Claude for Financial Services Keynote

Please welcome to the stage head of revenue at Anthropic Kate Jensen. — Good morning everyone. Thank you so much for being here today. I'm Kate Jensen, the head of revenue at Anthropic, and I am so excited to welcome you to the future of finance powered by Claude. I want to invoke a scene that's probably familiar to many of you in the audience today. You're three coffees deep. It's 3:00 a. m. You're furiously checking that last model. You have a 100 tabs open. You have a 5 a. m. prep call ready for a 6 a. m. client call. Today, that's all about to change. We are so excited to announce Claude for financial analysis. This is the industry's first unified intelligence layer that transforms how financial professionals will work with AI. Now you will have a virtual first class virtual collaborator powered by AI that will help you to get your work done better and faster. What this is a tailored version of cloud for enterprise. It's specifically built for financial analysts, and it's equipped with the nuance, accuracy, and reasoning that you need to handle the complexity of your work. Since 2021, Anthropic has been building AI systems that are helpful, harmless, and honest. At our foundation is safety and trust, which is exactly what financial services companies need. When you're managing billions in assets, good enough just isn't good enough. AI is transforming every industry, and nowhere is this more prominent than in financial services. The complexity of modern markets, the velocity of information, the sheer volume of data that you need to process, it has reached a point where human intelligence alone, no matter how great, just isn't good enough. From speaking with many of you, our customers, we're hearing that the world is bifurcating. There will be two types of investment firms. Those using AI institutionally and those who are losing their top talent to competitors who are over the past year, we have been partnering with leaders who are breaking new ground across every segment of the financial industry. The teams at Bridgewater have been using Claude since 2023. This is to power their investment analyst assistance which solves complex models. Their CTO tells us that Claude is really powering their efforts to push the boundaries of what's possible. Down in Australia, Commonwealth Bank is making a massive bet on AI. Their CTO Rodrigo sees our partnership as the foundation of their global AI strategy. And here in New York at AIG, Peter tells us that they have completely reimagine underwriting with Claude. What used to take weeks is now taking days. Timelines compressed more than 5x and accuracy has gone up from 75% to 90%. Underwriters can now serve their customers better and faster. Now, the solutions we're sharing today, we didn't create alone. They're the result of deep collaboration with the entire financial ecosystem. At our foundation, we have our cloud providers. AWS and GCP provide secure, scalable infrastructure that financial institutions demand. I am thrilled to share that today Claude for enterprise including claude for financial services is now available on AWS marketplace and it's coming soon to the Google cloud marketplace as well. We hear all the time that AI is only as

good as the context it has. We're excited to partner with platforms like Box, Datab Bricks, Palunteer, and Snowflake to bring your company's data into cloud so that you have the context you need to get work done that's actually relevant to your day-to-day. And today, we are especially excited to announce new integrations to critical external market data sources for informing your most important financial decisions. We're partnering with the companies you already trust. Fact set delivers comprehensive fundamentals and consensus estimates. S& P Global enables access to Cap IQ financials, market data, and transcripts. Dupa provides AI verified fundamentals with source citations. And between Morning Star and Pitchbook, you have both public investment research and private market intelligence at your fingertips. Our consulting service providers are also turning this technology into real business transformation. Deote and KPMG are modernizing entire organizations and deploying AI agents at scale. PWC and Turing are solving critical regulatory challenges, navigating constantly evolving compliance requirements and Slalom and Tribe AI are modernizing core operations from migrating legacy cobalt infra to accelerating due diligence with intelligent document processing. Collectively, we are building the AI ecosystem that will define the future of financial services. As our head of revenue, my team and I spend all of our time talking to customers and partners. We heard from you how important it was to bring all of this context, both internal and external, into cloud to help you make use of this technology. We care deeply about defining the future of AI and human collaboration. and thank you for your feedback. We are thrilled to bring this solution to you today. Now, I'm really excited to welcome up to the stage two of our partners who help to exemplify the power of our ecosystem in action. Peter Lerzie is the chief strategy officer at Kensho Technologies, S& P Global's AI innovation hub. He's been defining how AI transforms financial data into actionable intelligence. And Vikrambot is vice chair and financial services industry leader at Deote where he has been helping global financial institutions navigate digital transformation. Please welcome them up to the stage. Did you pick that song? — Yes. — It's actually in my writer. — I love it. Thank you so much for being here. You have both been instrumental in shaping the frontier of AI adoption at your companies. From your unique vantage point, Peter with data and intelligence, VROM with implementation and transformation. what's the one breakthrough that you've seen that has really created an aha moment at your companies? — So, yeah, I think I mean for me it's really adoption. Um I mean I've spent the better part of the last decade trying to get financial services uh companies to adopt AI and we've had some really great successes with traditional machine learning. I would say in the world of generative AI that conversation is very different. Uh it it's a it's kind of ho how can you serve us as quickly as possible the data that is in a format that is optimized for these large language model use cases uh for generative AI and how can I make sure that the data that you're providing is trusted that it's uh accurate that it's kind of foundationally uh a part of the workflows that we've always relied on and so I think for us you know the big kind of uh surprising thing about generative AI has just been the speed of adoption, the rate of adoption. Um you know we it has it's been very influential in our roadmap. The products that we've kind of foregrounded uh for us our uh LLM ready API uh which is the data behind what we've done with MCP uh has been we've seen like enormous traction with that. uh our clients want to make sure that when they're interfacing

