Confluent, Inc. (CFLT) Earnings Call Transcript & Summary

Please welcome our MC for the day, Confluent's Vice President of Investor Relations, Shane Xie.

Good afternoon, everyone. It's nice to see everyone in the room. I know it's been a busy week in San Francisco, so we appreciate you joining us both in person and on the webcast. Before we dive in, though, I'd like to direct your attention to a quick slide. This includes forward-looking statements management will make today, which are subject to risks and uncertainties as described in our most recent Form 10-K, and we undertake no obligation to update these statements after today's presentation, except as required by law. Now a few housekeeping items as you continue to review your favorite slide, number one, today's presentation agenda has been posted to our IR website under the Events section located at investors.confluent.io. Number two, after the management presentation today, we will post a slide deck immediately to the IR website as well. And finally, for those here in the room, we invite you to join us for a cocktail reception, hosted by our management team after the webcast portion of the program concludes. With that, let's get started. And it's my pleasure to welcome to the stage Confluent's Cofounder and CEO, Jay Kreps.

Hey, everyone. Thank you so much for joining us. Every quarter, we put together a little snippet about the business that we get to take out and Confluent has, I think, a very exciting story, but a lot of depth to it. And so the ability to get everybody together and do something that's a little bit of a deeper drop that goes through all the elements of the business, I think, is really exciting for us, and I hope for you guys as well. So over the course of the day, we're going to go through pretty much everything, products, go-to-market, some of the emerging AI use cases, partner and customer discussion and a financial overview of the business. But to start with, I want to really position Confluent and our platform in the wider ecosystem of data and around what's happening with AI, some of the emerging use cases. Any company that gets to any scale has some tailwind, something that's pushing it along. And data streaming has actually grown enormously over the last 10 years, the success of the open source, the adoption in all kinds of architectures across hundreds of thousands of companies. But what's really driving that? Why is this happening? Why is it happening now? What is this moving towards? What is the role that this is taking on for AI. And where is the puck going next? So to talk about the future, I want to actually rewind a little bit and talk about how we got here. Why are data systems the way they are and what's Confluent's rolling out. So if we go back to how software was really rolled out in organizations. It happened piece by piece. In every organization today, there's a bit of a sprawling mass, but we didn't get there all at once. We started with a set of relatively siloed UI-driven applications that would store their own little pile of data and look up bits of that to display the users. So this is the CRMs and HRIS and ERP, all the different systems that organizations adopted over the years and it was done bit by bit. And Ultimately, the data infrastructure back end for this kind of application, it's a storage system. It's a little pile of data that you need to look up the right bits and put it on a screen. And what emerged as companies had multiple systems like this, each of which had a snippet of data about the business was some need to see across to be able to do analysis that would cover multiple parts of the business that would talk about what was happening with our products that would happen -- cover what was happening with their sales, that would actually put it all together. And this is the rise of data warehousing. And again, the premise is very much storage-oriented. We're going to suck all the data out. At the end of the day, we're going to put it in one place. We're going to be able to run a set of reports. We're going to be able to run a set of analysis on it. And so this is kind of the early days of the emergence of software and companies. Over the last 10 years, what's happened is this is accelerating, right? The analytics ecosystem got more complicated, the operational applications got more complicated, there's ultimately a lot more piles of data spread around an organization. And this is in part a difference in scale and complexity but it's also in part a difference in kind, like the types of applications that are being built are different. They're not just the simple UI apps. They're not just siloed things that are augmenting some human-driven team. They're ultimately running much bigger parts of the business. They're actually taking on some of the action in companies. And this paradigm is really different. Ultimately, the early applications where the software used by people. You have an application, ultimately, the human types in the deals into the CRM and the CRM is able to show it back in different orders with different functionality around it, right? That's kind of the paradigm that we grew up with, with software. The action is happening periodically when the human shows up at the desk and logs in, right? It's not happening all the time. What's different in the modern world that software is taking on activity in the business is, first of all, the software is not going on a coffee break or going home at night, it's actually working all the time. The action is happening continuously. And these use cases are really much more end to end in a company. So if that sounds a little bit theoretical, let me give an example. So one of the most interesting industries to talk about is always the car industry, right? Because everybody likes cars, and there's lots of transformations and things happening. So a customer we worked with for a long time is BMW. And I think they're kind of representative of the larger trend in this area and in other businesses. And the work that we started doing with BMW was in their factories. It was about what's the actual data coming out of all these machines to be able to really instrument and drive efficiency and quality in the manufacturing process. And that led into work in inventory and supply. How are we actually keeping this Fed and how are we doing this globally in all our factories around the world. So it was really about the production of their products. It's not a UI-driven application. It's something that's happening continuously that's driving efficiency in the production of the product. But it didn't stop there. So as their business evolved, one of