Advancing Claude for Financial Services

0:00 Good morning everyone. I'm Eleanor and I lead our industries team here at Enthropic. >> Hey, I'm Nick and I lead product for Claude for Financial Services. >> And welcome to the launch webinar for the next milestone of Claude for Financial Services. A few quick housekeeping items before we start. If a friend or a colleague wasn't able to join today, don't worry, we'll be sending out a recording within 24 hours of today. We likely won't have time for Q&A at the end, but please submit questions throughout and we see many of you have already started submitting questions and that's amazing. There's a widget at the right hand of your screen. Our team will be monitoring and answering questions throughout and anything we don't get to, Nick or I or the teams we work with, we'll be sure to follow up. Lastly, we love feedback and so if you have a chance either during or at the end, there's a survey widget at the bottom of your screen. We would really appreciate it if you could take a minute and submit a survey. Now, let's set the stage for why we're here today. At Enthropit, we've built our reputation on depth over breath. The same discipline that has made Claude bestin-class for coding. Our deep domain understanding, rigorous evaluations, and a relentless focus on realworld performance is exactly what we're bringing to financial services.

1:15 This is not just another vertical for us. This is infrastructure for the entire world. The allocation of capital shapes how companies grow, which innovations get funded, which communities thrive. Getting AI right for financial services means getting AI right for the global economy. This is why we're approaching this with the same intensity, precision, and partnership that has defined our work with developers and we hope will now define our work for financial services. Since our initial launch three months ago, finance professionals have been making cloud an essential part of their daily workflows. Investment banking teams at city are using cloud to accelerate deal execution. Asset managers like noises are integrating cloud directly into their investment analysis. Insurance leaders like AIG and Alons are transforming underwriting and claims processing. And corporate finance teams across industries are leaning on claude for acquisitions, planning, and reporting.

2:12 Let's start with banking where both speed and accuracy are critical factors for success. Global banking leaders like City and RBC as Nick mentioned are driving huge initiatives around the world of becoming AI native banks. They're starting to think about reinventing processes that have existed for decades if not hundreds of years like know your customer and controller reconciliation that are built on top of human and system processes that haven't been modernized in generations. With cloud for financial services, AI can handle the systematic work so that senior bankers and strategists can focus on what they do best, decisionmaking and deal execution.

2:52 Asset management faces similar problems and we see similar patterns. Asset managers like Bridgewwater and Noris need continuous monitoring across dozens and hundreds of companies. And while AI can maintain this detailed coverage at scale, portfolio managers can spend less time on the continuous and very tedious information gathering and more time on making fast, accurate, and good investment decisions that can drive returns. >> Now, in insurance, getting to an accurate quote faster than your competitors is really important.

3:25 Companies like AIG and Amins use Claw to generate policies and process claims in days instead of weeks. without sacrificing the thorough risk assessment you need to really stay competitive. And for fintech and payments, the challenge is maintaining oversight while operating at incredible speeds. AI enables continuous analysis that wasn't possible before. Monitoring transactions in real time, tracking compliance across multiple industries, all at the speed money actually moves. Today >> there's a similar set of problems across banking, asset management, insurance, and fintech. Front office teams need speed to insight to drive revenue.

4:09 Middle office teams need a consistent and reliable approach for managing risk while these front office teams are innovating. And the back office needs modernized operations to continuously reduce cost. All of these have to happen at the same time throughout this ecosystem we're in right now of rapid innovation. Enabling this level of transformation at this scale all of you on this webinar work at requires domain expertise and a deep investment from companies like Enthropic in both the research what goes into building the underlying models like Sonnet and Opus and Haiku and product the application layer that sits on top of those models.

4:47 We need frontier intelligence that actually understands the practical needs of the financial industry that you all live and breathe in daily. And we need to build that intelligence into the foundation and the product so that we can deliver the lab level of value that you need every day. >> Now finance presents a different challenge for anthropic. There are many developers internally who can test our coding models and products every single day. Now that is a big reason why Claude is state-of-the-art for code. But not all of us are investors and underwriters. And we're not just hiring teams of junior bankers to train our models. Instead, we have to spend much more time with our enterprise customers to understand the real problems they're hoping to solve. KYC, underwriting, investment analysis, train Claude on those problems together with our research teams, and build products our users can easily adopt to test Claw's intelligence. Now, we take the feedback directly from our users to our research team so that Claude can be really great at the real world tasks you all really care about.

5:57 When we meet with your teams, we're usually talking about the products. What problems do they solve? What problems do you have? What user experience are you looking for? What's your implementation strategy? How are you thinking about rolling things out for the next 3 6 9 months? What's our security posture? But if we zoom out for a second, let's take a moment and think about what it means for those products that we talk about with your teams every day when we bring domain expertise and the enterprise expertise of how you all work directly into the models that sit beneath and power these products.