with this technology that they can rely on trusted data and data in workflows that they that are already existing that they can then automate and optimize with large language models. And so that's really been uh a huge surprise. It's going into these client conversations and not saying, "Okay, but here's what AI can do here. Here here's the kind of magic that you can unlock. " They already know that part. Uh it's now how can we truly make our company or institution or group an AI forward group and also uh make sure that we can trust the outputs and for that you need trusted data. So that that's kind of been the big thing that we've noticed. And Vic, what about you? As you think about bringing this technology, what are those aha moments that get people to really adopt it? — We have really an AI first strategy in everything we do. Some of it is internal in every service that we have. We have starting to embed AI, you know, kind of front to back. On the client side, what we're seeing is there was an initial focus on productivity as a key theme of value in through AI. But what is really happening is that is expanding to can we develop new products based on AI? Can we re reimagine distribution? Can we reimagine frontto back processes? It is really changing the conversation from a pure productivity as a value driver to revenue generation. Can we do better risk and control? Can we drive mo more employee and client experiences? So as the appetite and the aperture of value has really expanded, it is really brought AI into everything that does and every piece of the client organization. — I love that. And we'll start with you this time, but what do you think you need to do differently as an organization to make sure that you're really capturing that full value, not just adopting the technology? — Yeah. So I'm probably going to give a in incorrect non-political answer. In financial services, — we are a highly regulated industry. Everybody knows that. So the real challenge is in this battle of I'll say innovation and risk management. How do you coexist in a manner that there's not 20 checkers for one doer and the equation is actually flipped that you actually have more people driving innovation across your organization and you work in coexistence with your audit teams with your model risk teams you know all the other players that are necessary in order to get adoption um successful. The other uh example is that AI first strategy that I mentioned. I think more and more clients are actually saying AI is not just about driving 30 to 50% productivity in the CIO organization or anywhere else. It is really about changing the way we do business and to do that they are how do you actually bring the humans we have this philosophy of humans and machines coexisting in the age of AI and how do you drive the human change at speed and scale together with the pace of change that's coming through AI — how do you think you do that how do you get them to coexist and adopt at the same time — I I think it it is not one answer. There's a number of different levers that one can pull in order to drive that. Um for example, some organizations have done um executivemies where the top 300 executives really drive training. Some are doing hackathons to really think about certain processes or value streams that they can completely reimagine. um there is you know complete AImmies and trainings that are being pushed forward at you know for developers but also for general users. So it's not one lever I think in the end that's going to drive the speed of human adoption of AI. It's a number of things that need to be done together. — Love that. — Peter, what about you? What do you think people need to do differently and institutions need to do differently? Well, I think I mean I think that honestly it comes down to like and I think Vicram touched on it. It's kind of a culture thing and I think that a lot of that culture does have to be uh top down, right? You have to have uh your leadership teams really buy in to this concept of AI first. I totally agree with that. I think uh but part of that top down is to also embrace the fact that in many of these organizations innovation's coming from the bottom up, right? people who are in the trenches doing day-to-day work are surfacing uh amazing ideas uh not just for productivity uh which I think is oftentimes the focus but for huge expansions of revenue opportunity I think I think that's absolutely

correct uh you know I think that it's kind of changing a mindset around innovation and I do think a big part of that is understanding especially in our space there are going to be workflows where being maybe a little bit less risk averse is really important and being able to create those that room for experimentation uh is going to be really important. I mean certainly for us at Kencho uh right we we've been able to have that room and the way that it works for us is that we kind of operate somewhat independently and as a result we kind of have a little bit more room to experiment but then you know crucially it's also understanding where those boundaries are uh and where that risk is unacceptable and and that's I think where for us we've also seen an enormous benefit uh in having S& P's trusted data as our kind a foundational core competency. You know for we we've been on this journey uh creating this grounding agent uh where for us it's ensuring that when a large language model is interfacing with our data that we're taking responsibility for automating where you know where in our entire data universe we're going to pull and query those data sets and then surfacing that data with citations. And so it's both right enabling experimentation but then also making sure that your response is and whatever the output is truly grounded in verifiable fact. So I think there's a balance there and creating that trust is something that's really important to us. — I love that push to innovate, that desire to really harness all the energy that you have from your organization, but to do it responsibly. It's really at the core of a lot of what we do. Thank you both so much for coming up and thank you for the partnership. — Absolutely. Pleasure. Thanks. — And now I'm thrilled to welcome up to the stage my colleague Nick Lynn who leads our product for Cloud for Financial Services. Thank you all. All right. Thank you, Kate. My name is Nick Linton and I lead product for Claw for financial services. I am also a recovering investment banker and private equity investor. So, I was the analyst K mentioned frantically trying to debug the model at 3:00 a. m. while building the pitch deck running on no sleep. That's why I am especially excited to share what we've been working on. As Kate mentioned, we're announcing the Clawed Financial Analysis Solution, Anthropic's first offering built specifically for finance. Our solution is built on three main pillars of what comprises of an agent. First, the models. Now, something I always think about is that the models we interact with today are the worst they will ever be. Even so, Claude today is not just state-of-the-art for code. Claude has been specifically trained for finance domain knowledge and excels at tasks like data analysis at scale, financial reasoning, and even Excel manipulation. Now the research product flywheel is a critical part of an enthropic strategy. We firmly believe in working with customers like yourselves to really understand where the models are working, where the gaps are, and to borrow a term we use internally, hill climb, to improve the next generation of our model capabilities. The finest agent benchmark published by our friends at vowels. ai has picked up a lot of traction in the industry and is well representative of many financial reasoning tasks. You can see cla and sonnet far outperform competitors. Similarly, cla is also very capable of actually doing the work. One of our partners, Fundamental Labs, has built an Excel agent called Shortcut on Opus. This agent was able to pass five out of seven levels of the financial modeling World Cup competition and scored over 83% accuracy on complex Excel and financial reasoning tasks.

