The AI Balance Sheet

Good morning, everyone. Thank you for joining us. We're so appreciative to have you here. I'm senior finance writer Cheryl Estrada. Welcome to the session, the AI Balance Sheet. We have a great group of speakers to lead the conversation, who you'll meet in just a moment. But before we get started, I want to thank Block for hosting today's session. We appreciate your partnership. So this is meant to be a group discussion. We want to incorporate as many voices as we can throughout the session. We welcome you to share your thoughts and interject. And we have mic handlers who will go around the room, who will pass you the mic in order for you to speak. You just raise your hand and we'll bring the microphone over to you. And a reminder, this session is on the record. Now I'd like to introduce our discussion leaders. Amrita Ahuja, Chief Financial Officer and COO at Block. We have Sheila Glazer, Chief Financial Officer at Synopsys. We have Michelle Gatlin, co-founder, president, and co-chair of the board of directors at Cloudflare, and Heather Zorn, vice president of customer experience, products, and services at Amazon. Thank you, ladies, for being here today. So, Amrita, I'm going to kick it off with the first question to you. Many companies just start adding AI to processes they already have. What made you decide Block had to redesign the business model to really rethink workflows? And what decision made your cost structure adjust or change? So first of all, thanks so much for showing up this morning. I didn't think a panel with the term balance sheet it would get such a great turnout at 7:30 in the morning. And thank you, Cheryl, for doing this. Oh, sure. So at Block, we didn't think about, you know, sort of sprinkling AI on top of our day-to-day work. We thought about fundamentally refactoring how we operate as a company using intelligence at the center. So what does that mean? First, I think you talk about how our teams are structured. Then you think about how we do our work. And then finally, you think about how that actually shows up for customers. So in terms of how our teams are structured, for thousands of years, really, and for the last hundred years of business management, oftentimes well-functioning teams have sort of a pyramid structure with layers of hierarchy through which communication flows. What we see with intelligence is by being able to store sort of that world model of how a company operates, how you think about customer interactions, what's important that needs to be transmitted through those different pieces of sort of workflow or commitments to a customer can be stored in an intelligence, can be stored in a way that you're not necessarily having multiple layers of communication that separate the strategic sort of decision makers from the actual builders. And so the first thing that we did was refactor our organizational structure as a company. We both reduced the size of our company pretty dramatically in a bold fashion earlier this year with a 40% reduction, and we de-layered our company. And we fundamentally believe that in a world of AI, you absolutely need the human at the center. You still need the tastemakers, the goal setters, the strategy deciders, and you need people who are builders. And by the way, that's in every function. In finance, too, we need people who are thinking about building on a day-to-day basis, not just their own functional excellence. But the role of the person who's ferrying information or reviewing the thing before the reviewer, you know, the prep meeting to the prep, you don't need that anymore. And so, first, we completely refactored how we operate as a company through that organizational transformation. And I think it's led to clearer lines of accountability and ownership. I think it's led to faster decision making. And I think it's ultimately led to strong product velocity for us as a business. The second and third pieces are around how we actually use AI both internally and for our customers. And to maybe just quickly speak about some of the high level there, it is not about taking our existing processes today and automating them. What actually you need to do is first completely simplify and oftentimes delete what you're doing, your processes, and rethink them in the mode of, again, operating as an intelligence. And so what we've been doing for the past few years is consolidating and simplifying our processes. We've been using AI within our business now really since early 2024. So this wasn't just about the step change that the models provided, you know, about 10 months ago. That helped catalyze a lot more bold thinking. But it's really thinking about how we automate our workflows, but automate the right workflows first. And what that's led to is a series of different intelligence tools that we use that has catapulted how we do our work. So our engineers have a 150% improvement just from the start of the year to the middle of the year in terms of code review changes. Even non-engineers have a 60% improvement. In code review changes, by the way, non-engineers who are touching code, every single line of code is AI-enabled in some way, and 15% of our code changes are almost fully automated, meaning there's only a human in the loop at the end for the final decision. Everything else is completed by our intelligence tools. And it's those sorts of changes that's led to, for instance, in our Square business, the business that serves small businesses, we have a 3x improvement in product features that we released in the first half of this year, 130 features, three times more than in the first half of last year. So it's really the through line of how we structure our teams, how we use the tools, and in what ways, in what workflows we're using the tools, that then accrues to actual, what matters, customer value. How are you measuring the effectiveness? What metrics are you So I think initially as a CFO, you think