The application of it inside the organization and at the edges. We'll talk about the power of AI in the hands of a smart CFO, how it can free them up [00:01:00] to become the futurologist chief, future officers even.

Hi, I'm Nicole Carrillo, managing director for financial services at Workday, and for this exciting conversation, I'm joined by Vikas Agarwal, Financial Services Risk and regulatory partner at PWC.

Vikas, please introduce yourself to our audience and share a bit about the areas that you've been working on lately in the AI space. 

Vikas Agarwal: Thanks, Nicole, for having me here with you guys. Just to introduce myself quickly, like you mentioned, I help lead our financial services risk and regulatory practice at PwC.

We work with banks, insurance companies, asset management companies, and payment companies. Helping them solve problems across financial risk, non financial risk, compliance and risk technology, as well as the internal audit domain. So we really touch a lot of areas of the business, doing a lot of interesting things.

Everything from helping people respond to emerging risks and emerging threats, to responding to regulators, to helping them grow their businesses with new technologies and new areas that they're expanding into. When I think about the topic of AI. Right now, I completely agree with you that this is nothing new for financial services.

But what is new is what we're seeing with generative AI And what makes generative AI very different is that you now have a capability of AI that's very self service. That people can use, that people can interact with when you think about these transformer models and applications with GBT, Gemini, Bedrock, with the hyperscalers.

You can interact with the AI very easily, becoming a self service way to use AI in your everyday work. And that's helping create the potential to transform how people interact with AI, how they use AI, how they gain trust with AI, and how they make their work smarter and faster. 

Carrillo: I think all of that was spot on Vikas I think for me it's interesting really to see how the AI that we've been seeing at the front end of many of our banks and financial institutions that we're used to interacting with as consumer, now start to move more into the back office to help.

The workers and the people within those functions to become more efficient and to become just as technology forward as we've been on the front end. So let's go ahead and get to it. I have a big question to kick us off. I was reading recently how AI is detecting 12 percent more early breast cancers and helping educators provide more personalized learning for students.

It's also helping to prevent wildfires and even mitigate the human impact of hurricanes and typhoons. So my kickoff question really is, if healthcare needs AI and teachers need AI, the climate needs AI, tell me why does financial services need AI?

Agarwal: So I think it's exactly what we're seeing in these other industries that we needed to make us smarter.

You know, we have a saying that I think is used a lot in the industry nowadays is that AI isn't going to replace people. It's going to be people that use AI that replace other types of people. And using AI can help you make better decisions. It can help you gather and analyze more information, and that's key in financial services.

When you think about whether it's M&A, it's investment banking, it's wealth management, it's even in retail banking when you're doing your operations, there's a lot of things that rely on a lot of words and a lot of data to make decisions. And that's what generative AI allows us to do much faster and also just with more comprehensiveness from that standpoint.

Carrillo: Yeah, absolutely. I think about it that, you know, in most financial services. Our business is managing risk and taking risk based decisions, and why wouldn't we want to have the power of so much artificial intelligence behind it to help us make those risk based decisions? Not just for our customers, but for our shareholders and the businesses as well to continue to thrive.

It's interesting because in another generation's time, how do you think we're going to look back at banking and insurance? Before it felt the full power of AI.

Agarwal: Look, I think it's hard because I think this is a transformational shift. And I think when we think about other transformational shifts that have happened, whether it was the Internet, whether it was email and even going back all the way to electricity, I don't think people could imagine how the world was going to change with those things.

I think everybody knows that the technology is going to change things, but I think fast forwarding is still difficult for people to wrap their heads around from that standpoint. So I, you know, I do think that, you know, in the future, all decisions will be guided and aided and assisted with AI based models that are going to be acting as another opinion that offer people more insights and allow them to make more precise decisions.

But how we actually imagine that coming to fruition, I think is going to be hard, and I think it's still going to take time for people to figure out the challenges that they're having to get there.

Carrillo: Yeah, absolutely. And so. You know, taking a line from my daughter's favorite frozen movie. If we're going into the unknown here, what are the practical steps that we think companies and executives should be taking in order to make sure that they're ready to go into this, you know, kind of unknown future.