6:29 >> Now the foundation is intelligence. There's a lot of questions around whether the models are plateauing. Now, we think we're just at the beginning of building domain specific intelligence. Now, this is a benchmark you all have seen before. Published by our good friends at val.ai, the finance agent benchmark is well representative of many realworld entry-level finance analysis tasks. Now, within its general training corpus, Claude already had a vast amount of finance domain knowledge. In August, we saw Opus 41 topping the leader chart at 50%.

7:10 Since then, for the first time, we've specifically invested in finance vertical training as a part of Claude Sonnet 4.5 launched in September. Within just one month, as you all can see, Sonnet outperformed even opus by five full percentage points. Now, these benchmarks, what do they mean? They directly translate to can claude spot the information faster than your peers. Can it identify the errors in your discounted cash flow model? Can it stress test your assumptions?

7:43 And we're just getting started. We spent the last few quarters I'm so sorry. Sitting with you, sitting with our customers, and understanding your problems, your needs, hearing your feedback, in many instances a lot of feedback, and supporting your implementations. But one major constant theme has emerged as you think about this new way of work that we're all entering into as part of this AI transformation. You want Claude to meet you and your teams where you work.

8:12 >> That is why we have three big product bets for 2025 and beyond. Expanding the surfaces where Claude works. augmenting the data sources Claude has access to and increasing the depth of task Claude can own end to end with skills. >> Let's start with services. >> if we were all able to see you now, I'd ask for a show of hands on who works every day in Microsoft Excel. And I think it's a reasonably safe assumption that all of your hands would be raised.

8:42 I know our finance teams would be. >> Now, as Eleanor mentioned, we want to meet our users where they work. You all are making Bordex in PowerPoint. You're building models in Excel and we heard you. Visibility and transparency is really the only way for you to trust Claude for these core analyst workflows. We're excited to announce that as of today, Claw for Excel is available in research preview. You can work with Claw directly in Excel where you can analyze, modify, and create new sheets and workbooks while seeing all of Claw's changes in real time. Now, we've been really impressed by what Claw can do, but try it out for yourself. Put it to your comps, your models, and data packs, and let us know what you think.

9:32 Second, connectors and data sources. Now, we want Claw to be your unified intelligence layer. Our incredibly rich ecosystem is the only reason that allows our platform to be flexible and adaptable to your specific needs. Now, Eleanor and I have been so excited about the momentum from a July launch where we partnered with seven of the largest industry players to deliver native connectors to claude. This time we're adding eight more to the list, including real-time market data, transcripts, and news from LSEG, ERA, Moody's, and MT Newswires, expert network intelligence from ThirdBridge, private market, steel rooms, and portfolio management with Chronograph and Ignite, and even core enterprise knowledge systems with Office 365. And we'll continue to add to this list with ecosystem partners over the next few months and years. But our team can also work with you to set up custom connectors that are essential to your daily workflows. Whether that's you, your team, your department, and internal data sources that when you combine them with these real world external data sources in real time can power infinitely more powerful workflows.

10:48 >> And finally, our agent skills library. We launched skills a few weeks ago. Now, skills are a really easy way for anyone, not just developers, to program Claude to be really great at these core workflows you all really care about. They're essentially bundles of instructions clock can easily follow to do comprehensive research like earnings deep dives and initiating coverage, complex analysis like discounted cash flows and leverage bio models and creating client ready deliverables like dozers and company profiles.

11:26 Similar to the last slide where we have built these integrations for you and these connectors out of the box, we are providing these stills as part of cloud for financial services. But stills can cover an entire spectrum from work that's just for you again to your team to the department to the organization and across the financial industry as a whole. I for example send out a weekly email on Sunday nights that recaps my team's progress from the previous week and focus areas for this week. I have a still write that now. And while the team thinks I'm spending hours of my Sunday dedicated to updating them, I'm in reality watching TV while Claude does this work for me.

12:03 We'd love to hear from you. Take a moment and think about the task that always sits at the bottom of your to-do list, no matter how urgent. The thing that makes you want to go for a walk when you know you have to do it in order to get to the real work you want to focus on, and what skills you want Claude to be excellent at. Please submit them in the chat so that we can take them back to Nick and the entire product team and hopefully prioritize and delight you by introducing these skills in the future.

12:30 We are going to move on but definitely take the second right now because we really really want to hear from you and >> I think that is a perfect example of how we use customer insight to feed directly back into our road maps. Now with these new features, Claw can go much deeper into your direct workflows. And that is across hedge fund, private equity, and investment banking. And we would love to show you what this actually looks like in action. Starting with hedge funds.

