Artificial intelligence is rapidly transforming financial services, but where will it have the greatest impact? Recorded live at Point Zero Forum in Zurich, this episode of Macro Matters features BWC Executive Director Emily Slater in conversation with Sarah Hammer on the launch of BWC’s latest report, Generative AI and the Transformation of Risk Management and Compliance. Together, they explore how generative AI is reshaping financial institutions, the opportunities and challenges for regulators, and what responsible adoption means for the future of the financial system.
Sarah Hammer co-leads BWC’s Artificial Intelligence Working Group alongside Bill Dudley. Drawing on experience across the private sector, academia, and government, she shares practical insights on one of the most consequential technological shifts facing finance today.
Stream Macro Matters on all major podcast streaming platforms
Apple Podcasts | Spotify | Amazon Music / Audible | iHeartRadio | Youtube
Transcript
Slater: So delighted to have you here with us today, Sarah. We are recording this episode of the Macro Matters podcast here in Zurich, Switzerland. We are here for the Point Zero Forum, which is a really exciting convening bringing together technologists, innovators, market participants alongside policymakers and regulators to have conversations about what the future of finance looks like. And I mean, we’re on day two here and we’ve had a lot of really interesting conversations. I think everything is centering around digital finance, AI, quantum, and what the impacts are going to be on the financial system going forward.
Today we are going to talk about AI, as you are an expert in artificial intelligence. So before we dive in, I will quickly introduce Sarah to our BWC community. For those of you who don’t know her, she’s currently the managing director and global head of trading and trader compliance at Charles Schwab. Before that, she was executive director at the Wharton School and head of the Accelerator and an adjunct professor at the University of Pennsylvania Law School. You’ve had extensive experience also in public service and in the private sector in trading portfolios. You have a great perspective on all public sector, private sector, and academia, so you kind of have seen all angles of this. I think you bring a lot to the conversation. And of course, you’re leading BWC’s AI work alongside Bill Dudley. Anytime I talk to Bill about our work, he says, “I’m an imposter here. Sarah is the expert. I just sign on to these things.” He obviously thinks very highly of you, as we all do at BWC, and are so appreciative of what you’re putting into our organization.
So, we’re going to talk a little bit about the latest paper that we’re just releasing and that you have worked on. I want to start a little bit broader with some of the topics we’ve been hearing here at PZF and some of the landscape issues around AI. Many would say we’re in the middle of this AI revolution. Massive amounts of investment are happening. The speed of technological change is happening at a pace that a lot of us can’t even keep up with. From when we wrote our first paper, you wrote the first paper to now, I feel like things have shifted incredibly, right? There’s been analogies to other periods of technological advancement in the past industrial revolution, the wave of globalization, the so-called China shock in manufacturing, and changes of the 90s. You can look at other eras of innovation as well. There’s a lot of efficiency gains, a lot of productivity gains because of innovation, and a lot of economic growth gains. There’s also major implications for sectors because of this rapid innovation in terms of reshaping industries and workforces’ skill sets income distribution. When you look at AI right now in the financial sector, tell me how you see it changing the complexion of the financial sector and the financial system going forward and in terms of what that’s going to mean for jobs, for work, for the workforce, and for skills development as we think forward here.
Hammer: Thank you so much for that, Emily. I’d like to start by thanking you, Bill, and the BWC for the opportunity to be here with you today. It’s a huge honor and privilege to be able to work with BWC and the working group. The work that we are doing is really the contributions of so many incredibly smart, experienced professionals. It’s just fabulous how BWC brings everyone together.
I think it is a really exciting time. I’ve been working in AI since before it was considered very exciting, going back to my time on the board of the ITU and the UN Specialized Agency for Digital Communications. It was a slow progression back then in 2018. Fast forward to 2023, Chat GPT and the banking crisis. Now it feels like there’s an incredible announcement almost every day. It is fast and exciting, but it does come with risks as well. I think a lot of financial institutions have been going through this process where they’re trying to risk manage how they will implement AI. For a long time, they weren’t quite sure how to do that in terms of what AI would face the public. So internal use of AI for things like operations, for example, or we’re going to talk about risk management and compliance today, that’s different and subject to a lot of different regulations and considerations as well. But for client-facing AI, which we also write about in our second paper at BWC, there’s a whole set of rules around how we communicate with all kinds of clients and what sorts of controls are required when we do that.