about returns, right? And there's certainly a cost to this. It's obviously a big transition for our industry and for every company. And there's a literal cost to it as well in terms of AI infrastructure and tokens and training your employees. And I think there's a couple different ways. The industry is still, I would say, in the moment of time of measuring cost. So how do you offset this cost with that cost? and you might think of the costs that you're offsetting or around personnel, I think the industry is starting to transition to actually thinking about returns. So how do we drive more customer value because of these tools? And again, I think a lot of it, you know, the metric that we're trying to come up with, and I will confess we have not yet fully identified it. So if anyone in this room has a great way to measure it, I'd love to hear it. But the metric that we would like to come up with, we're calling timed customer value, which is being able to trace, by the way, I think traceability and replicability with AI is like the next frontier, but being able to trace this code reviewed change led to this product or feature change, which led to this customer input or interaction, which led to this value to them and value to us and that throughput. And then seeing the speed and the velocity of that throughput is the way we should be measuring AI returns. We're not fully there yet. So we're measuring things like product velocity at our company, which again, we're seeing tremendous improvement on and we want to see more. That speaks to the importance of data. Yes. Clean data. That's right. Absolutely. Sheila. Yeah. So we are a 40-year-old company and we just bought a 50-year-old company. So we're two middle-agers getting together said this is year one of the new company and so what we're really thinking about is what we do for a living is we create software that semiconductor or GPU companies used to make GPUs or semiconductors and those complexity of workload and schedules have never been higher complexity and shorter schedules so our whole goal is to simplify complexity of our customers and that includes how do we simplify friction that we create as a company so we're very much looking at ourselves as we need to drink our own champagne the kind of thing that or eat our own dog food whatever you want to say in terms of simplifying customer interactions with us because we don't need to add to their tension we want to simplify things they have hard jobs so we've used this sort of year one as rethinking how we do absolutely everything in the company and how we remove friction and everything. And similar to what Amarita was just saying, really think about, well, what is the customer journey? How do we simplify that? And really anything we do in the company should relate to the customer because in the end, that's who we all serve. So we've thought about AI very much as creating capacity for ourselves. Part of the acquisition, we had some synergy targets. So how do we create capacity? how do we have more product releases how do we have more capability that we can do the customer and then how does everything else basically shrink or get eliminated and so right now we very much are on the journey to you know measure outcomes and we want to get to customer outcomes because in the end those are the outcomes that really matter because just because we released a product if the customer can't use it or the customer can't quite ingest it that didn't help on customer outcomes. So that's what we're on the journey for. That's the return that we want to measure, the return to the customer, because ultimately that means we're building something that the customer can't live without. So that's the journey we're on. As a CFO, do you find yourself kind of at the center of helping create the strategy to create value with AI? Yeah, it's been interesting. And actually the cost really resonated with me because I don't know if anybody's gone through this journey but the cost for AI was pretty reasonable until it wasn't and when it wasn't it was really not reasonable but everybody had a habit you know it's kind of like having snacks in the break room like everybody had the habit and so we had to kind of reformulate who got the you know premium tiers who maybe is going to use the internal model all those kind of things And that started a lot of intentionality around that. And interesting, some of our AI providers didn't have that metering, that actually wasn't something that they had. And so we had to build sort of that metering. And it's been a journey to kind of educate all of ourselves on who needs that latest, greatest cutting edge model. Where do we need to deploy that? And where can we use some internal models on some use cases? And suffice it to say, the engineering folks that are working on innovation are going to get the latest, greatest tools. But it was, I'm going to call it was a moment when we all had to get together and sort of realign on how we were thinking about using this in the company. Because everybody had, you know, developed habits and things like that. And it was helpful that I was seeing other people having the same problems. that we didn't feel so alone. But yes, in the end, working with the gentleman who runs our kind of IT organization on this, that became really important. And we are investing in AI. It is an important investment. We want our team to have the latest, greatest tools. But just along with everything else, it's being super intentional about where that investment is. And that then forms the strategy of how we think about, you know, how we get customers. I'm going to come to everyone to see if you have any questions or comments in a minute. So we pick up some questions. So Michelle, Cloudflare is actually actively transitioning into an energetic AI first operating model. Can you tell us what that means? Sure. So nice to be here. And actually, I think the I feel like