12:59 Let me show you how Cloud helps credit analysts move from market signal to investment decision in minutes, not hours. We're following Yuri, a credit analyst at a multistrategy fund, Steady Capital. It's 10:00 a.m. and he notices the Walmart 2030 bonds moving tighter. Portfolio review is at 2 p.m. Yuri wants to know, is this a trading opportunity or fundamental shift? He starts simple. Show me Walmart's bond curve and compare it to Costco and Target. Claw connects to LSG. Within seconds, an interactive dashboard is ready. He receives full Walmart curve with live data. The 2030 is tight, but there is a steep jump out to 2037 and 2038. The market wants compensation for a longer duration. Claude automatically pulls comps for Costco and Target, and Walmart trades tighter than both of them. In under a minute, Yuri sees that Walmart is priced as best-in-class. Now, do the fundamentals support it? Yuri asked Claude, "Dig deeper on Walmart's credit quality, analyze last three earnings calls, check third bridge experts, and give me a write up."

14:04 Claude pose error earnings transcripts Q4, Q1, Q2. While searching Third Bridge for former Walmart execs, everything happens in parallel. Within minutes, Claude has a complete credit analysis ready. It found patterns across three quarters, consistent sales growth, e-commerce is surging, advertising is up nearly 50% and new high margin businesses are scaling. Every claim is verifiable with links that go directly to Arrow's platform. An exact transcript Claude analyzed. Claude found former execs who actually built these systems.

14:35 One confirms the e-commerce profitability story. Another explains their competitive advantage on tariffs. Supplier relationships and domestic sourcing give them structural protection. Claude synthesizes everything. This is a trend, not a blip. 3/4 of strong execution. Experts confirming it's real with every data point linked source. It's 1:45 p.m. Yuri walks into the portfolio review with his dashboard and comprehensive deep dive. The story is clear. The market is pricing Walmart as best-in-class and fundamentals back it up. Clause view.

15:06 Short bonds are priced to perfection and longer bonds offer better value if this transformation continues. 30 minutes, three platforms, one story with full transparency. That's the difference between data and conviction. Now, thinking back to my days in investment banking and private equity, I can just imagine how much more time I would have had to focus on what really matters, right? Forming my judgment on the company and building relationships.

15:42 And we hope you're excited to join us at the frontier. Our new connectors are available immediately on a directory. Both clot for Excel and our finance agent skills are now available in research preview. Join the wait list with the QR codes on the screen. >> And now, no launch webinar is complete without a distinguished panel of industry guests, and ours is no exception. And so, I am excited to welcome four industry experts who are joining us today.

16:11 Christina from Moody's, Andre from RBC, Andrew from Carlile, and Josh from Anwins. Great. Why don't we go ahead and get started? why would love if you guys could introduce yourselves until I'll share a little bit about your background and then we will dive into questions. Andre, can we start with you? >> Sure. Thank you so much. hi everyone. My name is Andre. I lead the AI and digital team here at RBC capital markets. I've been with RBC for about 10 years now. always in AI, always in capital markets. happy to be here.

16:54 >> Amazing. Andrew, >> great. Thanks for having me. Very, excited to participate in this conversation. I'm Andrew Levventhal. I am one of the leaders of a number of applied AI initiatives across the Carile group. Been with Carile for about two years, always focused on AI. and principal areas of focus are essentially building tools to support investors and deliver output for our LPS. >> Amazing. All right. And Josh, >> hey, thanks for having me. I'm Josh. I'm CIO for Amlins. I've been with Amwins for 11 years. I actually come from a banking background and unlike the others on the panel, I did not start doing AI. So, it's been been a journey.

17:40 Well, then let's actually start there. I think on the audience we have folks varying levels of the AIDO adoption maturity curve and I'd love to start with the beginning and build our way to where we are now over the course of this panel. Could each of you share what specific problem or opportunity led you to begin exploring Claude or AI in general inside of your organization? And Andrew, why don't we start with you? >> Sure. our our journey at Carile probably begins six years ago five six years ago and much more in the context of traditional data science so our team has done quite a lot of work across large data sets to build things like machine learning models etc.

18:24 and as AI became more prevalent and more advanced, I think we saw a lot of opportunities to take not just predictive modeling in a more standard sense, but start to layer intelligence on top of that, and really leverage these tools to help drive insights for our teams. so that's probably one main one. There's a very long list of other things that we're interested in using AI on. >> Amazing. Andrew, what about you? We started in 2023 we we started with capital markets research. RBC has a long history of investing in AI. but we started in capital markets generi specifically in our research division and the reason why we started in research is because if you think about you know going back also to the demo earlier if you think about what a research analyst does is they they have to read a lot and they have to know the industry deep and they need to know the the the stocks they cover really deeply.

