Given that we've only got a limited amount of time, I thought I'd crack on and Kash will join us in a second. Okay, so Janesh Moorjani, CFO of Autodesk. Great to be here. Simon Mays-Smith. Thank you, Vinny Kash. Matt, you're gonna step in? Well, yeah. Here he is. Okay. Yeah, yeah. Let's go. Oh, great. Okay. Thanks. Thanks. No, you're good. All right. I'm done. Okay. Okay. Oh, welcome, Kash. We've got the launch buffer as well, so we can go two minutes over. We're live up on stream. We're streaming live. Yeah. Okay. We always make an inference with the CEO. Yeah. Is everybody having a good time despite my couple of minutes of being behind on schedule? I speak faster, so we'll compress it all. Thanks for coming once again. Thank you. Such a delight to see you. Yeah. Congratulations. I've not hosted you after you became CFO of Autodesk. It's been many, many, 10+ years since we got to know each other. Yeah. And your days at VMware? It's been a long time. Yeah. Yeah. Yeah. We've known each other for more than a decade. Yeah. Absolutely. Matt's gonna chime in here with some questions as well, but I wanted to set the stage after joining the company. First of all, give us an overview as to what your—what are the key things you're working on? As CFO, every CFO coming in has a mandate, right? What is kind of your mandate? And also, as an adjunct to that, what does success look like for you and Andrew and the team in the next four to five years ahead? Yeah. Great question, Kash. So, look, as I joined Autodesk back in December, when I joined the company, coming in, I had two distinct pieces, right? One was just around our ability to continue to prosecute the market opportunity ahead of us. What I saw was a company that had been a consistent, steady grower over the past couple of years, that is an industry leader in the spaces in which we operate, well regarded by customers, well regarded by partners, and had been at the forefront of driving many business model transitions and industry transitions, over its existence. And at the same time, a company that had a massive opportunity in front of it in terms of everything related to driving the convergence of design and making the cloud and with AI. And so, job number one was obviously to think about how we can position ourselves to capture that opportunity while building on that strong foundation. And the second piece, which was equally important, was making sure that as we drive the business forward, we drive a degree of optimization and operational efficiency from the standpoint of expanding profitability, which has been a top-of-mind issue for many of our shareholders for some time. And so it was all about expanding margins and expanding profitability and doing that even as we grow in months. I'm pleased to report that we've been able to do both, and you've seen that so far in the results for this year. And looking ahead, I think it's more of the same. If I think about our opportunity to continue to build on the strong investments that we've already made to build to capture the future with AI, and the convergence of design and making the cloud, that's where we are focused. Five years from now, if we're successful, you'll see that reflected in our financials, and you'll see that reflected in the growth and the profitability of the business. That's what we're focused on. Excellent. Excellent. As you went on this 10-month journey, Yeah. Many of us, when you go into a job, we have an expectation of what it is, and then there are certain things that catch you by surprise, positive, negative, whatever it is. Yeah. So how would you characterize the things that surprised you relative to what you expected anyway going into this job at this company? To be honest, there was no real negative surprises. Yeah. There were only positive ones, and maybe that's just because, you know, in terms of how I think about things, it maybe comes with CFO DNA that you tend to think about how you de-risk things and as you look ahead. So I did a fair amount of work looking at the company from the outside, talked to, clearly, the management team, board members, a number of other folks in the industry, and so I had a very good thesis going in, so I'd say there were really no adverse surprises. In fact, only positive nuggets that have uncovered along the way. Got it. Simon's gonna say, "I've heard this so many times, Kash, you stop boring yourself, painting yourself into the corner of being the oldest guy in the room, but the first software application that I used even before Microsoft Word and Excel, 'cause those things didn't exist, was AutoCAD and LT. Like back in college. Yeah. So I never knew that that would be a public company and that it'd be covering the stock and having the CFO of the company on the stage with us. There was a time when the software used to ship on floppy disks. Mm-hmm. And now. This was before floppy. This was tape drives. Yeah. We had an air-conditioned room, and only three of us could access it, and I was not the best programmer, by the way. Yeah. I kind of fumbled around, but that was my first exposure. Let's talk about go-to-market. Since there's been a lot of changes, I think Autodesk went through its first big change, at least in recent times, in 2016, the subscription model transition. When you came upon the scene, it was this transactional model. So how has the GTM function evolved at the company, and how do you see that evolving as a result of the transactional model? Yeah. We've been on a journey. I talked about business model transitions that we've been driving, and we've been on a journey to evolve our go-to-market, from the standpoint of how it's