Agarwal: You know, I think about crawl, walk and run with generative AI. And the first thing we advise our clients for doing on, on the crawl side is How do you start to get an internal chat GPT like function stood up? So today, if you're a client, or if you as an organization, Want to actually go to chat GPT and use it your prompts what you're typing into it becomes part of the public domain your data isn't private anymore So if you're asking it a question on how to advise a client or what the top risks are in a certain area You're exposing yourself from a privacy perspective. So the first step is how do you set up that basic capability inside your organization?

That allows you to use these transformer models to just do some very simple things. Maybe it's summarizing a document. It's asking questions. It's just doing what you do every day. It's editing an email and getting that set up in a way that your data on your information is protected, and there's different ways to do that to set that up, whether it's through enterprise agreements that you can have with the hyperscalers or whether it's, you know, through kind of figuring out, you know, setting up your own GPUs and your infrastructure within your organization.

Then, I think as you get to the walk phase, you start to get to, well, what are the use cases that are going to make our business better? And really, when I think about a business, you think about, you risk growth and efficiency. And I think we see a lot in financial services today around risk and efficiency.

You know, how do you think about, how do I become more efficient in certain areas? And how do I control the risk in certain areas and domain specific use cases that rely on domain specific data to help you do that. So one use case we've seen a lot of is taking regulation and translating it into plain English.

Very simple, something that people spend a lot of money doing with paralegals, but it requires a corpus of good data in terms of examples of what does that look like? And then the transformer model is smart enough to help you with humans in the loop, accelerate the work that people do today. And then I think as we get to the run phase, it's going to be about that growth area and how do you use it to actually grow your business?

And that's going to mean making decisions for customers with customers, being able to provide models to customers. But I think we're still a little bit far away from that when you think about controlling some of the risks that exist.

Carrillo: Yeah. Even just thinking back to the days when I was, you know, the CFO at a bank, I remember the people in legal who were the ones that could really explain and translate some of those regulations were always.

So busy, so bogged down because there was, you know, always the go to one or two people. I can only imagine the efficiency that would be gained to be able to kind of translate some of that for people to, in a way, be able to help self serve as opposed to having to continuously go to the one or two key people that, you know, they're used to going to.

So that, that, you know, things like that would absolutely be incredible, um, for not just efficiency, but, you know, letting people branch out and, and continuous learning as well. 

Agarwal: Absolutely. 

Carrillo: So, there's really two sides to this when we talk about AI. We can't really talk about AI and machine learning without talking about data.

And the sheer volume of data that's being handled by banks, much of it, like you said, is intensely personal, private, hugely valuable to anybody with ill intent. You know, we can't go out and start asking about clients out in, chat GBT, as you said. I don't know how much that data is. I think it's like 2. 5 quintillion bytes of data being produced worldwide every day.

And I'd say financial services probably owns a fair chunk of that just because of the nature of what, of what we do and how we process. So that volume probably presents an operational and regulatory headache right now, but also a huge opportunity for unlocking value. So really it's kind of a two part question.

The first at the, as the kind of the efficiency angle, AI relies on such a high quality, high velocity stream of data input, the garbage in garbage out type of, principle really rings true here. So how would, should we be looking at AI to work with our data management strategies, where can it be of most help?

And what do you think that customers or clients or companies out there really need to do? To make sure that their data is ready for AI. 

Agarwal: So look at the better data. You have the better output that you're going to have. But one thing that I think is very exciting about these models and these transformer models is that the data doesn't need to be perfect.

But now what matters is what controls that you put on the outside of when you get answers and when you get things. One of the things we talk a lot about is this control of humans in the loop, and that view your data is imperfect, but also view it as this is something that can accelerate helping you both clean up the data and also get answers that may be 80 percent correct and require that 20 percent of refining.

And that 20 percent is the gap in the quality of your data. So the more that you close that gap in your quality of your data, the faster your people will become and the more efficient, the smarter answers that you'll get. But we always tell people back that say that you shouldn't hold yourself back from getting started just because your data is not perfect as I think about what organizations can do to get prepared as they really need to look at what their most valuable data assets are and how do they really catalogue those data assets and start to make them as precise as possible.