Again, far outperforming competitors. The next pillar of our solution is agent capabilities built on top of these models. Now model intelligence needs to be translated into something that's actually useful for humans. Claus's agent capabilities are flexible and composable and really aim at solving the core problems you all work in every single day including building multimodal reports like pitch stacks investment memos analyzing and visualizing data like benchmark analyses or stock price and volume charts and natively reading and writing Excel and PowerPoint documents. I'm excited to announce that for the very first time, these expanded output capabilities are now in research preview exclusively for select customers of our financial analysis solution. We've also expanded usage limits to really support the deep work analysts work in every single day. And we've integrated cloud code to support use cases like analyzing much larger data sets, Monte Carlo simulations and risk analyses. Finally, the model intelligence and agent functionalities are delivered through our very flexible platform. Our solution is the first agent with unified intelligence across all the core finance data sources you all work in. As Kate and Peter shared, we're excited to partner with industry leaders like yourselves to build those critical integrations into the agents with many more to come in the future. Importantly, we're delivering white glove finance specific implementation and onboarding to really assist you all with deployment, education, and change management. Now this is of course built with enterprisegrade security and trust. We're sock 2 type2 certified and by default we do not train any of our models on your data. Now our solution is really targeted at solving the problems you all know well. In investing, insights are alpha. The ability to sort through the noise, build a thesis, get to an investment decision quickly is paramount. As we've heard from many of our customers, this is of course true for the sell side and the buy side. To summarize the pain, analysts work in a deluge of documents every single day and spend hours researching into data sources that are difficult to verify. They painfully prepare analyses and tick through models sell by cell and they manually prepare investment memos and pitch decks often pulling allnighters on a tight deal timeline. With Claude, we can start to address all of these pain points. Now, let's show you what that looks like in action. Meet Sarah, a hedge fun at Acme Capital. It's 2 p. m. Her portfolio manager burst into her office with an urgent question, which I think many of you know well. Our target company, Velocity Athletic, reported terrible earnings yesterday. revenue is down 12%. But somehow the stock price is up 17%. Trading at $71 per share. But PM needs to know, is this rally justified by their new strategy or should they sell into the spike? And he needs an answer before the market closes. Now, let me show you how clot transforms this typical four to five hour scramble into analysis under 30 minutes. First, Sarah connects her tools. Look at what is available in one workspace. All the tools she works in S& P Global, Morning Star, Faxet, Dupa, and even her firm's internal box documents. No more juggling 14 different browser tabs. Everything she needs is right here with unified access. Sarah starts a comprehensive query asking claw to pull from multiple data sources simultaneously. Watch what happens. Cloud orchestrates across platforms. S& P Global for transcripts, Morning Star for reports, Box for internal analyses, the Loopa pulling eight quarters of financials.