set up and naturally then the financial implications of that. Mm-hmm. That's a journey we started on a few years ago. Many of our investors have patiently been with us on that journey. We're coming to the tail end of that, overall business model transformation. And that's gone well so far with our partners. There's been some areas of operational friction which we've addressed along the way, but by and large, I'd say it's gone well. Mm-hmm. And so starting in February of this year, we then initiated the optimization phase of that go-to-market transformation, which meant making some changes in terms of how we're structured from a go-to-market standpoint and focusing on certain areas where we needed to divest so that we can reinvest in other areas as we are building capabilities for the future. And even that has gone reasonably well so far. We're now about six months into that. So far, I'd say it's gone really well, and we're starting to see the results of that in some of these early successes that we talked about. Yeah. Can you share with us a little bit more detail on the optimization? That, that word I've heard since you joined the company. Yeah. That is, it's certainly like you got hired to do optimization in addition to a million other things, right? What does optimization mean in your lingo, and can you apply it to a few business functions so we can kind of visualize and let you size what that means? Yeah. It's mainly in the context of go-to-market, and I'll be the first one to say that it's not just me, right? This was a body of work that had been already underway, and the company had been working on it. So it's not that I show up and I start to optimize, but the team had already been doing a lot of work in this area, and we already had a plan when I joined in December. The plan for this year was almost set already. Mm-hmm. But, to put a little bit more meat around the bones on what that means was part of this initial phase, we focused on some of the non-selling functions. We focused on, in areas like marketing, where we wanted to cut back in some places so we could reinvest in other places. We focused on our customer success functions. We focused on some of the sales operations and sales technical functions because we wanted to build new capabilities for the future, and we're reinvesting some of the savings from those actions so that we can continue to build tighter channel partner integration so that we can continue to build greater self-service capabilities to service our customers. Those are all of the things that are underway right now, very specifically. Got it. Simon, I'll loop you in on the transaction model piece. I just wanna get a. You've got a PhD in the transaction model. I mean, even before Janesh showed up, he was talking about the models. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. PhD now, but maybe just give us a mark-to-market on kind of where we stand in terms of kind of the rollout of the transaction model, the expected benefits, and if you're realizing any of those today, given that we're still pretty early in. Yeah, so on track, the way I describe whenever you're doing big implementations is you have a bunch of assumed one standard deviation things, and we've talked about things about channel partner products. Mm-hmm. And deteriorating during the transition just 'cause they had a bunch of other work, but we haven't had any three standard deviation events, thank goodness, and that's a function of good execution of the process. The other thing to understand is that this isn't a single sort of one-off, one-and-done. It's a bunch of optimizations, to Janesh's comments, both across revenue and across cost over time, some of which happen quickly. You know, so for example, the 9% RIF we had earlier on this year, and some of which happen over time. You know, so for example, we've talked about some of the lower-end customers' channel shifting and coming directly to through the e-store. That happens over multiple different quarters. So you just talk about it on sort of different durations. And then the sort of final bit is the sort of financial impact is that you have essentially the mass impact of the pre-revenue cost shifting from pre-revenue into sales and marketing. Sort of that comes through billings first, and then it flows rapidly through the P&L, over time. And so we're sort of past the sort of the billings peak, and we are approaching the peak of the revenue peak, and then it'll start dissipating as we go into next year. And I think one of the, like, end-state advantages of the transaction model is having more data-driven interactions with your end customers, right? So are you starting to see the effects of that with the rollout of the transaction model? Is that translating into incremental cross-sell opportunities? Too early, as you know, so we're still very focused on getting the transaction model right. We're still in the sort of lift-and-shift phase. That's the sort of data phase, which is, you know, around lift-and-shift and optimization, and then you get to a data phase. So that's the vast majority of that opportunity is still to come. Okay. Great. And then, Janesh, maybe just stepping back a little bit, you've seen a lot of momentum in the construction segment, right? So can you tell us a little bit about what's driving this ongoing momentum and where you see the next potential leg of growth for the AEC category? Yeah. Construction has been a bright spot for us. It's performed really well here in Q2, continuing its streak of strong performance from prior quarters as well. And a lot of that is strong addition in net new logos in construction. A