This year, we did see a selloff in stocks and software stocks earlier in the year. It was because, in part, a lot of the analysts and industry watchers felt like those companies might be disintermediated by AI. Financial companies sold off too. I think the industry, at least from my perspective, saw that we better get moving on AI because people are looking at us and saying, “Why are they going so slow?” We better show people that we’ve been working on it. Since then, a lot of institutions launched client-facing Gen AI, but they’ve also begun to now think about looking at the institution from the top down. Some great firms are saying, “Let’s start by just looking at the whole firm and prioritizing, looking at all the potential use cases. Where do we think the most value can be added? And where do we think it’s really important for us to use AI to actually reduce risk?” For example, in risk management, they can reduce risk in some areas by using AI. From that, they’re thinking about how to measure that and whether they’re going to buy, build, or partner. They’re thinking, “How will I decide that, and then if I do partner, how am I going to do that?”
There’s a lot of discussion at Point Zero and elsewhere about what it means to really work closely to co-develop with one of these foundation model companies. I think it’s going to mean a lot of changes to the structure of business models. I think banks will have a different impact from asset managers, and they’re subject to different regulations. Banks are subject to very strict model validation rules and requirements. Asset managers are subject to a lot of other rules and requirements. For risk management and compliance, there is not that client-facing element.
There’s a lot of data, manual processes, content, and still a lot of regulatory requirements for things like reporting. What we talk about in this paper is that that’s a real great use case for Generative AI. By using Generative AI, you can not only apply it to a process, but reimagine how the workflow is going to look in this world of AI. Maybe we don’t need all of these different processes, but we would determine where the control points we need and where humans need to be involved. I think what that is going to do is elevate certain decisions to humans that are equally, if not more, important. It will also demand a high level of expertise- not just about the process and the regulation, but also about the models and how they work.
Slater: There’s been a lot of discussion about that, of course, here at PZF, including the human element and decision-making point needed to build trust in the technology. You also mentioned the business decisions that are going into this and where one can find increased efficiency and productivity. What jobs are no longer going to be required? How do I reskill those to some of these areas where there is going to need to be human intervention, or how do I rescale those for growth, company growth, and more competitiveness?
You mentioned how companies are looking from top to bottom at how to implement AI. We’re here in Zurich – UBS is obviously based here and has featured heavily on the agenda. It has talked about their experience here at the forum including how they are thinking about AI from top to bottom. Their CEO spoke this morning. He made a really interesting point that I would like to get your thoughts on. He said he thinks markets are underweighting the economic impact of AI, but overweighting the speed of implementation.
I just would like to get your thoughts on this tension. We’re seeing massive valuations and expectations around AI, but everyone thinks this is going to happen so quickly. What is happening right now? What do you think is the time horizon we’re really talking about here to see some of the economic benefit?
Hammer: I think it’s a really important statement, Emily, and I can understand that position. There’s a whole process and evolution that financial institutions like UBS are going through, and some of them are way out ahead like UBS. Others are really not quite there yet and need to get ready, because it’s quite clear that you need to be AI smart going forward.
I think there’s a two-stage process for those institutions that are ahead and thinking about the top-down approach. The first is looking at all those cases within the institution like we talked about earlier to value them and prioritize. The prioritization can be based on a lot of different things. It can be based on economic value to the firm, cost savings, headcount savings, or reduction in risk. We have all these manual processes we’ve been concerned about, and now AI can do this. We’re just going to be taking human checkpoints. There’s a lot of really exciting stuff happening there, but it’s also incredibly complicated. The tech infrastructure involved around AI – everything from cloud to compute to the foundation models themselves – all has to be thought about in this context of what use cases are we going to implement.
Part two for a lot of those firms is how we are actually going to monetize what we are doing. We can have internal cost savings and internal risk reduction, but what are we going to offer our clients? In some cases, firms are doing sort of an internal beta. Then they think about if they’re doing this for themselves, maybe their clients would be interested in that as well. For that second part, it will take a little bit more time. Some firms are really far ahead. I mean, it’s just incredible what they are doing. But others, it’s going to take a bit of time. And I think it also goes to the point of human capital and talent. This is really complicated technology. There just aren’t a ton of people existing in current roles who would have that expertise. So, firms have to be thoughtful about upskilling, bringing in talent, and what works best for their organization.