the changes they went through is from selling big shipments of cars to dealers to selling cars online, putting up a website where you actually go buy your car, and it goes and is built and you have a lifetime engagement with that customer. You can buy more features that come along with the service. There's an app that comes with the car. They handle the maintenance. It goes on even to the next car you might buy. So suddenly, there's a digital relationship that spans the life of the customer from purchase to the end. And it doesn't stop there, the car itself has a significant amount of software is sending back real-time data about what is happening, has a set of features built around that. And so why do I bring up this example? Because I think this is emblematic of what's happening in the use of software in companies. If you think about this, now end-to-end in this business from how you engage to purchase the product to how the product is actually manufactured to how it's delivered to you, to its operation through the lifetime of your ownership, there's a significant software component and all the parts of the business are now plugged together. As your car is being built, you're getting notified that this is happening. So all of what were disparate UI-driven systems in the back end of the business are now coming together into something that actually runs significant chunks of the company. And this is ultimately the trend that Confluent is built around. Our platform allows you to take all the streams of events of what's happening and share them across the company, act on them in real time, build applications that react and respond, really put all the parts together. And if you think over the last 10 years, this has been the drive. And it's ultimately, the story is about how companies are rebuilding themselves in software. And so what's happening now? When we think about what's happening in the tech industry, what's the big change? It's ultimately the rise of AI. And this is an incredible force to supercharge this transition. So when we think about what was a limiting factor in putting parts of the business into software, a big limiting factor was what you could express in hard-coded rules and logic. Ultimately, the limiting factor for what you could do with software was how hard it was to take some of these software problems and actually apply them at scale in a big nuance business. And so this whole trend of rebuilding parts of the business and software is supercharged by this. And I think we're entering a bit of a new age when we think about data and data platforms and the use of data. So in a sense, we're kind of moving from like a world of BI, like business intelligence to AI. We're moving from a world where being data-driven was hey, can I run a report that tells me what's happening in the business and make a good call and tell Joe to go act on it, right? Or can I have an analyst go run some analysis offline, and we're moving into a world where software systems are taking some of these actions. And this hasn't -- it's not like this is just starting. This has been happening, but is accelerating at a really amazing pace. And this is a really different world because we're no longer taking action in the large. It's no longer one big executive call. We're making decisions about pricing, about product delivery, about user experience, case by case, user-by-user, product by product, transaction by transaction is really a different world in terms of how you can actually run a business in this software-driven world. And why do I bring all this up? Well, I'm doing it because ultimately, the change in companies is reflected back in the infrastructure platform. So there's -- the infrastructure exists to serve what companies are trying to do. And the infrastructure platform that we have had for business intelligence has been data warehousing. And if you think about what this was, there was a set of functionality around first, bringing a store of all the data into one place. Second, being able to run processing inquiries on top of that to drive value, to drive insights. And second, a set of human oriented reporting and analysis use cases that drove early data warehousing. And this is ultimately the stack. This was the platform for BI. When we think about what we're building, we're obviously not attacking the same use cases. But the data streaming platform, what is that for? How does that compare I think there is an analogy, right? Ultimately, we are also trying to bring together all the data in a company, but we're not doing it as a batch dump of data at the end of the day. We're doing it as a continual stream of what's happening all the time. And we're bringing together a set of processing capabilities to act on that to drive new data to drive value and insights and decisions, but not purely for pointing back to users, but actually for driving application workloads. The things that run the business. And when we think about the next generation of workloads, I think they are that. I think they are the software systems that are operational in nature that are running parts of the company, and I think the adoption of AI is going to increase the reach of that. And so this is not to say that these 2 things compete. That's not what I'm saying. But in terms of the place where all the data comes together, that's ultimately that hub of what's happening in the company, that is moving into real time and that is moving into data streaming. And one of the things that encourages this movement is ultimately what's happening in the use of AI in the enterprise. So I worked at LinkedIn before founding Confluent. And one of the areas I was in was the machine learning team that would train up big models and a lot of the data infrastructure that we built was ultimately to power those use cases. And that was very much in the paradigm of classical machine learning. And the way that, that works is we would suck all the data into some big data lake, and we would have a bunch of really smart data scientists who would extract dozens or hundreds of features that would predict things, and then we would use that to build a custom bespoke model. And we would chip that off and integrate it in the product in some way to try and improve what we are building. So that was the paradigm for classical ML. It happened off-line in the data lake. It was ultimately a kind of priesthood of data scientists that were doing the work. That is actually not the paradigm for Generative AI. Why is it different? Well, it's different because the model building got 1,000 times harder, it's happening not on some bespoke enterprise data set, but on a large-scale