So I think about ourselves at PwC. One of our most data valuable data assets is our library of risk and controls. We've been doing risk and controls for 200 years. I don't know if we took it as seriously as saying this data needs to be really clean. We need to have really well written risks, and we need to have really well written controls.

And now what we're realizing is the better and more precise we are at making sure our controls are really well, well written. The more and better our AI gets. And ironically, we're using our AI to actually analyze our controls for the five Ws, for the who, what, when, where, why. And using that to improve the data, and then that is giving part of our databases, our vector databases, that our AI is using to help us write even better controls or write new controls against regulations and risks.

So, it's that life cycle that people really need to look at, and it's really honing in on those specific data sets. That you think could be really valuable for your organization. 

Carrillo: Are you seeing customers start to build this into their data management strategy early enough? And I'm talking about this really in relation to where their data sits.

I know within many financial institutions, there's multiple data warehouses. And sometimes AI is going to need to use data from multiple of those locations. Are you seeing people starting to think about that in their data management strategy, as they think about where they house data as well?

Agarwal: We are. And I, and I think, especially as people move to the cloud and move to more digital enterprise applications, that forces people to really rethink the data from their source systems that are feeding these applications and thinking about how do they put more controls in place, both at the front end.

And in the processing of their data to actually clean up that data rather than while things are happening and processing is happening, rather than thinking about data management and data governance as a back end process where we're cleaning things up all the time. But I think that's where the opportunity comes in well, where people are upgrading their systems and they're becoming more cloud based and they're using the next generation of ERPS. From everything from their core banking system to their HR systems to their finance systems and really integrating data management into those transformations is key and we see the best institutions really doing that. 

Carrillo: Yep, absolutely. And so now moving from kind of that efficiency angle to the value angle, if AI can really help us spin gold from that data and find the needles in the haystack.

What do you think are the game changing applications for AI analytics, within, let's call it the financial services space? Are you seeing any experiments, successes or failures that you're really excited about, with the customers and the clients that you work with? 

Agarwal: Yeah, I mean, look, the way I would classify it, we talked about it in the beginning a little bit, is that we are seeing people, you know, enter that crawl phase and that walk phase.

But They're really focused right now on where can I take areas that are very manual, that are very word intensive, and how do I make those people smarter and faster. So a lot of applications in legal, a lot of applications in risk management, a lot of applications in compliance, internal audit, you know, some of those areas, credit risk is a great example.

But, you know I put in the credit risk area and I differentiate, one, we are helping clients and we're seeing clients use it to write, you know, a better credit risk memo. But I think people still hesitate on are we there yet to understand biases that can exist to say, are we going to have the AI and the Gen AI actually recommend, you know, a credit risk decision.

And I draw that line because I think there's still a lot of unknown unknowns with the models and with the bias that may exist in a model to really have the confidence and trust to do that with precision. You can help it, have it help you write the memo, have it help you write the documentation that you need, and that's a humongous uplift for organizations.

Carrillo: Yeah, absolutely. The credit underwriters who are helping to make those decisions that knowledge and that judgment that they use, that is an interesting thing. How do you, how do you tell when AI is smart enough to help even make those judgment calls? But like you said, it's AI supplementing people, not replacing people as part of these functions.

So moving on to something that I think our CFO listeners are going to be interested in is whose job is it to think of an organization's future self? Almost predict the unpredictable is really what sometimes we're looking at our CFOs to do. While they're not busy being financial superheroes, that is, you know, can you tell us what's going to happen and when it's going to happen?

I feel like if anybody needs AI, it's our CFOs. So beyond the Black Swan events that come out of left field, you know, for our industry, how accurate do you think we can get with AI in terms of foreseeing and navigating big trends? Do you think it can replace a team of analysts or again, really be that working alongside of them as they build out modeling?

How far ahead do you think we can get with that? 