I think the markets are going to be volatile. I think it was the only thing I would say about the markets that I think I can predict. You know, they’ll be volatile and it won’t just be AI. But I do think there’s value longer term that we’ll see down the road.
Slater: It’s hard to predict the time horizon, right? There’s going to be winners and losers, early adopters and first movers, and others that take a wait-and-see approach. I find it really interesting here at PCF how we’re talking so much about digital assets, payments, digital money, and where that conversation started 10 years ago to where it is now in terms of the market maturity. We’re also seeing the policy discussions around it, where 10 years ago it was the same thing. Regulators had no idea what we were talking about with digital currencies and digital payments, Libra, stablecoins, and CBDCs. It’s come very far now in 10 years. I think for me, AI is kind of the equivalent, right? We’re at the early stage of AI, and I think the speed is accelerating.
Switching a little bit to the use cases that you’ve been talking about and some of BWC’s work, we want to hear you share a little bit about this latest paper on risk management and compliance and the use case there because that’s really compelling. We’ve also released a couple of papers to date – we just had one also in April that we released alongside the spring meetings. Tell me a little bit about the use cases that BWC is looking at, and then let’s dive into this current paper and what you’re seeing there.
Hammer: Sounds great. The working group has done such fantastic work planning and thinking about the structure of this publication series. It’s amazing how fast things change, and yet we had sort of a framework set out that continues to make a lot of a lot of sense. One of the first things we did was publish a paper talking about what we plan to do and evaluate. This goes into BWC’s core mission and principles, bringing people together from different jurisdictions to talk about what the challenges are and how we are going to govern together what changes need to take place. That first paper talked about a lot of different use cases at a high level. This included everything from trading to lending to mention of risk management, and then financial advice, which continues to accelerate as well. It really set out a core framework around AI and what generative AI and agentic AI are, and how those are determined. It also included a bit about governance and how complicated it is regarding AI regulation, sector-specific regulation, and firms’ own internal policies. I’m excited to continue to build that out with a group, and it’s been a wonderful venture.
Then we plunged right into AI and financial advice. I think that working groups’ timing was perfect because our paper came out around the time that those firms started to put out their client piece in Genia that a couple of banks put out, not just chat bots, but advisors. There’s a continued push towards agentic financial advice, and we’ve actually seen that now from a couple of firms. Some real fast runners in the industry are getting approval from the regulatory agencies to launch these agents. Advice is a complicated topic. You can move from just education, saying, “Here’s what a portfolio might look like. Here’s what asset allocation is,” to a recommendation, such as, “You should buy this stock or you should buy the SP 500 Index Fund.” If you do that, then you’re making a recommendation and you’re subject to best interest requirements here in the U.S. You can also give financial advice, which is fee-based and requires fiduciary duty to attach. That’s a whole different set of regulations. This brings in the whole question of can agentic advisor carry forward with judiciary duty. We’re moving forward on that question, but it’s not quite clear yet. As firms think about whether they want to be in education, to move towards recommendations, or advice to offer clients everything possible using either generative AIA or agents, they have to think about those different frameworks. The implications are huge. The requirements for review controls and monitoring are different, but the advantage of it is that digital advice can be offered any time of the day for anyone. But the advantage of it is that advice can be offered anywhere in the world. Tt can be offered in different languages, at greatly reduced cost from human advice, and so there are a lot of advantages to moving down that road.
This is the benefit of technology. The 24/7 piece of it is important. However, the financial system doesn’t work on a 24/7 basis. That’s a lot of the tension and friction that will have to be sorted out as the technology continues to develop. Some of the conversation with digital currencies, settlement, and 24/7 settlement include some of these issues there will have to be figured out.
Let’s dive in on some of the risk management and compliance issues that you’ve raised in the paper. I think this is really compelling for near-term use case for generative AI because there’s real efficiency gains happening in terms of AML, backlogs, and whatnot. So, give me the state of play on how this is AI is creating efficiencies here. Also in the paper, you introduced agentic AI, which we haven’t talked about in some of the prior papers, but it is a big piece of this. Give us an overview of what this latest paper is and what this use case is all about.