data set of like everything on the Internet and well beyond, the number of parameters is trillions. And it's being done ultimately by a very small set of companies, right? It's OpenAI, Anthropic, meta, whomever, right? These are the people who are building the model. It's not being built for each problem. Instead, you're taking this prebuilt model maybe some fine-tuning, but by and large, you're combining it with your data as is. You actually have something that has relatively general intelligence capability. So what does this mean? In the infrastructure world, it means that the use of data of enterprise data is moving to the inference side. It's moving out of the training side. Enterprises are not shipping their data off to be part of the training run. By and large, what they're doing is combining it at run time. They're saying, "Hey, this is the context about the problem that I'm facing, Here's the reasoning model that can take that context and actually produce output, produce decisions. And this means that the use of data is moving into real time. It's moving into something that's fresh. It's moving to something that's part of the operational domain. And that's ultimately a really fundamental change. And we're going to talk about this in more depth. So one of the sessions we have today will talk about some of the emerging AI use cases, the patterns around RAG, applications, what we're seeing with AI agents that actually react to events happening in the business that respond that produce output. This has a natural affinity to the streaming world, and these use cases require real-time data. And this will be covered in more depth by Andrew, who will come a few sessions later. One of the implications of this change is these systems which are closing the loop, which are a software system that's actually directly making a decision, taking an action. When you close the loop in software, you increase the speed of how things operate. And what that means is that you go from actions kind of happening periodically to happening continuously. And the nature of the underlying software systems become something that is continuously operating. The software systems are not going to sleep at night. And this is part of this transition from a world of batch processing. At the end of the day, we run the job that sucks out all the data and turns it and get it into the right place, into a world where we're continuously processing data as it's produced. And the reason that, that's important is because the action on that data is now part of the operation of the business. It has to happen in the flow of the business. That's ultimately at the heart of this transition to streaming. This is what is driving Confluent. And I think this is a really fundamental change in businesses. So what does this mean in the larger world of analytics and in data? Well, there's a big change that's happening here as well. The analytics world has been dominated by these data warehouses, which traditionally, we're kind of a big box that you put all your data in and every use case for data would kind of run through some query interface that would charge a bit of a toll to get at that data. But that world is opening up. So in -- there's emerging sets of standards around iceberg or Delta, which take all your data and open it up in cloud object storage and make it available to any system in the enterprise. And the entire world of analytics is orienting around these. And this is a fundamental change in that business because now your data is much more broadly usable. The set of systems that can access to it is much more open. All the work that you do to get data into the right format to be usable is now reusable across all of these. You're no longer stuck in 1 box. You can now use a whole ecosystem of tools. And this is a really powerful change. And so at Confluent as we looked at this, we felt like, hey, this is an incredible opportunity. We've always had a nice connection into the analytical world, delivering data. Kind of our point of presence has been in these operational applications, connecting them, but there's an opportunity to bring these things together. If Confluent is connecting these operational apps around streams, and the analytical world is increasingly connected around these big stored tables of historical data. We can tie them together. And we could start to populate these tables continuously in real time off of real-time streams. And that's the work that we've done with a feature called table flow. This is something that we announced last year that's going to GA imminently, we're extremely excited about this capability. And what it enables is taking any of the data that's in Confluent, which we were already maintaining in cloud object storage and opening it up in a way that is directly accessible in the analytics world that can power all the use cases there. And this is a really important trend because there's a difference between the analytics world and the operational is kind of going away. These AI systems kind of blurred the line. They're doing complex data processing, but they're doing it in the context of serving the business like an operational application is actually somewhere in between. And so the ability to have things that are in real time that are arriving quickly that are happening with the business, but to have rich analytical capabilities is a really big deal. And we're lucky in this to have really great partners. So one of the companies that has built itself around an open data architecture with Lake houses and that has really pioneered the ecosystem around AI and use cases there is Databricks. And we recently announced a partnership with Databricks to really take this out to our customers together. And this is I think, incredibly well received. We've heard really great feedback from our customers. They are using Confluent. They have these real-time streams. They're using Databricks. They're very excited about this world of open data and be able to connect to the 2 and get reliable data to these systems is a really powerful thing for them. And this is a deep product integration. We'll go into some of the details in a bit, but it opens up all of the data into their delta format. It registers at all with the Unity catalog. It really connects into their whole ecosystem of tools. And so this kind of bringing together of operational and analytical use cases, this is a very powerful thing for customers and is directly where the puck is going around AI applications. And to talk about it, I'm really pleased to be able to invite up to the stage Databricks CEO, Ali Ghodsi. Ali, do you want to come up and join.