19:23 so that's where we thought that Genovi can kind of help extract some of that key information. So our research analysts can spend more time talking to clients focus on more higher value tasks and sort of leave the the the grunt work to to to the AI. >> In unrelated news, Andrew will be the voice over at our next demo for for our next launch. >> Great. So perfectly perfectly put. Josh, can you share where you started at Amwins when you first saw the opportunity to bring in claude and AI?

19:56 >> Sure. I think with like a lot of firms we started very much looking at operational things. I think like a lot of folks you look at the things that are already measured that are quantifiable your your widgets in widgets out you know what the price model is the whole nine yards and so it's a really great place to start because it's very easy to talk about what your ROI is when you automate those operational tasks. That was a really good start. We learned a lot from it. we were able to use that we like to refer to it as pattern recognition figuring out what works what doesn't how to take that forward and we've slowly taken that into other areas of the business whether it be in our underwriting groups working on how do we take some of that information and that intelligence that they have and leveraging that enhancing it through AI or automating a lot of the work that our brokers do so it's they're able to get things done faster and more effectively >> now you know agents are really top of mind for everybody, right? With skills, we're also hoping for claw to more autonomously complete bigger pieces of complex tasks. Now, I think all of you are thinking about agents today. I'm very curious, how are you all building the foundation to enable more agentic workflows and how do you make sure that humans still have the proper governance and controls in place? Why don't we start with Josh? Yeah, this is one we we've been thinking about a lot over the last year easily. It's one of those areas where this is where the value really gets very manifest very quickly when you think about a lot of the workflows people have been doing here before it's been very single task or how do I do this one task using AI as opposed to how do I just automate the entire thing? How do I make think about the entire process differently? And so for a lot of us, that's caused us to realize that there are lots of little areas of friction that are in the way of making that real. We tend to talk about them as agentic infrastructure, things that you never thought you needed an API for before or that you never realized, oh, this isn't actually conducive to an automated play where there's not a human involved. And so there's a lot of work that goes into that, you know, that infrastructure to make that real and make that viable. In terms of the human in the loop, that's one of those things that we're we're pretty passionate about. I think one of the the failings we tend to see when people start looking at these u these workflows and and these different approaches for the first time is they think about it in terms of it's got to be perfect. It's it's got to 100% be accurate every time. It's computer, right? Like it >> people forgive humans and they don't forgive AI.

22:42 >> Exactly. And and the problem is you are probably already measuring what your error rate is today. That's your target. That's your baseline target. Can I make that? And can I if I can do better than that, great. But let's not create unrealistic expectations for what we're trying to accomplish. Perfect is great, but it's not what you have today probably. >> Yeah, absolutely. You know, we think about evals a lot internally and eval sound like these mystical concepts that nobody understands. But in our minds, they're really simple, right? What are the problems you're hoping to solve and what does good look like for those problems? And these should be problems you're thinking about, you've been thinking about for a long time already.

23:26 So to your point, Josh, there should already be metrics you're tracking that really gauge success against those eval. So very aligned with that. Andrew, what about you? >> This is an area that you know, we we've spent a lot of time thinking and and still I I would say like you know, no one has the answers to to to this problem. we are working really closely with our technology partners to to set up that governance that that that Jos was just saying. That's extremely important. We you know, we're strong believers of still a human in the loop.

24:01 these agents they can do and they are starting to to take more complicated tasks and complete more complicated tasks but ultimately as they do that the risk increases. So we're taking a really strong approach from a risk and governance perspective like we want to make sure that every like everything they they they can do and they are doing is monitored ultimately by there's there's a direct responsible human at that chain.

24:34 >> Yeah. You know I think we think about our job in BM auto provider to provide you all our customers and the users with the proper auditability as well. You know, cloud is really great at exposing all of its thinking. And for all of our MCP data connectors, we also want to make sure that every piece of data that's retrieved from an underlying system has citations and audits associated with them so that the users can exactly verify where the information is coming from. So, you know, trust and safety is something that is really top of mind for us as an organization. And we really want to work with enterprise customers like yourselves to make cloud an even more trusted model partner. Now Andrew, what about you?

25:18 >> Yeah, so we've done quite a lot of work with agents and agentic patterns across a variety of workflows within our organization. I think we're still in the early days of being able to deploy these sort of in production, so to speak, which one speaks to the continued need for human in the loop, but two, I think there's two like really foundational challenges for any agentic pattern in terms of delivering on we'll just call it broadly like process automation. and those those two things are one you need deterministic output meaning either it always knows that 2 plus 2 equals 4 in the case of of our world in finance or at bare minimum if I put a set of inputs in I'm getting the same set of outputs consistently and and obviously that ties back to your comments about eval the second piece that I think is really critical and this is the much more challenging piece I think that that we've seen across our organization across our our broad portfolio of companies is most processes are comprised of a whole set of business logic and business rules and many organizations despite their best intentions don't deeply document those.