And here's the key. Sarah gets synthesized intelligence, not just raw data dumps. Output full earnings transcript analysis with red flags from Q& A. The CFO disclosing a 400 basis point margin hit from the tariffs. Competitors look like they're doing better. Pace running has been in Vietnam since 2019. And most importantly, you can see full financials linked to the original data source like we're seeing here with the Loopa. So Sarah can verify the results instantaneously. To go deeper, Sarah asked Claude to create visualizations and specific analyses like an annotated stock price chart, comps and benchmarking analyses, and even discounted cash flow model. As you can see, Claude is now pulling key corporate events from SEC Edgar and the web, price data from S& P Global, historical financials from DOUPA, consensus estimates from fact set and creates several visual artifacts. First, an event annotated stock price chart showing the 17 SP percent spike with all key events marked. Emergency board meeting, CFO stock sale, earnings surprise. Next, a comprehensive comps table. Velocity is trading at 21 times IBIDA against peers at 16 times despite having worse fundamentals. And here is where it gets powerful. Claude has built a fully auditable discounted cash flow model with a functional case selector, projections tied to assumptions, and even a perfect whack calculation without needing to be prompted. Now, this DCF model projects a base price of $54, suggesting that perhaps $71 is overly optimistic. Finally, Sarah actually needs to prepare a deliverable for her PM. She asks Claude to create an investment memo using her firm's templates from Box. Claude searches Box for any relevant internal information, including a memo template and any past eventdriven trades as reference. In seconds, she has a professional memo with an exact sum recommendation to fade the rally, clear rationale with supporting data for market performance and comps, and specific action items and risk analyses, all properly cited. The bottom line, the market is overpricing a complex operational challenge. So, take profits now, buy it back later, cheaper. Now, let's recap what just happened here. Sarah delivered institutional quality analyses in under 30 minutes, a typical process that takes 3 to five hours across multiple platforms. So, she didn't just save time, she uncovered insights she might have missed like the CFO sale or the competitor's existing advantages. Now, we believe this is the future of financial analysis where AI fully augments and an analyst capabilities where the best insights across all of the platforms come together in one intelligent workspace. Now one example of these capabilities coming to life is from the Norwegian sovereign wealth fund otherwise known as MBIN the largest sovereign wealth fund in the world to share a quote from their CEO Nikolai Claude has fundamentally transformed the way we work at NBIM they've achieved 20% productivity gains that is 213,000 hours back a here to focus on what really matters that is better decisions and returns for the Norwegian people. As Nikolai put it, Claude has become indispensable. Now, building these solutions requires deep partnership with all of you. Everyone in attendance today will receive a one month free trial to the solution. I am personally very excited to spend much more time with all of you to dive deeper into the problems we can solve together and to shape the future of financial services with both our

products and model capabilities. And with that, I'll pass it to my colleague Jonathan Pelosi, our head of financial services, and our distinguished panel of industry leaders to share their own journeys with Claude. Thank you, Nick, and to the DJ. Well, I didn't choose that song. I love it. Let me put these down. And thank you to everybody for joining us. Hopefully you get a sense of why we're so optimistic about being able to support every one of you and everyone of you dialed in on all of your tasks as it relates to financial analysis. But don't just take my word for it. I've got a much more impressive panel that can speak to what they're doing with this type of technology and how they're doing it. And as Nick alluded to, in my role overseeing all financial services for Anthropic, that effectively means we support the biggest banks and insurance companies, hedge funds, and asset managers in taking this technology and practically applying it to your firm in a way where you can use it to drive real value and real transformation. And I say practically because all too often we'll read about the theoretical headlines of what this stuff can do and you're like, "Oh, it'll transform the way we work and it'll improve productivity by 50%. " But by the time my team and I sit with you, we often pull back the layers and say, "Well, how do you actually make that happen? tailor this to your organization? How do you actually work on change management? " getting these individuals comfortable with using this technology. Those are the types of things we obsess over and we're excited to chat further about. So, with that said, I'm going to call up my esteemed panel. If you could come on down and give them a chance to introduce themselves. Thank you, team. We're going to do quick intros uh because I don't think the titles necessarily do justice to all of the AI work they do. And we're going to start with my good friend Michael here. Take it away. — Thanks JP. Uh my name is Michael. I co-head the COO group at uh DE Shaw. That means is my team and I uh work on big firmwide transformation initiatives. And there is no bigger firmwide challenge and opportunity that we're spending time on uh right now than AI. Thank you, Michael Lloyd. Thanks, JP. Lloyd Hilton. I am an AI lead at HD Capital. So, for those of you who don't know us, we are a major PE firm, one of the top 10 largest globally. And we're a sector specialist. So, we have about 50 portfolio companies in the B2B SAS and services space. Um, and we've been driving a major transformation effort with those portfolio companies for the last two and a half years now, partnering with the anthropic team. — Don, — morning everyone. Thanks for having us JP. My name is Don Vu. I'm the chief data analytics officer at New York Life. For those that aren't familiar with New York Life, we're Fortune 69, over a trillion dollars of insurance in force. Um, and we're one of the largest insurance companies in the world. I'm responsible for our overarching AI and data strategy and its execution. — Thank you, Don. And Frod, last but not least. — Yes. So, my name is Frod. Uh, I am I'm part of the Norwegian Sovereign Wealth Fund. We are a $2 trillion sovereign wealth fund. uh and uh 70% uh equities, 30% uh fixed income and bonds. And we are only 700 employees worldwide. So we have um uh a single owner, the Minister of Finance who represents the Norwegian people and I'm part of uh the team here in North America that manages the bulk of the equity assets. — Thank you, Fro. F and I were chatting earlier and I think the stat I think has to be true that per employee no organization manages more money on earth than MBIM. So it's an impressive stat indeed. And for those in the audience we'll have a chance for you to ask questions as well after I ask a couple to the panelists. So keep that in mind. Uh we're really looking forward to hearing from you as well. But let's start the AI investment thesis. Naturally, your big organizations, complicated firms, at what point did this decision to go allin and do this at like a firmwide level take place and who drove that? And for this, we'll start with road. I'm curious to hear from you. — Yeah, sure. So, I mean, our AI strategy is really to be a leading user of AI in investment management. Uh, and that just means embedding AI into everything I do, I think, in a in a responsible way. Um so that means uh creating efficiency gains, reducing costs, uh enhancing returns and really improving risk management. So I think the what's