Hammer: Thank you, Emily. I am really excited about this next paper. And it’s funny, I said to Bill, “We can’t put the word compliance in the title, because no one will read the paper. No one cares.” And Bill was like, “Everyone should read that paper if it has the word ‘compliance’ in it.” But fast forward a little bit, and all of a sudden, compliance is the most exciting use case for AI. We’ve seen A16Z put out a paper about it and there’s more application in the industry.
Part of the reason that the working group discusses is that risk management and compliance are naturally very data heavy, very regulatory heavy, and there are a lot of manual processes. Those workflows have developed over time based on what the requirements were. They’ve just grown and grown and grown. And there are humans deeply embedded in those processes. In some ways that’s a huge challenge because as regulatory requirements grow, there haven’t been enough people to provide the resources that a firm might need for compliance or risk management. When I say risk management, I’m really thinking of all aspects of risk management- not just compliance, but technology, risk, and operational risk.
We haven’t talked about third-party vendor risk, but that’s a really important one as well. What AI can bring to that is the opportunity not just to apply technology to improve specific processes, such as looking at hundreds of pages of a document and summarizing it for someone, but actually looking at that whole workflow. What triggers a requirement in compliance or risk management? What came in? And then from there, what happens to it? Where are the humans embedded? And where are the controls embedded? Then we can think about, “Do we actually want to do it that way? Do we need to do it that way if we can use Gen AI to do X, Y, Z? Because humans are already embedded in those processes, we already have humans in the loop. But again, generative AI and agentic AI will work to just bring the decisions or the approvals or certainly things that are required by regulation to the human. And the human will have a higher level of expertise and knowledge of the models. The models can still hallucinate.
We’ve talked about convergence in the working group as well. So, you know, it is an exciting opportunity for the technology, it’s a huge opportunity for firms to reduce costs to free up human time to do other things like learn or plan or even approach different areas of the firm that needs to be addressed. It’s been great to see the working group really dig into that.
Slater: I certainly echo your comments about the working group, just a fantastic collection of perspectives and people who are all really passionate about this and really contributing to ideas and thoughts, contributing to the work.
So, you mentioned earlier that in the initial paper, we kind of outlined some of the regulatory challenges and some of the governance challenges around AI. So I want to talk about that a little bit more. And then, you mentioned sort of some of the regulatory challenges within the current applications of AI, right? Who regulates this? Is this, if it’s a third party, is it a tech? In the tech firm, they’re regulated differently than the banks are regulated.
I mean, even within the financial industry, like you said, if they’re making recommendations that triggers different regulations. There’s a lot of jurisdiction by jurisdiction regulatory issues that are going to have to be addressed, right? But then there’s this whole sort of cross border and global coordination and governance challenge as well.
You know, we’re seeing the US putting a very pro-innovation kind of posture. You’ve got China with obviously just more of a state-based model, but very innovation-forward. You know, you’ve got the EU AI Act, you’ve got the UK putting out kind of policy guidance. So, tell me how you see this globally in this kind of global AI race that we’re in right now. How we can think about policy guidance that is not a zero-zone game and that is a little bit more collaborative and cooperative.
Before we get down a path of just sort of nation by nation, jurisdiction by jurisdiction, patchwork policy frameworks and regulation, are we too early to even be talking about this? How do you see that?
Hammer: Yeah, that’s a hard question, Emily. It is a patchwork- even more than a patchwork, it’s just so incredibly complicated. As you mentioned, some jurisdictions like the EU have a real horizontal approach to AI. So, they’re looking across the jurisdiction and they’re determining what is high risk, what is lower risk, and then applying different requirements based on what those risks are, what the use cases are, really the use of the technology. You have other jurisdictions like the US that may have sort of a number of different frameworks that are relevant. We have been very focused on innovation in the U.S., which is great. We all love innovation. And we do have very stringent regulatory requirements in the financial sector, in my view. And as you mentioned, it varies depending on type of institution or business model.