my panelists have helped tee this up. I mean, everything is changing. And so I can agree more with him, Rita, where like in businesses, we all have to rethink the factory floor, the process, like rethink it and not add it in, but like restart from scratch. And so we did the same thing at Cloudflare a few months ago. We started later than Block. And I think what's been amazing is how much progress you can make right now in a pretty short period of time. And it really, everything really did change last November. And I think about like how far the industry has moved in less than a year. You just think, okay, where are we going to be a year from now? And I feel like we're, it's very, it's kind of a tale of two sides. First, it's like so exciting to have a front row seat at all of this. We are all so lucky in our career to be able to be shepherding this. On the flip side, it's hard too. Like it's challenging, it's hard, there's unknowns, there's hard problems. And so on the agentic side, I mean, that is an extension of our team. Like we hire both employees and agents at Cloudflare. And like that's just like the high level. And then it's both internally how we're getting our work done. and so there's so many examples within the business of saying how do we take every single person at Cloudflare we have about 5,000 people at Cloudflare and make everybody a builder so if you're in finance you are now a builder and you have the tools built to go do that if you work in the cyber security our internal security team you're now a builder too and so everyone's becoming builders within our business and they're using agents and our company has given them a place to do that and then the people closest to the problems can go solve it and it's been amazing to see where the best builders come from it's not where you expect it's like um like one example is like one kind of person who's really risen up is someone i mean she was a director in our finance organization who was the heavy power user within finance building these agents and now our CFO is tapped her to say hey come help be my head of AI transformation within the whole team and so it's been amazing to see who the stars have been and it's not where you would have predicted beforehand so I think that's pretty cool but the second thing I would say is like I love that we're doing this but also like how do we help businesses around the world do this and I think this is the other piece of the puzzle it's like all of our organizations have to be going through this transformation but the real powerful one-two punch is how do you help your customers get what they need more how do you deliver that value and so for us we sell to businesses and so we have become the place where businesses are building their agentic workflows and all of their AI on the AI cloud which is a really good place to be and so that means things are super super busy right now but exciting because you're saying hey how can we help um you know so and so business achieve their goals and it feels really cool to be part of the solution not part of the problem so in these moments of transformation you want these things to be tail tailwinds and i know this is about the boardroom and you know i sit on my board at Cloudflare i also sit on another public company board and i sit on a non-profit board and what i see is when board members say hey what are we like are we worried about AI is it an asset is a liability you for sure want the answer to be a tailwind it has to be true but you want it to be a tailwind we want we're doing the transformations internally you need to have the proof points of examples hey here are the products that we're now in market and they're selling and actually the ultimate i would argue the ultimate KPI is revenue growth that is the ultimate KPI we don't need another KPI we just need revenue growth that's the actual KPI that i would argue Michelle in the process of incorporating agentic AI, I think there's a change management element because you're teaching, you know, your employees how to work with an agent, can you tell us a little bit about how you approach that? I mean, that's true and I feel like that was a really good question six months ago because at this point I just feel like there's been so many articles we're kind of like in that change curve, not everyone's along the way but there's a little bit like if you're not on this program you're already behind I would say and so organizations absolutely have to give the training and the tools to the teams, it's up to the employee to start to learn how to do it too, like it's, I think it's a two-way street I had a chief people officer many years ago, I mean I work, I'm a startup founder, my company is now large but this idea there's no better job training than on the job training and there's something about why apprentices are powerful, it's like you see what you're doing, like there is something about using it and doing it but one of the really effective things we've seen is we hired a lot of interns this summer this year, so we kind of committed to hiring over a thousand interns over the course of 2026 and we're not quite done 2026, but we've, we've more than have 80% of them who've gone through our program, that's been an amazing catalyst for change. So the hiring managers, a bunch of different teams brought in interns and those interns helped the team be more AI native and ship things and show them how to do things. And everyone got better that way. So that was like one tip. It's a little bit, I mean, we didn't call it that, but it's a bit back to everybody was learning by doing. And I'm a big believer in the more that you can ship things and focus on the work to be done by doing and building, you learn so much more than reading and talking, that's true, yeah I do want to just, I know you have a question for me yes but one thing I do want to add in there is I think that the important thing as a leader