changed is that we have a coordinated strategy with really dedicated organizational support um and a dedicated AI team that uh helps out in the organization and this is really driven from the top right. So it's uh we have this AI maniac at the top of the fund uh Nikolai Tongen and uh he is just all in on uh on the change that we see in society right so his leadership his enthusiasm about AI is really setting the scene internally in the fund and driving the cultural adaptation that we see I think that's just absolutely crucial — thank you fro and not surprisingly this kind of top down buyin is critical if you're trying to drive adoption and really meaningful change across the Oregon. I think the frozen Nikolai has done an amazing job of that. Lloyd, how about you as it relates to your firm both internally and with your portfolio companies? — Yeah, so I mean I think that the top down buyin really resonates for us as well. Um so although we are investors, a lot of our leadership team are sort of technologists um and engineers. Um, so very early on I think we sort of recognized that this was a platform shift. Um, a lot of HG's success has been predicated based on helping our portfolio companies navigate the sort of on-prem to SAS platform shift. Um, we've also been investing in AI for a while. So we have a central data team. Uh, we actually had access to GPT3 back in 2021. Um so we we've been experimenting with this for a while and I guess when um you know the early LMS were sort of launched around 2022 2023 we realized this was a pretty major opportunity. Um I think for us though you know people find conviction at their own pace and what really helped was we ran an event in Silicon Valley so we worked with the anthropic team. We actually had Mike Creger the CPO come to speak at that event and we've fully immersed our execs in this AI topic and got people really hands-on. So for us that was a real catalyst. Um, and I guess over the last two years, the thing that's really surprised us is just how quickly this has moved. So, we started probably thinking about, oh no, we can get 10 to 20% productivity gains through integrating AI into our existing processes. What we're now really focused on is fully transforming our portfolio companies with AI. And we think that there's going to be a huge amount of value uh upside for us and for our portfolio companies through doing that. So, that's really where the focus is now. Um, and we've kind of replicated that same approach. So, you know, we got that top down buy in centrally for HG. We're now making sure that we uh create that same catalyst and that same spark for our portfolio companies and we're finding that sort of now permeating pretty nicely through the HG portfolio to drive that transformational change. — Fantastic. Thank you, Lloyd. And for the folks in the audience, I can imagine this is something on your mind and for those dialing in. All too often we hear, you know, should I build this internally or should I buy something out of the box with APIs now, it's increasingly easy to build really bespoke AI powered solutions. And at the same time, there's a lot of great solutions out of the box. Nicklin just walked through one with you today. So with that said, we'll start with you, Don. How do you think about this notion of build versus buy and basically deciding which path to go? Do you go both? Walk us through that thought process. — Yeah. No, I think it's a great question. I think when we first started this generative AI wave, a lot of organizations, especially financial services companies uh went through a decision like should we leverage uh APIs and build our own custom rappers or should we partner with a company like Anthropic and leverage like these enterprise solutions? There's obviously trade-offs with those. I think um for us we were actually very intentional that you know continuing to it innovate on the UI layer and all these different sort of capabilities like canvas uh and things of that nature connectors etc was going to be really challenging right I think uh companies like anthropic are just moving at such incredible velocity with respect to innovation uh we wanted to ensure that our internal developers and engineers were focused on our most you know proprietary and specific uh organizational solutions so For us, it made a lot of sense to more so partner versus try to build our own custom ones and let's leave like the custom integrations and solutions for those things that are really important for us competitively. — Thanks, Donna. And Michael, how do you guys think about that at Daw? — So, for those in the hedge fund industry at least, I think we're pretty well known for having a big focus on building things. uh we hire amazing technologists and we give them the opportunity to build great things. And so the instinct when something comes along for us is to build the best version of it ourselves. But resources aren't infinite and with great people comes big opportunity cost. And so we always think hard about build versus buy. What I think is different this time and I think Don alluded to it is the speed at which the technology is changing and

with that speed comes a different balance between uh build and buy. uh to be able to build and deploy at that speed and scale that some of what's being offered in the commercial market can provide uh is often somewhere between difficult and impossible internally. So that shifts our calculus. Uh we're still doing both and we're being extraordinarily deliberate in h where we choose to build and buy. — Fantastic. And yeah, we hear that So often as well when they'll develop a partner will develop an internal AI chat and they'll share a feature that they're particularly proud of and by the time we meet with them you're almost you know afraid to let them know it's like that's already a thing of the past and look at this and this that's already been developed. So I could couldn't agree more. Recognizing you're all big firms and leaders in your respective spaces. Another huge challenge we hear about is this change management and driving this kind of transformation at scale. It's one thing that this technology can do cool stuff. How do you actually get people using it? So when you think of like transformation of the size that you're operating at? And Don, I'll start with you. Obviously got a lot of employees. — How do you think about like practically driving adoption with technology like this when you've got