26:28 and so what you get is this sort of rat's nest of logic and rules that mostly lives in people's heads. And so the question becomes how do you start to take that tease it out of people's heads and present it in a way that a model or an agent can actually learn from and make use of. And so we've been spending a lot of energy trying to figure out how do we lay the foundation for ensuring that we have good clean data good documentation for the particular workflows that we're we're tackling. I think the other key point to raise with respect to human in the loop is you know we anticipate that humans for certainly for the foreseeable future will be making ultimate decisions with respect to the outputs of these tools. there's two pieces to that I think. The first is you know that's what our LPs hire us to do. they want to leverage our reasoning, judgment, creativity and experience. and the second which is a much longer conversation is you know what's going to be the regulatory posture towards fiduciary duty with respect to these tools going forward.

27:30 No, Andrew, I think one thing you said really stuck out to me, which is, you know, the model is only as good as the context it has access to. And that's a mantra that we think about internally a lot. We create a model context protocol so that the models have access to the proper context. But everything starts with having proper documentation and governance in terms of what exists in people's heads. So we think about this a lot in terms of how do we enable a much cleaner enterprise knowledge graph and this is where we spend a lot of time with our enterprise customers to fully document you know internal processes teams histories of projects you know decades of information and how do we process that in a way that is easy to access for the models and when it comes to runtime actually be able to retrieve those with accuracy as well. So, I think that's an area that we're really excited about, but we still have a lot of work to do in Christina. Do you want to give your audio a try and we can we can see if we can have you?

28:30 >> Yes. Can you hear me now? >> Cena, >> this to prove to everyone in the audience that we're doing this live >> and that technology hasn't solved everything yet. Yes, I can try to well call on financial services version three will so okay so u I'm Christina Pretti I had a a business unit in Moody's called digital continent innovation basically it's a distribution of all the content that Moody's owns or produces through different channels including Moody's.com APIs data feeds and of course genai and in that regard you know this is is very for today and what we're calling out smart APIs and we've also been leading the charge on aic solutions. I think you had a question around you know when did we start actually with machine learning we had started years ago because we are a very quantitative company you know producing a lot of models so we had been working with machine learning for a long time we had actually done some tests with natural with NLP and LLG to see where it was I remember back I found the innovation unit here back in 2016 and we kind of realized it it was not really there and that's why we were like so wowed when you know November 30, 2022, you know, we announced GPT and then, you know, we've gone in this incredible journey as as as an industry of what has happened. So, we quickly jumped in that bandwagon and we're doing things with agents both internally to make sure that we can you know drive efficiencies and we can empower our people using agentic solutions. We also have a big commercial push in terms of both putting our data out there through FCPS and smart APIs but also supporting our customers in their genai journey by delivering genai solutions. So that's a little bit of of the the when and what we're doing and I'll take it from you if you want me to catch up in any of the other questions.

30:36 >> Yeah, Christina, I think that's amazing. You know, as I quickly alluded to in the presentation, one of the core pieces of claw for financial services is to be that unified intelligence layer across all of our customers data sources, right? And we've been really excited about the ability to bring together not only, you know, historicals, fundamentals, information, and credit ratings, but also real time data and analytics, which is a big theme for our launch today. Christina, I'm curious, how has real time intelligence changed decision-m for you and your clients and what do you think still remains challenging?

31:13 >> Yeah. so I I think this is a big game changer, right? And I I I always say when I when I do this type of of speaking engagements and this is the biggest change I think we're going to watch in our lifetime >> or at least for me given where where I am in my career, right? And I think this this basically it's it's a whole it's not only about the real time because it it puts it puts you know a lot of information our fingertips immediately but also as you said you know in the case for example of cloud financial services we're aggregating a lot of data. So it's it's it's it's both right it's the fact that you have real time data at your fingertips and it's not only one set of data but you can have you know a lot of relevant information for the topic you're working on at the same time. So I I would say that's advantage number one. I think that those are some of the early things we we saw.