lot of that is continuing to see larger expansion. Construction is a, it's you know it's a single industry, but there's many different use cases or many different types of construction. And one of the nice parts about our business is it's highly diversified. And so when you see strength in some areas, that can easily offset slowing down in other areas. So in this past quarter, for example, we saw strength in data centers. We saw strength in industrial and infrastructure buildouts, and that more than offsets some of the softness that you're seeing on the commercial side of construction. That's actually helped us quite nicely, and I think that business will continue to perform well for us. Yeah. And And I wanna drill into the ACC product specifically. Simon, I know you have thoughts here around kind of the broader competitive landscape, but just give us kind of a broader understanding of kind of where you think Autodesk fits into the kind of construction management category and sort of maybe the competitive, you know, win rates you're observing against some of the other players out there in the market. Yeah. The opportunity is the same as it's always been, which is the single most important strategic priority for our customers is to wrangle their data. You know, so if you speak to, you know, Bill Oakland of Oakland Construction, he's saying, "Look, I've gotta get control of my data, which is partly around putting it into the cloud, but then it's building value on it." The reason they wanna build value on it is so that they can work remotely 'cause of the pandemic, so they can manage supply chain shocks, so they can apply AI to it. Pretty much everything that's turning up in their everyday life is driving more cloud and more data wrangling. Sort of more prosaically, if you speak to the construction customers, what they'll tell you is that 70% of the problems that turn up on a construction site are because of decisions taken in the design and pre-construction phase. Yeah. So that's why they wanna connect it upstream, as well. And then they also wanna connect it downstream once they hand over to the customers. You know, so you've got a data center that's got a lifespan of however long, 20, 30 years, or however long it is, and optimizing that, the construction customers wanna have an ongoing relationship. So having a digital twin that allows them to have an ongoing relationship on the operations and maintenance side and to extend the relationship downstream. But at the core of it is wrangling data. I'm curious, like, from a field execution perspective, like, how important is it that you guys own sort of the pre-conceptual design phase in terms of selling in this kind of broader construction management platform? So the answer is the horizontal connection is already important. There's gonna be turbocharged importance in an AI world. And the reason for that is that we will be able to pull forward, essentially simulation much earlier into the construction process. What I mean by that is that you will be able to, when you have a block of land over time, before you've even started building anything and just playing around with blocks, to have the software tell you, "No, no, no. Don't do that. Otherwise, you're gonna have a $10 million problem when you start building." And so being able to pull forward problems that you're going to have further downstream much earlier before you're set on a certain path is a significant driver of value. And maybe we can talk about sort of AI monetization at some point. Yeah. As you move from, up the scale in terms of, AI. Yeah. Let's talk about AI, from a product perspective and then maybe from a monetization perspective. So I think, like, Autodesk has effectively sort of two approaches, right? I think, like, right now, you're infusing AI in some of these kind of core more rote functions with, you know, solutions like auto constraints, but then you have a big grandiose vision around things like project needs. Can you maybe unpack that opportunity that you see from a product roadmap perspective? Yeah. The same thing, it's the fact that AI is a continuum, not a fixed point. So a continuum from feature automation, which is part of a workflow, to workflow automation, which is the traditional silos you think about in design software, to system automation, which is essentially cross-silo or end-to-end when we talk about. And so that's some of the conceptual design stuff, being able to make influence on the construction phase that I was just talking about. As you move up that stack, two important things happen. Firstly, the model sizes get bigger. From a relatively small model for feature automation to workflow automation to system automation, the reason that's important, the model size, is that for feature automation with a small model, you can deliver that and have the compute happen on the desktop. And that means you can deliver feature automation value in the way you would with a traditional value of product, which is delivered as incremental value to a subscription product and the value captured in price. And we're doing that today already. Mm-hmm. But as you move up to bigger models, you can't do it with a traditional subscription product because you're delivering a lot more value, and you need to capture that value, but you're also delivering more variable cost through more compute, and you can't put that through a fixed subscription price point. So as you move up that stack, you will shift to more consumption type models to capture that value and cost. That's the first thing. And the second thing is that as you move up that stack, the breadth and depth will move