So banks, for example, have been subject to certain rules from the banking regulators about model validation. We discussed this in the third paper, banks have been subject to a rule called SR 11-7 up until recently that required model validation. And that is a complicated process that involves something called collective review, where different parts of the firm will look at a model. You can validate a model once, but the model can change or the use cases can change. And so it’s kind of a living process. Living target. And then more recently, that was rescinded and replaced with SR-26-2, which is much more principles-based. There’s a threshold, $30 billion to where it’s applicable. But it also specifically carves out Gen AI and Agentec AI. And the agencies put out a request for information to develop their thinking on that subject. It’s just so complicated. So it’s moving, it’s changing. What I would say is for banks and other institutions to keep thinking about how to understand risk anyway.
So just put aside all the regulations and ask how we are already managing risk, how are we already treating our clients, and how can we be consistent in that with our approach going forward. I don’t think changing the regulation means that all of a sudden a lot of financial institutions are just going to do whatever they want. I think they’re going to continue to be responsible stewards of finances.
The same is true for asset management. Lots of very specific rules like we talked about earlier in the context of advice, and I think we’ll continue to see a lot of changes both on the sector specific and then on the jurisdiction specific. Another example as you mentioned is China, which has a very privacy cyber security focused, very stringent approach on those topics. And then Singapore actually has put out what I think is some very useful guidance and toolkits. Very clear toolkits for risk management, for AI, for example. Now, if I’m a global firm, I have to look at all of that and say, well, how am I going to operate in this environment? And it is difficult. I do think a lot of firms are still trying to determine what the best approach is.
I think part of what we’re doing today at Point Zero is talking about that. There are the right people here to make some of those decisions or to continue to move the ball forward, and we’ll do that tomorrow as well at our own table. Bretton Woods Committee has just really been a leader in this type of challenge. Bringing together people from different parts of the world, different jurisdictions with incredible insight. So in future papers we’ve talked about addressing governance and addressing regulation, but it’s not going to be easy. With this technology, we won’t be able to say, “Here’s exactly the statute that we should have 10 years from now.”
Slater: There’s probably not one. I think in this geopolitical economy that we have, countries are going to make different sort of economic policy choices and therefore regulatory frameworks because of the choices that they make about what is important to them from a from a policy perspective. Market participants have to, as you said, sort of navigate that.
There’s not going to be a single statue, a single rule, a single global, harmonized kind of, “Here’s what everybody agrees to.” I mean, we don’t even have that hardly in traditional financial services or banking sector, right? It’s a challenge anyway, isn’t it? It’s a challenge with any cross-border issue. And it’s kind of the defining challenge, I think, of international cooperation right now. The defining question no matter what the issue is, no matter if it’s AI, digital assets, or trade. We’re having to figure out how countries can make different policy choices, are going to choose different policy pathways, and how the market has to navigate that, because it’s not going to be this integrated, singular standard any longer. There is just more variance, I think, in the frameworks and policy frameworks and in the regulatory environment.
That’s been certainly a lot of the conversation, as you said, here at PCF, certainly around AI, digital currencies, digital payments. So, I’m excited to continue that conversation. We’ll learn more. It’s been a fascinating conversation here at PZF. I think it’s BWC’s first time.
Hammer: Okay. Yeah. Well, BWC is really making a splash.
Slater: And yeah, I’ve been really pleased at the number of members that we have here as well, speaking or attending. And just the quality of conversations and where the dialogue is. It’s been a great experience for us so far. We’ve got a couple more days to go and a lot more conversations to have. But I want to thank you, Sarah, for everything you’ve done for BWC, for the latest paper, which we’ll be putting out here in the next couple of days. I want to just thank you for joining and sharing your insights. It’s been so great to have you.
Hammer: Thank you, Emily. I have to say, I think it’s really been BWC, Bill, and the working group, and just these brilliant people who speak up with incredibly important points. You know, putting our heads together around global governance is not easy. To be able to bring those people together, have smart conversations about it, and move the ball forward, it’s an honor to work with the group. I’d say you and the staff at BWC are so fantastic. It’s just a great opportunity to have the chance to work with you.
Slater: Tara, we’re very fortunate to have you. Thank you for the paper and thank you for all of the AI work. Thanks for joining this episode of Macro Matters.
Hammer: Thank you so much.