32:00 I think there's there's also the the change in the work right the change in maybe a lot of the time for example when I we look at our clients that they were looking into to finding the information integrating that information making sense of that information that's all kind of automated or reap for automation from by by Genai right if you think a little bit about the challenges and I think we discussed a little bit about this during the prep I think this is a big change right and and it's a big change for many of things. It's a big change in terms of of all of us as workers and what does the future of work means and entails for us. but it's also a big change in terms of managing risks. For example, you know, when when we think about we're putting a lot of of we're putting a lot of trust in the solutions and that's why it's important that we develop best-in-class solutions, right? like like block financial services is an example where we know that you know there's the right there's the right compliance and governance around it because at the end we're asking you know all the actors in the financial services industry and I think some of the speakers were talking about it before to trust something that is not their own work right and I think that's that some of the biggest changes how do we how do we put the right controls to mitigate those risk how do we get comfortable us with you know and all the participants in financial services that those risks are okay and can be handled. And then the third and probably one of the bigger and toughest ones is how does it mean for for my you know the future of work and how do we how do we manage the change of that in the in the workforce?

33:39 I do think that's a really important point, Christina. I mean, we when we talk to customers and even when we think ourselves about this transformation, I think this one is so huge for so many reasons, but it isn't just professional. It's also personal and it's impacting every part of everyone's lives. The the question I did the most is what should my kids study in college? And and how should we start thinking about the future of work and the workforce? And so I want to take us in a slightly different direction and and come back to the the future of work because I'm very curious for how you're all thinking about this for your teams, but to first talk about what does adoption look like inside of your your firms and your organizations because this isn't about adopting a new tool or adopting a new technology. It's exactly what Christina was saying. This is adopting an entirely new way of work and in many instances changing the entire DNA of your organization and everything that had previously propelled you all to the heights of your careers of your firm's success now looks completely different and you have to rethink almost from the ground up what this looks like.

34:45 And so it isn't about change management. It really is adoption at scale and and rebuilding and rethinking about these internal processes that touch so many different teams and touch your clients and your customers. And so I'd love to hear from you all on what cultural or organizational factors have been key to getting folks excited to moving beyond early adopters to starting to see broader adoption and how you're thinking about addressing some of the very real anxiety your teams might feel. Josh, it'd be great to start with you. Sure.

35:18 For us, education's been definitely one of the the the tip of the spear in terms of how we worked on culture. At the end of the day, a lot of this is is very new for folks. And there there are a number who've come in very eager to learn who are in that pioneer category who are just out there, you know, just discovering all the ways they can they can make things more efficient or save some time or or make things better and leveraging those people to share with their peers. But then making sure you're coming back on that education piece and helping fill in the gaps and not turning it into, hey, this is a this is a way of approaching work. this is a way of thinking about things as opposed to just a tool used for this one thing. I think one of the things that's made the biggest difference for us is finding those use cases that are directly applicable and valuable to people in their roles today. one of the things that we we say often is if folks don't find value in a tool that you roll out, they're not going to use it. Doesn't matter how much you promote it. doesn't know that doesn't matter how many marketing campaigns you have. They're they're just going to put it the side.

36:27 And so finding things that that drive real tangible value and showing people how to use those, that's what gets for us at any rate buy in and engagement. And that's where we've tried to put a lot of focus in in our education and our training efforts. >> I had a customer say to me the other day that with his team, he needs them to do and then feel and then think. And it's very hard to get people not to try to think upfront but to just do and that that paradigm in and of itself has been one of the most challenging things but where his focus has been.

37:01 >> You know we think about education not only in terms of the seminars that we can we can provide you all but also within the product itself. We have an entire ministry of education team internally at anthropic just focused on the former but you know we're thinking a lot more internally about how do we actually build the products to be useful and that understands what your workflows look like. I think you know software have the advantage of us being able to write code for those workflows right I think within the process of writing code we're injecting our domain expertise in terms of what best practices look like to interact with systems like you know Salesforce and Slack but for us AI is a lot harder right there's a lot less code to write because the beauty is an AI system so we want to really give claw the scaffolding with the tools it has access to these skills that it can flexibly deploy so that we start taking you So these important core workflows for our customers and breaking them down into smaller chunks that clock and start attack.

37:58 >> And now let's move on. Andrew would love to hear from you. it feels reductionist to call it change management. I want to acknowledge that. But how are you thinking about enabling your teams and driving this level of adoption and rethinking of how they come to work every day? >> we we have a two twopronged approach. we take, you know, we have a bottoms up and it's sort of a top down. Bottoms up meaning that the product just has to be good and the product has to be connected to the right data sources and it has to work well and if the if you don't have that then you you know whatever change and adoption you're trying to get like people are just not going to adopt it because the product is not good. So that's sort of like the the number one thing that that you have to have. Then the number two thing, you know, on the top down is is all the training. That's it's extremely important. Like you have to train people that don't know how to use a team. You will always end up having sort of the fast runners that know the tools, are using them on their personal lives, are experts in their business and are sort of becoming what we call internally AI champions.