Are we going to sit that far part.

Yes, I guess so.

Awesome. Thanks guys.

So Ali and I have known each other for quite a while, we both started companies around -- well, I guess, sort of open source projects and then companies around the open source project, not that far apart.

Yes, simpler journey.

And we -- that's right. We both apparently go to the same hairstylist.

Exactly.

Who's not very good. And so -- and for a long time, we talked -- we're kind of adjacent in the data world and had talked that there was something we should do together, but one wasn't really something. And I think one of the things that's exciting now is that, hey, there is something that's really useful and meaningful to customers. And so I think the goal was to talk about that.

Yes. Super excited about it. Yes, this is going to be awesome. It's going to accelerate both of our businesses.

Yes. Awesome. So maybe we can start and talk just a little bit about the motivation for the partnership. I'd love to hear kind of what you guys are seeing as some of the challenges the enterprises you work with are facing when they try to really operationalize AI -- like there's a lot of excitement, but there's some journey from excitement to something that's running in production that's actually doing something of value for the business. What are you guys seeing when you work with customers?

Yes. I mean, CEOs are yelling to their staff saying, we need AI. You need to get the AI now in production, we need to go faster, faster, faster. So people are stressed out in the industry. And the projects are behind, and they're not getting them into production. And people don't want to publicly talk about that because everybody want to say, "Hey, I succeeded. I have this many projects in production in AI. We have agents, the agents are awesome. But the challenge is really, honestly, most of the problems come back to the data because this whole AI revolution happened thanks to just using more and more and more data to train bigger and bigger, smarter models. So the issue is just getting access to their data is a huge problem. The data siloed in many different places and you need the context. If you don't have like the agent, we'll do will hallucinate and do weird things, when it doesn't have full context. If you just get it the right context, but that context you have it in the enterprise, it's just maybe, oh, that was in the SaaS app, Oh, that's locked away in that operational application, that's here or that's there. . So it's really getting access to all these different silo data that's sitting in different places. And oftentimes, in applications, oftentimes it is operational. So that's number one. The second is can you get it -- how timely can you get it? Like are you running a batch job and you're getting it a week later? That's not very helpful. If you want an agent that's customer-facing, you want to be able to get things right, you need the data to be timely. So this is a second problem that's people are struggling with. And if they can't figure that out, that's the second thing. The third is securing it, making sure that it's privacy preserving, making sure that they have governance in place. So all of these are huge challenges, and that's what's slowing down the project in the industry. So this is the challenge. They want to do agents. They want to get AI. They want to deliver all these great news, but they're getting stuck in getting access to all the siloed data in real time.

Yes.

It's similar to what you're seeing.

Yes. It's very similar to what we're seeing. That's right. Yes. And I guess the other thing that I was touching on earlier, architecturally you guys have built very differently than a classical data warehouse. Ultimately, you've created something that is kind of an open data system and kind of an open ecosystem. I think it has helped you to maybe bring a richer set of capabilities in the AI world. I'd love to hear how you guys thought about that, how that's evolved. Does that matter to customers? How are they thinking about this area? You guys have really pioneered a very different way of doing this.