so many different departments employees to think about? Yeah, I know for us our AI strategy has been pretty everything everywhere all at once. So we actually um treat our AI strategy as a portfolio of initiatives. So I think we started with like a lot of folks on targeted use cases specific to workflows or certain business domains. We had measurable value in production and that was great. We certainly realized that um we wanted to raise the floor and the ceiling of our aspirations by really democratizing and empower all of our employees with AI tools. And so there's multiple tools that we're really um leveraging to do that. And then we're actually also focusing on reinvention. So like business leaders at the top are really looking at each of their respective domains to understand how AI, agentic AI and these new paradigms can actually help us re-anchor our aspirations of how we can transform and the velocity for us to do it. And so that, I mean, again, very pervasive, tons of hands-on um uh training and uh hackathons, all these different things that you actually have to do like get in the trenches to really help people understand what these possibilities are. There's mindset shifts that are related to that I think maybe we'll touch on later in the discussion. But it really has been kind of a fullcourt press that's complemented by both a top-down commitment from our CEO and his executive leaders, but again this bottoms up sort of empowerment that uh goes alongside that. — Thank you, Don. And Lloyd, you have a bit of a different challenge and you're working with the employees internally, but also of course the portfolios that you support. So how do you guys think about scaling this and driving that transformation at the portfolio level realizing that's a lot of different companies you got to help train and educate? — Yeah. Yeah, for sure. So, I mean, I think a lot of what you said resonates, Don. So, we have 50 companies, about 120,000 FTE across those companies. All of our companies look a little bit similar and mostly, you know, verticalized software platforms. Um, so what that's enabled us to do is really sort of bring transformation at scale to that group. So, that starts with a central effort from HG. We have 20 AI specialists internally supported by about 150 or so kind of contractors and close partners. Um, and we've been really driving that central change um across the portfolio. Um it's also allowed us to experiment with lots of different tools, lots of different initiatives and really see what works quickly across that sort of laboratory and you know understand what the best processes are. Um but I think the yeah the top down point has been absolutely critical. So we've been kind of systematically uh working with each of the leaders of our portfolio getting them engaged getting them hands-on and then there's also um the piece I'd add is that we we're really trying to be disruptive in the way that we think about this. So you know rather than retrofitting existing processes and adding some AI we've re-engineered a lot of our functions. So if I take software engineering for example um on average across the portfolio we're now measuring about 30% productivity gains in software engineering. We have some companies who have taken their development squads down from nine people to two people. Um so getting really material leverage. We have one company that's deployed a thousand instances of an Aentic software engineer. So they've kind of added 50% productive capacity to their engineering team. Um and that's the sort of reinvention that we're now sort of driving out across the portfolio. Um I should say there what we're not doing is you know taking out any cost. What we're doing is then rein uh reinvesting in um productivity um basically clearing down tech debt and also building AI products. And that's the other key focus for us where we're seeing you know a huge amount of transformational value through embedding um the anthropic API but other solutions as well into our core uh software solutions. And that's adding material growth to some of our businesses already.

already. — Thank you, Lloyd. And on the topic of change management, this also comes up a lot with our partners. How do you think about, hey, this technology, there's a lot about that scares me and my employees might interpret this as something that could do their work. and how do you tow the line there in empowering them to use it while also recognizing that there's some fear on is this going to kind of do my job for me. So we we think about like the cultural implications at the firm and how do you make this a part of the culture and so fro this one's for you because I think mim's done a masterful job navigating this how do you and I guess starting from Nikolai but as you as a senior portfolio manager and your teams how do you see this fitting into the culture of the firm and becoming just a core part of how you guys operate? Yeah, I mean starting with a top down perspective obviously very important right I mean that's uh really driven it and I think we had uh focus on tech for a long time we've been a very techdriven organization now we we've taken a journey from moving from uh our infrastructure to the cloud uh having uh data on snowflake u integrating cloud for enterprise and being able to really chat with our data and that fits very well with our quantitative nature and our inclination to use data right so I think from a culture perspective is a very good fit and um uh I'd say in my team for instance uh a lot of us are really introverted right and uh we like to sit on and uh and code and do things and uh I think just having claude and having the projects that we can now build assistance and just share and collaborate on it is just uh really accelerating the cultural shift that we're seeing organizations I think that's super fascinating and I'll also add that being a sovereign we belt fund with a single owner it's just having AI into the organization is just such a massive enabler for us, right? Because we can do a lot of stuff with this technology that others just can't, right? Because we're not hampered with, you know, uh client and customer dynamics. So, we can we have a really open uh and lean organization with a collaborative culture and uh we're just 700. So, this is just fascinating. — Fantastic. Thank you, F. really I really appreciate that this idea of like demystifying what this technology is all about at the individual employee level is so critical because so quickly you can address those concerns or fears at you know a cultural level because very quickly they see oh this is just a great assistant if you will it's a great partner when I wake up and I start my role in helping me do my work more effectively so it's wonderful to hear that example — yeah I think JP I think it's a great point I think we were actually very intentional. I think very early on in this journey, folks would talk about, you know, AI is not going to take your job, but someone using AI will. And so we wanted to really, you know, address that head-on and ensure that as many employees as possible. And we were intentional about every employee having multiple AI solutions and then not just that, but getting