beyond the capabilities of the human brain to be able to do it. And so what that means is that you will move from human-led processes to machine-led processes. So, because a human brain cannot make those connections between a block of land and what's going on in construction, and a machine can. The reason that's important is that we will be able to deliver significantly more and greater dollar value of value to our customers as one vector of opportunity for ourselves. And also, there's a second vector, vector of time, which is that machines don't have to sleep, whereas humans do. And so you have, you know, machine working 24 hours a day, 365 days a year. So, yeah, except for junior, you know, for analysts, yes, I used to be one of those, so I know, I know how that feels. And so there's two vectors of monetization opportunities. So for everyone sort of worrying around with their hair on fire worrying about seat-based models, what you are not thinking about is that machine, incremental machine consumption, which has two significant vectors in monetization. Simon, on that, how should we see the end-user productivity? So certainly, there's a vision of what AI can do in that scenario, right? How do you get paid, and what is the incremental value? Because Autodesk, as it is, brings a lot of value to its customers. And you've been on this digitization journey, digital transition, a lot of turns of the crank or turn of the crank, if you will. So what does AI allow the customer to do that they maybe did not need an additional engineer, or is it unlocking new scenarios that would've required some kind of time, resources element? How do you equate it to the value realized for the customer in tangible terms? I think it's gonna be both, to be quite honest, right? And Simon was describing very effectively that you can have certain things that are just tasks that are automated. They make the user more productive. And you see that in a number of applications, and you'll see that in our products too. And those are just monetizing the ongoing value of the subscription. But then, AI unlocks so many new things that previously were not possible, and connecting it back to the thread that Simon was talking about around the value of collaboration across the ecosystem and around the value of being able to envision problems before they happen much earlier in the life cycle, those are things that unlock tremendous value for customers. And so they will save money with that, and therefore, there's value that we deliver, and we can monetize that value in the form of these consumption-oriented models. That's the second piece of it. And then finally, as you're doing that, each user becomes much more productive. If you think about traditional seat-based models and P times Q, Q models, over a long period of time, what'll end up happening is that because there's more machine usage and less human usage, how we think about the transfer of that value to the customer will change, and therefore, our monetization approach to that will also change because when machines are doing the work and humans are that much more productive, we're delivering significant value to customers, and they're willing to pay for that value. You'll see those evolve as well in the form of the consumption models that Simon was talking about. Just to give you sort of two specific examples, for a complex, you know, manufacturing object, to put constraints on it, you know, and how the dimensions sort of fit together, for a complicated object can take four or five days. With our constraint automation, we can do that in seconds or minutes, and so just a dramatic saving in terms of time for the customer. That's one way. Another way, which is one of my favorite examples, is a customer who was building a hotel in Miami and put a sundeck on the side of the hotel. Considering Miami, you wanna get a 10. but when they ran an AI automation around it to show how the sunlight affected the building, they figured out that the sundeck that they put on was casting a shadow on the pool for 70% of the day. That's a bad day at the pool, okay? And that would've cost them millions to rip it out and put a new one in, okay? So just two specific examples of how this can help our customers improve their productivity. Interesting. So have you thought about, maybe it's still early, how you price the consumption credits? And you start off with a limited number of credits that are part of a free pack, and then there is an upside trigger that how have you given thought to how you could get economic realization from? We have. And some of it, as we said, will be baked into existing products and reflected in the price of those existing products. And some of it will be additional capabilities, particularly if it's more resource-intensive and causes the compute meter to spin faster at our end. We wanna make sure that, that we're monetizing that the right way. And so as you think about these more complicated, workflow automation and systems automation, kinds of, use cases, those will be things that we will monetize, through, you know, either tokens or otherwise. And we'll talk a little bit more about this at our user conference coming up at, at Autodesk University. Yeah. Yeah. Yeah. At AU. So, we'll share a little bit more at that point, but for sure, it's gonna be both of those kinds of models. Figma is not exactly in your space, but, design. They have taken the approach of investing aggressively to push Figma Make, which requires a lot of GPU cycles, and so we all have to take a big step down. Not anything surprising 'cause it's all contemplated and premeditated, obviously, but the pace of