Yes, but it's very similar to you guys, right? Enterprises, one of the things that slows them down is being locked in. And data lock in is huge. It's been for 40 years to get locked into 1 vendor to stop innovating. They still jack up the prices and you can't get off of it. And the migration take forever, many, many years. So the question was, what if instead of saying that, hey, here's Databricks an awesome company, give your data to us, we say, let's not do that. Why don't you own your own data and store it in a standardized open source format and let the best engines or systems or whatever you want in the ecosystem, they should access that one copy kind of becomes like the USB standard of data. And that way we can unlock a lot of innovation. People can move faster, and they don't need to worry about moving off of Databricks or some other vendor or -- so that's really the thinking and it's really resonating.

And what do you think has worked for you guys despite being more open, which would seem like, okay, we have less control and less lock in. You guys have been the most successful in this space. So like what enables those 2 things to happen together?

Well, I think we started it. I mean you guys did it on the real time, the kind of central nervous system of all the data flows you guys were -- Kafka did that. On the sort of more data warehouse saying side, we kind of did that. We were first to do it. No one else was really doing it. Now you will hear more about it, but that's how it works. Once you start this trend, you did it with Kafka, right? Then others will see, well, yes, it's resonating with the market. People want that open ecosystem. So in the more sort of competitors will come to space and say, well, there's like a competition for those open source projects. So for a while, there was actually a little bit of a sprawl in the open source ecosystem for, okay, who's going to be that open source standard. And we have proposed Delta, Data Lake as one of those but immediately 2 others, there was iceberg, and there's a smaller 1 called Hudi. So that actually led to us acquiring this company called Tabular to unify the formats to remove some of the interoperability so like kind of put this question to bed and say, look, this is going to going to be one open source standard because we would be really bad off. If it turned out that, okay, we do have an open standard and everybody is -- but it's like 8 different formats. So it's really working diligently towards unifying that ecosystem making it simple. And in a way, we're also cannibalizing ourselves. We're seeing let the best engine work on the open source format. Well, there could be some other engine than Databricks because we're not locking it into Databricks either. So we're exposing ourselves a little bit there, too, but I think it's better for the industry.

Yes, I think you guys have done a remarkable job of bringing that out to the market early on, giving you name and then evolving with the technology world as the ideas around that continue.

I also -- I'm curious if you're seeing the same thing, which is that most enterprises in the world are like, "Look, I want to know that I'm not blocking myself in. So there's an open source standard with Kafka. But I don't want to be in the business of running it myself. I don't want to -- I don't want to have [indiscernible] at 4:00 a.m. and if this thing goes down. So if you do it for me, as long as I'm not locking because it's an open standard, happy to pay you to run it for me as a SaaS service and I'll come to you.

Yes. Yes. I think that's exactly right. Like the observation I've had is the big companies are built over time. So they end up having some amount of complexity and sprawl and they need something they can kind of count on. And these open standards function is that there's just a much more interest in embracing that and really betting heavily on it. I think it's been probably a tailwind for you guys certainly for us as well. Well, as we started to think about this partnership, one of the things that made it easy was -- we had some really good customers that are already using the 2 products and using 2 products together. And that often makes things kind of natural. We were feeding data already into Databricks in 2 or 3 different ways. This table flow stuff makes it even better, so the governance capabilities and things that you guys have built now give us a much better way of propagating all the metadata and control information, so it can fit well. But we've seen this now across customers like Michelin and Meesho and E.ON and Accenture. And so it's been nice for us to have that validation. I don't know if you heard anything with the announcement, but at least our customers were very excited about the 2 things working together, and some of them were kind of like, well, we took you guys so long. We were already using them together.

Yes. No, I think it's super exciting, especially the ones -- the ones you mentioned, you mentioned Michelin. Michelin is an interesting one, right, because it's basically supply chain management for them and if you want to optimize your supply chain, especially now you can do it with agents and AI but you need real-time data. And a lot of that operational data is stuck away in like databases. In Databricks, we are in the analytical side of the house. So things tend to -- you have more time to process them. But here, they need real-time information to optimize that supply chain. So it's like made a lot of sense, it use the central nervous system, Kafka into Tableflow and then comes into Delta with Unity catalog and then can do AI on it and they can just -- that's one I like.

And they're an incredible Databricks shop. I know when I met with them you guys have done amazing things.