Agarwal: Look, I think in 10 years, we'll be a lot further than we are today. I do think where we are today is still,  a place where kind of using some of these models is helping us get a little bit smarter, but it's not at the point where you're really predicting the future, and you can really sit there and scenario plan in a sophisticated way to think about different scenarios, using a large corpus of data.

I think we will get there from that standpoint, but today, I think when I think of where I see it applied to CFOs, I think it's helping CFOs clean up their data. I think it's helping CFOs document areas of their business more rapidly. It's helping CFOs with areas that are a big challenge in financial services like regulatory reporting, and getting through that, and even analyzing all the issues that they have in regulatory reporting.

Those very word based areas, I think that's where we're seeing a lot of uplift with the CFO community. 

Carrillo: So pivoting from the CFO to the person who really is their right hand in all of this is the CIO. They really are pivotal in navigating this AI driven era, too. How do you advise your CIOs to go about recognizing the solution that's right for their tech stack?

Carrillo: What can they do to minimize that security risk we've been talking about, and also implement it in the most cost effective way?

Agarwal: Yeah, so, I think there's a few things. I think that partnership with technology is extremely important. I do think thinking about and really understanding and this is a partnership that it's not just with the C.I. O. But with the chief risk officer, understanding your level one and level two risks around AI and how are you controlling them? When we think about AI and generative AI to us, there's model risk, there's legal risk, there's cyber risk, there's data privacy risk, and then there's emerging regulatory risk, and you really need to get your hands around all five categories.

Agarwal: You need to make sure that you have a single choke point of ownership around those things. I need to make sure you design processes that take into account those risks, that you have the right controls to understand, like you said, when you're putting data in, how is that, where is that data going? How is it being transferred?

Agarwal: How do you ensure that your data is not leaking out? What are the agreements that you've made, with the partners that you're working with from that standpoint? And that, that becomes extremely important. 

Carrillo: I think that makes a lot of sense. So moving on from, you know, the roles within the company, let's talk about AI more broadly with how it's, how it's perceived by the outside world and also the inside world inside of a company.

Carrillo: We hear a lot about trust in AI, like we were talking about, when do we think it's ready to start making some of these more judgment based decisions? How do we close that trust gap generally with employees? I know we've heard 62 percent of leaders welcome AI. Compared to only 52 percent of staff. Are there cultural changes for financial service institutions when they're implementing AI to make sure that not just the company, but the employees themselves are ready for it?

Carrillo: What should we be thinking about with that governance and transparency with employees? 

Agarwal: Yeah. So I think building trust is important and part of it is bringing your employees along for the journey. Right. And I think people need to understand and some, really need to figure out that this is an accidental threat to a business model and that same concept that you want this, you need this.

Agarwal: It's just like saying again, that would you want to do your job without electricity? Would you want to do your job without something that's going to make you better, smarter, faster. But people need to see that they need to understand that they need to understand that it's not there to replace them. It's there to augment them.

Agarwal: And I think training becomes a critical component. How you roll things out, how you promote the culture of using AI in a responsible way and how you show people what good looks like and also what bad looks like and being able to cross compare those things from that standpoint. What's interesting is one of, I think, the biggest challenges that we see in financial services is, you know, how do organizations move fast?

Agarwal: But how do they couple trust and model risk and all the things that go along with the regulators with that? And to do that, I think organizations need to invest a lot in training. They need to invest a lot in digitizing their platforms that even help them roll out these models to show people that they are testing these models.

Agarwal: They are thinking about what they are. They have ownership over the model itself. And that they are looking at some frequency monitoring the behavior of these models. And I think if you're taking those steps, that's what begins to form trust with your people that, hey, these are things that I could use as part of my job.

Agarwal: On the other side, I would say it's very important, too, that, you know, we talk a lot about the trust is the complacency side, that you have to have the right controls in place that people just don't, you know, use models or use GPT to do their job and aren't reviewing what they're doing, and I think we see a lot of examples in the press and media about people doing that and outputs not being perfect because it's not perfect, and I think that, again, goes back to that training element that's very important.