like the training involved to help them feel even more empowered. Um, I thought that was really critical to your point to like get through some of the call it the FUD that might be a little bit of a headwind. — Awesome. Thank you, Don. — Yeah. And oftentimes I'll laugh because even when partners ask like, you know, how do I train? How do I build these curriculums to drive adoption? I'll pause and I'll be like, ask Claude. And jokes aside, whether it's Claude or another tool you're using, these tools are very good at that sort of task. We're like, here's my firm. And for those dialing in, same applies. Here's my size. Here's what we do. I want to build a simple training program for my employees. Maybe they've got three hours they'll spend over the next two weeks. How would you break it up? What would that training look like? I'm telling you, if you do that exercise when you go home, it'll be a beautiful, beautiful output. — Can I give you a double down on that? — Yes. — Uh if your staff say, "I don't know what to do. I don't know where to start. " Tell them to describe their job briefly. And ask the AI, how ought I start working with this? what are a few things I can try? It uh it actually works wonders. — I mean, I love that example because I'll even go so far as to say I've got two little kids, for instance, a four and a two-year-old. And if I'm looking for an activity to do on the weekend, I'm not very good at coming up with those. But I'll ask Claude, but I'll do one better. I'll be like, "Claude, I want you to optimize the following prompt. " And I'll be like, "I need to come up with activities for my kids. " And then Claude will rewrite it. I'll copy paste it. and then I get like the world's greatest to-dos. Um, so that's another good hack. Ask Claude to write your prompts for you. But deciding where to start is another thing we hear a lot about, right? So for folks in the audience, some of you are smaller firms, a lot of your massive firms, but okay, there's a lot. You've heard a lot from this group. Transformation across the or top down leadership. I'm going to start with Michael again here. Where do you

for the folks in the audience if we have to start somewhere like what's the use case? what is the first place they start if they're starting with a clean slate? What would you suggest? — So, it's incredibly tempting to look for a first place to start and my advice would be to resist that temptation. uh your colleagues, I don't know that they know uh more about how best to use the tools. They certainly don't before you give them the tools, but once they will discover uh things to do with those that you never imagined. So, we're a pretty decentralized place. We are a place that wants to make every one of our colleagues better using AI. So our core focus off the bat and what I would recommend to others is get really easy to use tools out there. Let people figure out what they're doing and then pay attention. Help accelerate the spread. in some cases centralized as you learn uh but not as uh not as your first step. I'll give one uh concrete piece of advice on that. I think some people will do that. They'll try something, it won't work the way they'd like it to, and they'll abandon it. What we found to be valuable is to encourage folks to come back, let's call it every six months. Uh the rate at which the technology and the products are changing is just foreign to most of your colleagues and the degree to which they can improve for a particular use case in 3 months, 6 months, 9 months uh can sometimes shock them and so pushing them to revisit is one play. We don't tend to be too pushy but that's one place where I think pushiness can pay off. So, I love that because I think to Michael's point, I want to hear from the rest of the panel. The notion of starting with one specific use case, oftentimes it can be like, oh, I'm going to build a bespoke analyst or risk mitigation agent. We'll always advise, similar to what Michael said, first things first, get this technology in the hands of your employees in a safe, responsible way. Obviously, you saw we have a solution. There's many solutions out there. And that should almost always be step one because so often, and I imagine this is true for everyone in the room. It's only when you start using this stuff. And honestly, that's true in your personal life. It's true at work that you really start to see what it's capable of. I'm going through a process today. I just bought a new house. I'm getting home insurance. I don't know when the last time you read a home insurance policy was. They're brutal and they're like 400 pages. and I get these three policies and I'm like even distilling the difference between coverage and umbrellas and all that and I work with a lot of insurance companies so I should know all of this. I just uploaded to Claude and I'd be like give me a breakdown of like what I'm looking at here and I'm sharing this as an example because those same aha moments and it gave me a wonderful answer. the fastest way to create them is to give your employees access to a tool like Claude and there's again there's other tools out there but just to get it in their hands in a safe and responsible way is is huge. With that said, um I want to go over to Lloyd as you think about advising this group on they've got limited resources, limited people. Where do they begin? Do you have any other thoughts on where they should start? And let's assume for the purpose of this they've already enabled a claude or an AI tool at the employee level. What's next? — Yeah. Um, so I was just going to add on to Michael's point. I fully agree with, you know, revisiting the technology every 3 to six months. The other point I'd add is really investing in your sort of prompt engineering or context engineering. I think there's a massive difference between a sort of average user who puts a kind of a lazy prompt into Claude versus someone who really thinks about that context. So I think finding those people who sort of become power users in your organization and just have a sort of natural intuition for that can really help to accelerate if you know put those people up as AI champions that can really drive a lot of change. Um I guess on other use cases once you've gone with broad enablement um I mean the way we think about that is looking at you know those kind of high friction tasks where people uh do a lot of mechanical processing that are quite repeatable that are using structured data. Those are often sort of perfect candidates in your organizations uh for something like claude. So for us um you know for the really critical decisions where it requires a lot of sort of careful thought and system two type thinking um we maybe use claude as more of a sounding board but for a lot of those sort of mechanical uh tasks that happen in all of our organizations it's kind of using claude plus other automation tools to really drive um yeah transformation of those processes is how we