right now is to make sure that you're giving time and space and permission for your teams to learn so that's why I would say it is on the employee but that's actually one thing that we're talking a lot about in our team is that some of the tools that have reduced toil internally it's not necessarily that immediately you get all of the efficiency in terms of throughput because what's actually happening is people are learning how to use new tools they have to incorporate these new tools and totally redesign how they're working right so there's some things that we might be redesigning but there's a lot of just like how all of us are working I was having conversations at the table last night and somebody was saying you know in my for my side hustle I'm using an agent that comes up with my agenda and my schedule and allows me to volunteer in this way that I couldn't otherwise do and so I do think that's a big part of leadership right now is making sure that you're creating a culture of learning um and the and and making sure that people feel like they have permission to spend time and understand that the kind of cost curve or the time curve you know rises higher than the returns and the last thing i'll just say is i like for sure and then we have a place every week where people can show what they're doing and share it with the company and celebrate it because part of it is we don't even ever it's new for everybody and so it's like hey what are you doing with it and so there's kind of very organized planning and then there's kind of hey this is what i'm doing here's the problem it's solved and then like a way to surface it and i think that's been like a good place to celebrate the place the builders who are doing it who've made the time and then other people say hey i want to do that too that's great insight um heather how are you approaching ensuring AI serves the customer rather than improving operational efficiency yeah i love this question and and this is one where Amazon's philosophy of working backwards from the customer and having customer obsession as the first leadership principle really helps us because everything we do starts with the customer. So even if you think about something internal, let's say toil reduction, right? There are a whole suite of tools that are available to us at Amazon that have been built by AWS Quick and Kiro that are super powerful, which are really about sort of internal productivity. But the whole notion is that all of that is starting from what's the value you can deliver to the customer and so it really isn't necessarily a trade-off for us I think that the whenever we're thinking about any new product or feature we start with you know the the ideation is how could this win for the customer and then you work backwards from that to say how does it also win for the business and so obviously there are a bunch of ideas that we come up with that win for the customer that don't win for the business and those are not necessarily going to get prioritized but that's where we start everything it's a process you know you don't immediately come with this is the exact formula that we're going to use it's an iterative process yes it is an iterative process absolutely and i think again um you you want to create win-win solutions right you want to create something that wins for the customer and then you have to find a way that it wins for the business as well and you have to figure out what's the time arc that you're willing to invest in order to get the the wins both for customers and for the business. Does anyone have any question or comment? Yeah. So my name is Sandy Carter. I'm currently the CEO of an AI company called Equus.ai. I used to work at Amazon. I worked at IBM. So I've been in the tech space for a while. So this question is a little bit different of a topic, but I don't know if you guys heard but the Grace Hopper conference just got canceled um and if you don't know grace hopper it has always been the world's largest conference for technical women to come learn this year was on ai fluency and typically it also helps for people to get roles and jobs as well and i'm just curious what your feedback is on that um i know one conference doesn't make you know a generation of women go backwards but it is a setback so i'm just curious all of you guys what do you think about it and what could we do about it i'll take that yeah i saw that i went to the Grace Hopper conference one time i worked at Intel for a long time and one of the women at Intel was awarded and i still remember was that at that tech mark that seems to never have been redone for two decades um but she gave one of the most inspirational speeches and she you know worked through a lot of hard times to be able to get to that place and that room you could have heard a pin drop was so powerful a woman who would reach the top of her game getting this award and when I saw that I was heartbroken because to me it is I don't know if it's a canary in the coal mine or what's going on, but it can't be because funding isn't available because we all know funding is super available right now. So I think we have to do something about that. And the thing that it reminded me of, and again, back to my Intel days, is I remember when Westinghouse could no longer afford the science fair and Intel took over the science fair. So that's been one of Intel's big sponsorship for decades. And I thought, I don't know, you know, who's the person that's going to take that over. But, you know, there's plenty of money to fund something anyway. So to me, it is not a harbinger of good things. And again, I remember sitting in the audience just I literally remember every word in that speech. It was just so powerful. And to see the recognition for this woman who had worked so hard. So I don't I don't know why it got canceled. I didn't understand the details, but it can't be funding. Yeah, but it can't. That math's not math-ing. So