margin compression that they had to endure to generate new revenue streams from Make was quite significant. How do you guys think about, since you have your own model, is it a little bit more efficient from a GPU consumption standpoint, or what are the puts and takes of going for this? Are there sacrifices that need to be made before you can start to see the benefits? Yeah. We've been investing in this area for a long time already, and so this is not new for us. And there will be efficiencies that we can gain in terms of how we think about just resource intensity of our own compute engines and so forth. So we will continue to stay the course on it, on investment in this area. We think it's a super important area for us as we think about how we drive growth. But the nice part about our business is we've shown is we have operating leverage, and so we can reinvest some of the portions in the business to go drive growth for the future, but at the same time, continue to deliver operating margin as well. And just because we've been at it for quite a while, and 10 years and far longer than most of our peers, we've been solving some of these problems as we go along, or at least slowing the rate. So, for example, what we've done is we're using AWS as our hyperscaler, but we've essentially built an AI stack, our own AI stack on top of that, and we are 70% more efficient because of that compared to AWS's own stack. So we knew this was a problem 'cause we've been doing it for some time and solving that problem as we go along. Similarly around particularly on the AI side? Correct. Okay. Correct. For Autodesk specific use. Autodesk specific. Yeah. Correct. Yeah. And then similarly around data, you know, a lot of data, one of the challenges that other AI engines have is that they're built on 2D data 'cause there's very little publicly available 3D data. We have access to a bunch of 3D data, obviously, and that's one of our sort of secret sources. But a lot of that data, not just for us, is just 'cause of the way that our customers build models is fragmented, and just figuring out how you take fragmented data and put it together in a way that means that we can extract value from more of our data than our peers, again, hard engineering to do. You know, so if you look at, and there's a bunch of other cases. So it's not just the sort of AI capabilities itself. There's a bunch of, you know, scaling questions we've asked us are answered. There's a bunch of data questions that we've answered and a bunch of other stuff, which means we think we are years ahead of our competitors in AI. Not that I intended to ask this, but as I've listened to you talk about that, this concept of a world model where the inputs are actually video and images as opposed to a text-based model, which is the world that we live in, how are you training Bernini? Open-ended question, so I mean, based on data that we have access to. So is it text data, or is it like video data, or? It's multimodal data. Okay, so our customers don't work in jargon. It's a pretty sophisticated engine. It's not like. Correct. And that's again, so use a simple analogy. The vast majority of AI engines, if you ask them, produce a water jug, will produce you a beautiful-looking water jug. But if you rotate and look through the top, they won't have a hole in the middle to hold water because they have no concept of depth. Now, there's a bunch of other things around how the geometry is also not manipulable as well, which our customers care about, and a bunch of other stuff. But in simple terms, you know, and having that concept of depth is a key differentiator. I watched a demo of a world model. I will not name the company that made this world. I can guess. Yeah. They had a slicing of a bowl of cereal, which had milk in it. The cereal, the hard part, was cut by the knife, but the milk did not spill. I said, "No, that's a big shortcoming, right?" So that it more or less kind of applies in the old world when you do simulation, slicing through. But there's a bunch of problems moving the deck and not have the shadow go up on the pool. There's a bunch of problems you have to solve. It isn't one problem. And we've been solving those steadily over 10 years. And, you know, I'm sure we're going down various cul-de-sacs and holes. And we've been able to do that with relatively modest amounts of investment 'cause we've been doing it over time. But those folks who are following behind us who are gonna have to spend a lot of money and do it quickly, they're gonna have to, the return on that investment will be less than ours because we invested early. How much of the other models do you have to use, like an OpenAI or an Anthropic or versus your own proprietary Bernini model? Well, we leverage them. We leverage them to a degree, but a lot of this is actually Bernini is proprietary research. Yeah. And it's built on it's been trained on the data that exists within Autodesk. Yes. Yeah. And as Simon was saying, a lot of that is actually multimodal data, including drawings and so forth. Yeah, so there's a lot of goodness that's being built within Autodesk, in Bernini. And again, you'll see some of that at AU next week. Mm-hmm. But and how close are we to Bernini being, like, avail able in its full form and functionality? So capability. I mean, you'll see more at Autodesk University next week as we're sure to have to. Yeah. Yeah. Got it. Got it. And Janesh, I wanna jump in here with the new long-term operating margin target that you guys announced for fiscal 2029, 45% excluding the transaction model. Just walk us through sort of the puts and takes in terms of how you get to there from the