think about it. — Awesome. Thank you Lloyd. One of my favorite questions to ask, so I'm obviously going to ask it here. Uh, for the purpose of the audience, all of you have a lot of experience. You've kind of gone through this. You've applied it in your organization. You've seen a lot of what works and doesn't. I'm going to start with Frod. Um, knowing what you know now, if you could do it all over, what's one thing you would do differently? — I don't know about what I would do different really. I think in Mim we really nailed it in terms of uh getting it. I think it's just uh maybe just start even uh even before right and uh start right I mean that's what I would say you know this is something you just have to lean into and and take on so I think um move fast and make mistakes learn from them and move on is really the key takeaway and I think if anything is just uh be there — thank you for Don how about you — yeah I'll lean into the deep do more faster bit but um I think one thing we haven't touched on that I think that we focused on but looking back perhaps we could have focused on earlier was this notion of it's a true like mindset and culture shift. Um we talked a lot about democratization and like a tool roll out but certainly this is more than just like rolling out tools and giving people access. It's like giving everyone in America a treadmill and expecting like heart disease to be cured. Um what we were very intentional about is understanding that this is actually a mindset shift as well. So I think a lot of people look at like a claude or a chatbt they see a text box and like oh this is like Google I should just like put in one question. I'm going to get back an answer and then I'm kind of done. I think the reality is the this technology is really designed to be more of like a human. This truly is a companion and a co-pilot, however you might describe it. And so having a bit of a mindset shift and actually helping folks understand that, I think is really critical. We've been really intentional about partnering with um folks from the outside to help our executives especially on that journey. Um, one analogy we've used is, you know, if you're going to Costa Rica with your family, you could go to Google and ask for some recommendations or you could just prompt Claude or Chat PT and say, "Hey, you are the Costa Rica prime uh, you know, head of tourism. I'm going to Costa Rica with my family of four. My kids are 10 and 12. One likes surfing. The other one likes, you know, wildlife. Can you come up with a customized itinerary? " And so that's a very different sort of paradigm between those two tools. And frankly, for most folks that, you know, first start to engage, that's kind of where they start. But we think it's really important to kind of help people on their journey. And if I could have, if we could go back, we would maybe even started that even sooner. So, — thank you, Don. How about you, Lloyd? — Yeah, I mean, I think I'd echo Frodo. We are running at this very fast at HD, and I'd like to think we're sort of ahead of the curve. We were building our first AI products back in 2023, but if I could say one thing, it would be run even faster towards this. um just the you know the exponential rate of improvement of models and the ecosystem that's developing around that is uh frankly unprecedented so yeah I would have said um you know run even harder even faster Michael — so we're pretty relentlessly datadriven and one thing we do is we look at the uh usage intensity across the firm and what you see is a lognormal curve something like this and the theoretical underpinning of that is it describes processes that are roughly the more you use it, it. So they're multiplicative. And if you combine that with this bottomup view, what you want to do is just encourage people to take more bites of the apple. So here are two things that I do to help make that happen. Uh, one, uh, at a lot of my team meetings, I just make people do a lightning round of some strange AI thing you tried in the last week. It worked, it didn't work, but you know, 30 seconds each share with the team. Um, and two, and this plays off something you said, the route to comfort and learning about how to use these at work doesn't have to only run through work. And often people are more comfortable taking risk in their personal life. Uh it's not as potentially embarrassing at least in their minds. And so encouraging people to experiment there I think can pay off uh as much or more than encouraging them to take those first jumps in the workplace. — Yeah. No, I could not agree more and that's a huge I think benefit to driving adoption for everyone here today and dialed in. If I had to recap what we've talked about today is enable. So give your employees access to one of these tools. Train. You don't have to be a master trainer or you know a buy a million-dollar training program. You can

work with some great partners of which we've t touched on a few of them today and you can use a claude to help you build a training program to give them baseline training. I'm not talking about 100 hours here. Baseline how to prompt how to interact with these tools measure. There's to Michael's point now. I think he's doing it in a really great natural way. Let's do a speed round at Enthropic. You can imagine it's like we are monitoring in a way are you using these tools to do a better job and like you're actually tracking that because whether or not you use them it's making sure you hold your employees accountable for building that muscle and getting used to using it. I'll share a p personal anecdote in closing and Michael touched on this. Sometimes folks are more comfortable using this technology in their personal lives to see what's capable of them work. I have a 2-year-old son who has a lot of medical concerns and the in learning to navigate what his challenges were and what we could do for him. The hospital sent over,00 pages of notes from the NICU from our neurologists and it's very hard to make sense of all of it. So even when you talk to different doctors, it's they all give you different answers. And I took all of this and I uploaded to Claude. I go, "Help me understand what's happening here. " And I took that timeline and I'm not suggesting Claude's a doctor by any means. Um, I just took that timeline, 1100 pages by the way, PDFs, and I took it back to our doctor. I'm like, "Does this seem right? Help me understand what happened here and what I can do about it. " And literally the doctor's jaw dropped. He's like, "How did you come up with this? " And I'm not sharing that story because look how great Claude is. I'm sharing that story because in moments like that when it has such a meaningful impact on your life personally, you can see my inclination irrespective of working in a company like Anthropic to try using this technology in work goes up dramatically. So as a takeaway, if you don't have one of these apps, these AI apps in your phone today, I'm going to say step one, use it. Set a goal for yourself. Use it. I don't care if you're taking a photo of a tree or you're analyzing life insurance statements or policies. Use it a couple times a day. Build the muscle and in turn you'll be better equipped to do that with your own employees. So, I want to say a big thank you to our panel. I really, really appreciate it. Thank you for everyone who dialed in and joining us today. And for those seated, stay put because we'll have a chance for questions. But a big