somebody needs to pick it up. I don't know who's the person to pick it up, but maybe we should get ourselves together and pick it up because I think it's that powerful. Hi, Shivani Huja from Salesforce. So coming back to ROI, at the macro level, you've got a lot of organizations that talk about cost being one of the cost takeout is one of the initial metrics that we would use. And then revenue generation is the ideal and the optimal. I'm trying to explore for even for Salesforce, us becoming the agentic enterprise. We talk about drinking our own champagne in a very classy way. Is what I'm trying to think about is yes. And from a developer perspective, it seems like there are greater metrics to be able to measure productivity to determine product velocity or code enhancements. I'm trying to find folks that are also thinking about or whether they are at the functional level. So we want everyone to be a builder. So how are people in human resources? What's the directional metric that demonstrates that giving them access to such coding tools, advanced or not models, that what's the directional metric that is showing that there's enhancement improvement? Have you been able to, at the functional level for finance, for HR, for legal, for procurement, been able to identify micro metrics for ROI? I can start really quickly. Like the progression that we look at at Amazon is adoption engagement efficiency and then outcome and so you just go along that arc right and so um i can talk about i have a design team and at the beginning of the year zero percent of them were creating prototypes or coding with Kiro and so the first thing and this was all based off of the the leader of that group was like wanting to really make the design designers more powerful and you really think about that role expansion so this was like completely on his own right there was nobody saying go make all of your designers able to code and build prototypes he was just like i'm gonna go do this and so the first thing was adoption right and that was um getting all of those designers to come together from all over the world and actually start getting on the environment together right so then it was like we got he got to 100% adoption and then it was starting to measure engagement so how many of them are actually going back now that they have the environment set up and actually going through that learning arc and starting to do the work and then you start to see efficiency right where you get to see some of those people who are adopting and then how quickly are they actually producing they're now producing things they couldn't produce before they're actually getting from a design idea to a working prototype that then can be handed off to an engineer versus having to have a front-end engineer doing the coding so you kind of remove this whole layer of work that needed to happen before and then you get to outcomes right where the outcome in this case would be sort of speed to hand off for a designed product i would i would add maybe just a simple metric which is time saved yeah that's the easiest one which is like you've you've got i don't know 30 hours in every board cycle that a team is is working to put together that set of materials and review that set of materials and it goes through the chain and now it's something that's actually when you think about it fairly formulaic and can be automatically updated as the numbers are updating too um and that saves 25 hours or whatever it is and you do that across a hundred different processes and before you know it you've freed up your team to do far more strategic stuff so these are the logically this makes sense at the again super micro level are you have you got a singular person that's aggregating this at the board level that like the the taking it from the concept and the theory of yes theoretically we should be able to reduce the number of hours for boards. Is that sitting and living within finance is a finance story to tell? Or is there someone that is aggregating that to a bigger story? And have you found a systematic way to aggregate the ROI? So I would say we're doing it at a discipline level. So customer service might be the best example, which is like how many customer interactions can we actually make better through NPS scores or, you know, fully automate, you know, find an answer faster. And so it is probably discipline specific. I don't know if there's a company-wide holistic measure other than our top line metrics. But, you know, I think there's the, so those are the bottoms up metrics. Then there is a tops down metric, which is, you know, are you able to do your work more efficiently? Are you able to like handle more scope with the same team or have a smaller team for the same scope? So those are probably the top down constraints that we would have. Yeah. And I would say for us, it, you know, again, it was thrust upon us. So we added like 22,000 customers and then we were able to take cost out. So the, you know, kind of the micro, okay, this used to take two days. Now it takes 10 minutes, but then obviously that has to ladder up into, okay, you can actually do that in a smaller cost envelope because you've created more capacity with that same cost one other thing we did just in addition to what i'm hearing here um it's so funny that even in a world of AI sometimes uh the good old-fashioned manual things sometimes work out so we we what we found is like we definitely have a team whose job is to help refactor the process of Cloudflare but to do everything is impossible so that we also want to make everyone builders closer to the problem and so we have a really good sense for the team what there we can track that really easily but everyone else is harder to track and so we kind of have created we use an email alias you know that's called AI wins and people submit them and we track that and we aggregate it and so and then we show it at our company all hands and