current kind of margin profile we have today. Yeah. I'm happy to. So if I think of it on an as-reported basis, this year we've guided to approximately 37%. The long-term margin is 41%. By the time we actually get to that stage, we'll be through all these business model transitions. So if I think about the new transaction model headwind, that will increase a little bit next year, and that's why you have that 400 basis points delta between the 41 and the 45. But if I think about the 41% as reported up margin, as we were building that target for the future, we've done well so far this year. Clearly demonstrated that we performed consistently, not just this year but for the past couple of years, from a top-line revenue growth perspective as well. And so, what we wanted to do is make sure that we set the margin up to be achievable under a range of growth scenarios. And so that was guiding principle number one as we set the margin. The second is, in terms of the sources of operating leverage, it comes from the go-to-market optimization that I talked about, which, as we are executing on that, I think that will continue to help us drive more to the bottom line. And then, of course, just the operating leverage that's inherent in the model, which we've been demonstrating over a period of time. And I think just putting those blocks together, as I said, over a range of growth scenarios, we should be able to achieve this margin target. Great. And then any questions from clients? Well, just yeah. Question about small businesses, but they probably still wanna leverage on all the AI tools that they're working hard on and develop. That would be very, very insightful. Is there a growth concept for different small size, growth for larger? A lot of the smaller customers will naturally benefit from these. So again, it will go back to the scale and complexity of the AI capabilities that they want to use. And if they are relatively rudimentary things, or if they are simple task automation exercises, those will be built into the products, and then we will naturally price appropriately for those, so they will get the benefit of those, and we'll capture the relevant amount of value, but as we were saying earlier, a lot of the benefit on this, if I think about AEC as an example, a lot of the benefit that even those small and medium customers will derive will be on collaborating across the ecosystem, and that they'll see when they, when they go onto Forma, which is the industry cloud for AEC, and when they're on Forma, they will get access not just to the capabilities in Forma, but thinking more broadly across the ecosystem and the benefits of, of leveraging the data models, the unified data stack, the collaboration capabilities, all of those things. And then depending on the kinds of workloads that they run, it'll either be included in the subscription, or they will have to pay on top if it's a more usage-based or more resource-intensive workload. Yeah. That's the final question. I think we have time. Yeah. Go ahead. You spoke a lot about the advantage of your 3D data, something which your competitors seemingly aren't lucky enough, like you guys, to have. Could you talk a little bit more about the challenges of using that data within your AI stack? And sort of more importantly, are any of your competitors even bothering to try to catch up, given the low ROIs that you were talking about, that they might be, you know, that they might earn from that? Yeah. I'd start by saying it's a lot more than luck. We've been thoughtful about it for some time. We are well ahead of our competitors in that regard. We do use the data internally. We disclose that to our customers, and make sure that we're always on the right side of being fully transparent with them as we leverage the data to build our own internal capabilities. Our competitors trying to catch up, I'm sure they are, but we feel very good about how we're positioned. Remember, there are very few customers, if any, who have sufficient data to build their own models. But they also don't wanna just dump their data with anybody 'cause they're worried about their IP. So it's sort of managing that, which is there's a bunch of data which is, you know, doing constrained automation or, you know, figuring out where all the toilets are in a hotel building. You know, nobody cares about that. And adding value to that and automating that, they don't care about. But they wanna make sure that their secret sauce is protected. So, for example, one of the things we're doing with our AI is when we inject the customer's data, we are tagging it so that if there is a design for an object that is unique to them, we are able to identify that and exclude it from the training data. And so that's on the sort of the back end of the process. And then on the front end, when we're delivering results to the customers, we also check to see whether the output is again aligned with a unique feature from a customer. And if it does, we'll exclude it from. So it's around automating what customers don't care about. You know, it's like the sort of spell-check function, you know, low-value stuff. But that's a big chunk of the data. And then it's making sure customers have security around the high-value stuff, which they really care about. On that note, I think we're a little bit behind time actually, but thank you so much. We wish you well on this transformation journey. Thank you. And also, looking forward to seeing what's ahead with Bernini and the world model in your own words, and what it looks like. Thanks for coming with us here. Thank you. Thanks for your attention as well. Let's have a great conference.