Meesho is cool. I mean, I don't know if you mentioned that one. Meesho is a big sort of in one of the e-commerce sites in India, hundreds of millions of users are using those. There, it's more about personalization, which is also AI and ML, but how can you get that? Again, you need all that data that's sitting tucked away in these operational databases. How do we get that to customize it for our customers so that we know exactly what do they want to see when they come to that landing page. And you don't want to recommend them what they just bought. That happens to like you go to these things and they're like, oh, have you thought about buying this? It's like I just bought it, right? So real time is really important there. So that's one. E.ON in Europe. That's a big German sort of -- they have a lot of -- actually heavy manufacturing in general and manufacturing. If you have real-time data on the equipment that you have out there, you can go replace it faster, you can prevent accidents, you can lower cost. You can actually -- it's both better for the environment, it's better for the people involve your employees. So that's another one, but real-time data ends up being super important. And then we're very excited about Accenture. We work closely. I know it's -- they work with both of us. And they're doing a lot of AI customization, but they need that data. They need the real-time data. So I think that's going to be an exciting one because Accenture helps so many other enterprises out there.

Yes, I totally agree, totally agree.

So yes, I think that's -- this is just the beginning. Hopefully, many, many, many more.

Yes, that's right. That's right. I think as we looked at it, there was a pretty incredible overlap of different customers, even the ones that aren't using us together yet, maybe there's an opportunity to expand there. So it's exciting stuff. Well, I -- we talked a little bit about the partnership. I'd love to hear just kind of from your point of view, and I can talk a little bit from our point of view, what does it bring to the table? Like what's -- any of these...

No pun intended?

Yes, that's right. That's right. Any of these things take some effort to put into it. Obviously, from our point of view, we really saw you guys as leading in the revolution around Lake Houses and Open Data really leading around AI use cases and the capabilities there and then just great overlap. And so our ability to kind of open up data that we saw that as kind of part of our mission is get data all over the company. But from your point of view, what's -- what's the value here? Where do you see this being kind of worth spending time on?

Yes. I mean what we've seen over the years, many, many years, like is that one of the most -- data company, we suck data out of everything. We work with all kinds of data unstructured, structured. One of the most important data sources for us has always been Kafka. It's like again and again and again, people have Kafka on-premise in the cloud. It's everywhere. It's in every enterprise. But how you get that data from Kafka and getting it into an analytical system like Databricks has always been kind of painful. I think we've kind of left it to the customers to figure it out on their own. And it's challenging, and they have to do a lot of things to make that work. So just making that super seamless is a no-brainer. Like every one of them, they're going to leverage it. So simply getting with table flow, getting the data that's in Kafka immediately into these tables so that you can do analytics on them. So it just automatically now will appear as Delta tables and their Databricks and you can govern it with Unity Catalog. I think it just eliminates so much friction for our customers. And I think it will also unlock a lot of use cases where they weren't thinking about because I have to go through the pain of connecting Databricks and Kafka and Confluent. It's like, no, it's right there. It's like, oh, you were -- any data you have in these topics, it will be in Tableflow. So I think it's game changing. It's better for everyone. I think this is the way forward for any real-time data or if you want to combine operational with analytical data, I think it's going to unlock a lot of value. And I'm excited also about the Agentic use cases because -- that's the one that everybody is...

Yes. That must be a transition for you guys where your use cases are kind of going from analyze the business at the end of the day to really run chunks of things in agents and with software systems just working faster.

Yes. I liked your slide that you had earlier. It's really like went from BI to AI, where business intelligence was like backwards looking to you had on your slides then on the right-hand side that with the AI, it's more forward looking. All the enterprises are kind of doing this transition now -- but they need data, and they need the data in real time, agents to frankly get the accuracy up because right now in the industry, the accuracy is not -- they hallucinate, they get things wrong. So it's crucial to give them the context they need to get the predictions right.

That's right. That's right. Well, yes, that's the product side of it, of course, we're also working on really taking it out to market. So what does that look like for you guys? How do you operationalize these partnerships? What's involved in making something successful with customers?

Yes. I mean it's in every market, in every country, having our go-to-market teams working closely together. But honestly, there's such a pull in demand from the customers. They're so excited to be able to combine their operational with their analytical data.

Yes, I agree that this one is not the most complicated, especially in these cases where either customers locked up in some legacy analytics platform that wants to move to something more open or where they're already using the 2 products, so it's just a question of connecting them together for the first time. So I think it's really easy.

Yes. I think there's so much demand, this pull from the market. They want to combine these things. It's frankly what has taken us so long. I don't know.