so it kind of shows by team and it's cool it's kind of cool to see the people in places team on there and like they it's self reported but like the point is it's just some way to show this is happening everywhere and we're all in this together and so just it's it's not like the most sophisticated way but sometimes quick and dirty does the job yeah um Liz Brand from Prudential uh you know we talk in a world that's very needle up needle down a lot of pressure is being put on both sides but you know there's an increase on the needle up in the growth and revenue side of things and i'm curious what you guys originally started the panel talking about getting rid of things you know throwing them away deleting and do you see AI being leveraged more in the actual like business landscape not just the underlying processes and technologies because you can get rid of processes you can redesign them get rid of units of things but the actual looking at your your what you're the the product that you're offering to your customer and potentially removing certain products or certain components of products based on their total cost to their revenue i think absolutely for us like if you think about what it takes to serve a small business no two small businesses are alike and we serve four million of them across the world and um and so the the opportunity for a small business to actually kind of build what they need using our primitives around payments and commerce and banking and software and hardware to to to tailor that um in a way that suits them So actually, they're the product builder. We're the infrastructure layer on which they're building. Now, we're not there today, but we are getting there. We imagine AI products being like the COO for a small business owner, the COO who's not going to take equity, you know, or get paid, but that can help them really manage and operate their business. And so, yes, absolutely. And we need to think of it that way ourselves. ourselves like we've got to make all of our team members builders and then we can bring all that learning to our customers too yeah for us it's much more um everybody is trying to build semiconductors yeah absolutely everybody because you want to have a workload down semiconductor because then that's not wasting energy um and wasting costs there's not enough semiconductor design engineers in the world so now everybody wants to build them and they want to build them in a year instead of three to five years and they want more complex ones and so what we're offering what we're incorporating into our tools is agents uh we just uh announced a big deal for a specialized model with OpenAI two weeks ago um because the world does not have enough semiconductor engineers and anything that goes wrong anywhere in your design process will mean the chip fails and so it's like you know five thousand nines is what you need and so that's how we're incorporating that so that our customers basically can get these complex chips done with high confidence in these tight time frames and so for us it's it's the only way that we can solve this problem is actually to give customers agents that they're in charge of and you know they direct traffic on but it gives them capacity to be able to you know solve these hard problems and build these products that they need for their business and I know you said about taking things away I think it's hard to take things away from customers so if you're gonna do it you gotta be really thoughtful about it and maybe offer something new at the same time and give it well you know it's a negotiation a little bit but it's just it's hard to degrade or take things away it's value for value yeah but if there's something hey we're sunsetting this but this is something new we're moving to I think that is a perhaps a good place to look i think we have time for one more question hi i'm bex paul i'm the chief people officer at octa um and rita this is largely a question for you but also interested in other perspectives as you did the 40% reduction and transformed the organization does the org chart still look the same because i can't help but think in the future that with ai you know, most of our companies probably have G&A, sales, R&D, that the org chart will look radically different. And I just, you seem to be very advanced on that journey. So I just wondered what it looks like and where you're heading. So, I mean, a couple of principles we followed, which is fewer layers and greater spans. And that's because we can enable managers to do more often with agentic tools. And we're still on that journey. We're still working through that, But but we're well on that way. I think where things are headed and I've already heard about a couple of companies doing this who are on the frontier is you look at your org chart, you know, that your company publishes and it will have not only humans, but agents on it because you have roles that are now fulfilled by agents where you have multiple people interacting with them via Slack, via email, via a voice even. and those people are responsible for that function. Those agents, I should say, are responsible. Those personas are responsible for that function. And so I think we are headed in that direction. And so you can imagine that someone who's an individual contributor today actually has managerial responsibilities over agents because, by the way, those agents, not too dissimilar to humans, need to be developed over time. They need more context. they need more information, they need to understand their goals, and those may change. And so I do think we're headed into a new frontier as to what the organization looks like, that block is not even an ad, but we're, you know, positioned, we're learning just like everyone else. Yeah. Yeah. That's all the time we have. That was such a great discussion. It was so good. Yeah. Thank you. Thank you all for joining us, and a big thank you to our discussion leaders.