Carrillo: Yeah, absolutely. I think the training and the change management, too. I think that's something many companies struggle with is how to approach transformational change, not just from the how do I roll it out and implement it, but how do I change manage people and departments and get them comfortable with these things?

Carrillo: Because truly, when a company rolls it out, they're helping their employees upscale. They're bringing them up to the next level and really giving them the toolkit and the skills that they need to continue in the profession for, you know, future jet for future periods with the most, you know, up to date technology.

Carrillo: So I, I would hope that if, if you go through the process correctly, employees really realize that, that they're, they're lucky, probably that they're learning the most cutting edge technology as opposed to seeing it as something that is going to potentially replace them. So, we know that there's a battleground and a lot of competition out there for market share for profitability, external events and digital first disruptors are making it hard to predict risk and create profitable products.

Carrillo: So how can AI help our customers fight back? What can AI do for us today to help these companies really win, with kind of the the products and, and essentially profitability back to their shareholders? 

Agarwal: I will say this and it's a little bit of a controversial view is a lot of people are focused on efficiency and they feel like generative AI is going to give them efficiency and you see the headlines from a lot of consulting companies and others that a like 40 percent cost out 30 percent cost out.

Agarwal: I have a fundamental belief that we are still far away from that and that AI is here is going to make us smarter. It's going to make us better. But get smarter and better is different than faster and faster is not as easy as flipping a switch and I don't think people I think people should be focused on the faster and better before they think about the efficiency side of it because I think you can get lost in the efficiency side of it and you kind of lose the forest or the trees about actually helping people make their jobs easier and starting to [00:24:00] get processes that were very mundane and time consuming faster but you're not going to like getting productivity at scale is hard.

Agarwal: Like, I think we saw the same kind of trend with robotics process automation and, you know, other things that have come along that have promised a lot of cost takeout that is not as easy to get through the organizations and large financial services companies. That being said, I think it will get there from that standpoint.

Agarwal: And, and that's where I think the use cases of really focusing on the back office and focusing on things that you can control end to end or where banks have the biggest opportunities today. 

Carrillo: Yeah, absolutely. I agree with you. And I know Workday has kind of taken a perspective on this based on what we're hearing from our customers as we, as we talk to them in the industry, that AI is really helping them to get a true picture of the financial performance of their business.

Carrillo: It helps kind of de risk the future. Hopefully in the future, like you said, free up costs and make innovation with AI enticing everywhere else throughout the organization, really starting in one place and carrying it through the organization. I know that we've seen many of our customers using AI to offer employee and customer experiences that newcomers truly can't.

Carrillo: If once you're into that space, you're able to really customize to, to your employees and, and your customers and give them that experience that they need. It's also helping to win the talent war and keep insurance and banking as human as it always used to be. The more that we can free people up, like you said, from that mundane, just writing, writing words, the more that we can use them to apply that judgment and have a better experience with their customers.

Carrillo: I think a lot of things that we're hearing also say that we should look at AI as the back office brain, really embedding it as part of your business, helps people move things forward, people and planning for finance. It's the connected core and kind of foundational intelligence that powers the outward facing aspects of banking and insurance, which really are what we should be spending a lot more time focusing on is that customer experience.

Carrillo: If we can free up time that we're spending on the back. And I think the other thing that we hear a lot is really just get started, jump in. I don't know that you ever feel ready to start on this AI journey, but you have to start somewhere. Like you said, I liked your I liked your suggestion a lot about really kind of developing the internal chat GBT function as a way to get employees used to operating with this type of technology.

Carrillo: So I think I would absolutely just wrap it up by saying, get started. Just start thinking about what's one place. What's one thing that you think you could automate and try to tackle that first so that hopefully it becomes something. That spreads the organization and helps keep our customers and our clients, moving forward forever forward with all of this new technology.

Carrillo: So I want to thank you so much for being here today Vikas, I really enjoyed this conversation. I [00:27:00] appreciate the perspectives that you were able to bring, and I know I definitely learned something new and I hope that our listeners did as well. 

Agarwal: Sounds good. Thank you so much for having me. Appreciate it.

Carrillo: Have a great one.

More Reading