39:01 And and so what we then did is we start identifying these individuals in different lines of businesses. we formed a group the AI champions group and we actually we meet with them every every Wednesday we come up with different ideas people share not only feedback but what what are they doing differently but what's very important in this in this construct is that then they go back to their teams and then they lead the AI adoption within their teams so it's not necessarily as like as as centralized it it's it's sort of federated in that way that where these AI champions are sort of teaching their teams, their their sort of neighborhoods or like in terms of physical location and and so on. and that has proven to be really really successful and probably the best thing that we've done and it's part obviously part of the education component >> and I think it shows that because this technology is so new and we're really just scratching the surface of the capabilities the best ideas could come from anywhere. they can come from these champions who are embedded in the closest to the business problems and the day-to-day execution. And while I like to think we'll all have the best ideas, I think our our customers have the best ideas, leaders and then folks doing the work day-to-day. So, I really think harnessing that creativity the way you're doing is is amazing.

40:22 >> And this is how we build products internally as well. You know, cloud code came out of a weekend project by Boris, the the engine lead on cloud code. So, we're really hoping to also foster an environment of creativity and letting all of these best ideas shine and grow internally as well. >> All right, let's do Andrew and then Christina. >> I think so a couple comments on this topic. The first is that where we started and and felt quite strongly with respect to our business is having a really clear and comprehensive compliance framework for how we were even going to approach this stuff before we started doing anything with it was really really critical for building trust across the organization especially driving sponsorship at the top layers of the orization. it really needed to have a lot of confidence the risk surface area was contained in a way that was manageable before we going to turn these things loose. I would say the second thing which I I really liked what what Josh said about the education piece. It's a really critical piece. We actually didn't start with that. we took a very specific view that where we wanted to start was with small pilots that we actually put this in people's hands and really didn't give them any instruction other than you know don't put certain types of data in because that's an issue from a compliance policy standpoint but beyond that it was really sort of like go figure out how to use it. and and obviously like we didn't do that across the broad organization.

41:50 We picked people that we knew were going to be testers and explorers and and be creative and thinking about how to use that. And I think we found some really interesting stuff that emerged from that and that's ultimately what for informed how we built our educational framework. and so you know we now have an AI university within Carile that we're pushing across the entire organization. So that's really driving those things. Then the last thing I would say is that although this stuff is very new and very different in many respects, I think the adoption curve looks quite similar to any other technology. The distinction being that the time frames are probably considerably compressed and so you know a lot of the playbooks that you would use to rolling out other technologies, some of which we heard mentioned here like champions groups and communities of practice and things like that are still super relevant. and so it's really the the combination of all those things together that I think have have helped us drive really high levels of adoption and usage across the organization.

42:50 >> I think the compression of those maturity cycles is such a good point when the models are getting so much better at such a tight timeline. You can take things that you would have normally taken 2 to 3 years to roll out and you have to think about what does this look like in two to three quarters and how can we all change our thinking to embrace that. Before we move on to our our last question, Christina, I'd love to hear from you because this is I know something Moody's has been thinking about for a while. What does that change management adoption, how you think about talent and talent development look like inside of Moody's today?

43:24 >> Yeah. Yeah. So, and you know, of course, the risk framework is super important and this is something we we did look and we and we continue to look, right? And we look every time. we are very much into bringing more genai tools and making sure we we give them to our employees but always with you know after we evaluated them through compliance. Now putting that aside you know we started very early with people have to get comfortable with ji tools.

43:48 So back in 2023 we gave access everyone to some to a select group of tools ji driven. The next step after that was look at the specific roles. So look at for example developers versus versus a product person versus a strategy person and see okay what are the specifics or a salesperson what are the specific tools that they could be leveraging their job and how do we drive that adoption we are now in a phase where yes we're still monitoring adoption we are starting to define what are the KPIs to see what does success need in addition to adoption and I There's a part that is extremely important which is the culture change right and making sure that people perceive that to be successful you should be leveraging genai right and and I think that's a big shock because I remember conversations from two years ago where people would say remember talking to someone from the marketing team as an example and they said really you want me to ask you know some sort of genai chat tool what you know what should the draft be and I said absolutely >> it's not cheating the type of things we're looking for. So I think that mindset is more about bringing more of that mindset of it's very important that we all you know jump in this journey because we I always tell my teams we want to be the disruptors not the disrupted ones right so very important to also have that mindset of you know yes let's identify the right tools for each job family to use let's understand what's how does that change their process and their work and then let's make sure that we jump on this on this disruption ourselves and we're leading the journey as opposed to you know being caught by surprise on it and I would say it's a it's a very similar thing of of what we we we observe with our clients.

45:42 >> Amazing. Thank you so much. >> I think we have time for just one last question. You know I joined anthropic before claw 3 launched and the enterprise AI landscape looks very different compared to 18 months ago. Now, what is one thing if the capabilities keep advancing at this pace that you're the most excited about for the year ahead? >> You think about those compressed timelines that we're working with. >> Why don't we start with Josh?

46:08 >> Sure. you know not to do the the topic dure but you know at the end of the day for us we think there's just a significant amount of untapped value in the concept of agentic workflows and and quite frankly there there's a lot of opportunity to make them easier to build and quicker to deploy and a lot of automation can be brought in on some of the topics we've talked about here like how do you do man-in-the-middle reviews how do you do evals how do you do all those things in a more automated fashion. I think one of the the the most interesting things about this entire space is how much of it can be used to build on top of itself using AI to build AI. And we're so early on in the the development of Agentic Workflows. I know everybody's starting to get in there and starting to build these out and starting to deploy and starting to see real value, but it still feels like we're really just scratching the surface of how far this can be taken.

47:07 Absolutely. Andrea, >> yeah, I would echo what Josh said. it's, you know, it's a journey. We just started. identifying workflows is something that we we've been obsessing over and you know we we actually took a page out of how anthropic and the anthropic economic index does their categorization of you know augment augmentative type tasks and automation type tasks. And that's how we're identifying our workflows and targeting the ones that we can automate the easiest first and and the ones that are truly automatable and then sort of leaving the tasks that are more augmented augmentation style we're leaving those for you know as the space matures and and we understand more then we can tackle them later.

47:58 >> Yeah. And I think the complexity of the tasks that we can start to improve and automate will certainly increase over time as well. And our job as a foundation lab company is to make sure that we provide you all with the models that can do those things. So let us know what those workflows look like and we will bring those directly back to the research teams to really improve the next generation of models. All right, Christina, what's on your mind?

48:24 >> Yeah. I would say, you know, what what I we always tell our customers is is kind of start small, but also start something that is meaningful, right? I I would say three things if I'm going to summarize. start small and don't don't try to do everything at the same time. I think to Andrew's point do something that is actually automatable by JAI. There are things where GI is not fit for purpose. So it's important to get that physibility assessment early on and then something that has impact right if you want the combination of the three things like a quick win something that can be done with genai and something where people are going to see the impact so you can drive more adoption across either your customers or your your own company >> I think the quick win point is so important right that's how we think about encouraging adoption for AI for any of our products and services right you need to make sure that claw can read through all of few emails to make sense of, you know, all of the most important things you have to work on before you can deploy Claude onto a discounted cash flow model. So, we want to make sure that we can help our users also climb the complexity curve over time.

49:30 All right. And Andrew, you have the last word. >> What are you most excited about? >> Terrifying thought, but I'll do my best. >> so I think what so so we build quite a lot of our own stuff. We use many different tools. We also do a lot within the application layer. But I think we found to date that across the application layer. There's just the the capabilities that are out there yet don't quite u suit the needs of a lot of our especially more complex processes which is frankly like where the greatest buckets of value reside. So I think you know the thing that I'm most excited to see over the course of the next 12 months is as we build tools across you know things like the cloud API and stuff like that. today it it's still hard.

50:16 you know, we we talk about these magic machines that can do everything. It takes real technical talent, real technical knowhow, to be able to build these things in a way that gives you the comfort that you're not exposing yourself to undue risk. And so I fully expect as the tools continue to advance and mature that that level of of complexity and expertise and things like that will come down to some degree. I I I counsel our portfolio companies quite a lot. and and and often the topic that comes up is like if you really want to do this the way you're describing it, you need to go hire AI engineers and ML engineers and and people who have really deep technical expertise. and and and if you're not prepared to do that, then we're we're probably not quite ready for it.

51:00 but I think that that that barrier is going to come down. >> All right. Thank you all so much. Any other final thoughts you want to share with our audience? And if not, with that, thank you all. We really, really appreciate the time that the panelists took to come and join us. I really want to thank everyone who was on this webinar and who supported Anthropid and Claude as we've taken this investment in financial services and worked with you all to ensure that we're building the right things for the right teams so that they can accelerate their adoption curve of AI inside of their firms.

51:37 >> Now, as you can see, we're committed to financial services as one of the most important verticals for Anthropic. We're just getting started and we're excited to build this together with all of you. So, please sign up, try out Cloud for Excel, the new connectors and agent skills, and let us know what you think. >> We'd love to hear your feedback, your thoughts, your dreams, and we hope you all have an amazing and productive Tuesday. Thank you so much again.

52:00 >> Thanks, everyone.