Yes. Well, that's awesome. So I'm really excited about the partnership. I think there's big things ahead and really excited where we take it from here.

Likewise, super excited. This is going to be awesome, especially being able to get these analytical workloads with operational open ecosystem. I think it will be very exciting.

Awesome. Thank you for joining.

All right. So that's a little bit of a taste of where I think things are going in the data streaming world. Obviously, we're going to get into every aspect of this throughout the rest of the presentation today. How we're taking it out to market, what are the specific products that we're building. But before getting into it, I want to talk a little bit about the team. So we've -- Confluent has always had a strong management team that evolves with the company. It's grown with us as we've gotten to where we are today. And I want to give one update on this. So after 5 years and going with us from early go-to-market development to over a $1 billion run rate Erica Schultz, who is with us today, has shared her intention to retire. And so she's going to be staying on while we conduct a search and onboard the new person and help us through the transition. But I actually just want to say, thank you, Erica. You've been an awesome partner. Erica started when we were just very early in this very funky office in Palo Alto. And I'm sure it was like, "Oh my God, what have I walked into as she joined and has helped us do really big things from developing the early go-to-market model to the transitioning to have a cloud product to going through a whole transformation around consumption and really orienting around that. And each of those legs has been really successful. So thank you, you're going to hear more from Erica as we do the deep dive into the go-to-market. But I just want to say thank you for everything. And with that, I'm going to conclude my section. I want to give a couple of takeaways. So first of all, I think data streaming is key to this whole process of closing the loop in software and key to this next generation of AI use cases that are driving that. Secondly, the platform that's emerging around streaming, we think, has the opportunity to be the most strategic platform for data in a modern enterprise. This is going to be where everything happens, where the data flows, where you have a view of everything happening in an organization, the nexus for a lot of the processing and applications that act to react on that. And finally, I think we're uniquely positioned to do this. This is what Confluent was founded to do. And we're at a moment in time that's uniquely favorable for that vision. As software systems are taking on more capabilities, the importance of data is rising. So I can't be more excited about what's ahead. And I couldn't be more excited about what we're building to go after the opportunity. And to talk about that, I want to welcome to the stage, Shaun, our Chief Product Officer; who's going to walk through a little bit about the products and capabilities that we're building. So Shaun?

All right. Thank you, Jay, and thank you to all of you for joining us here today. Now as you just heard Jay talk about, organizations today are facing a monumental challenge, a tangled, difficult, complicated infrastructure layer that makes it really hard for them to put their data to work. How do you keep up with the pace of modern AI on top of a messy and unbelievable architecture. Well, the Confluent data streaming platform is the key to unlocking that complexity. It provides a unified real-time foundation for data that enables organizations to simplify their data infrastructure, unlock real-time insights and deliver on the full power of AI. So I'd like to step back for a little bit and talk about our evolving product strategy, how we help customers solve their data problems and why the data streaming platform is the foundation for the modern AI data stack. Jay talked to you a little bit about the data mass, and he described how the mass is actually 2 separate masses. There's 1 in the operational estate with the online application to working together to deliver experiences. And there's a whole separate mass over in the analytical estate where ELT and ETL pipelines are sucking data out of the operational applications and then piecing it back together for analysis. Now we also talked about the fact that traditional AI and ML existed basically entirely in the analytical estate. And that's because traditional ML models were created sorry, may go back on traditional -- quicker back one. Traditional AI/ML models were created over a period of months, usually using your proprietary data from your organization. And then you would deploy relatively simple statistical models that would then go and make predictions. But in the era of Gen AI, things look completely different. The foundational models are pretrained by several major vendors. And so when you're trying to create a chatbot, you're trying to do task automation or you want to deliver an AI agent, it's not about training at all. It's actually about feeding that AI experience with data. You need an unbroken continuous loop of data feeding information from the operational applications to the AI and then from the AI back to the operational applications to update information or to take action. They actually need that continuous unbroken loop of data to operate at unprecedented speed and scale. It has to be fast because you don't want your AI agent taking action on information that's out of date and no longer useful. It has to operate at massive scale because you have to meet the demands of data flow in the enterprise, which continues to explode, but you also have to do that cost effectively because you want to be able to take these supersmart LLMs and apply them to all of your business problems, but to be able to do so and extract an ROI. Now any break in that continuous data flow if latency were to spike or if the context is broken because of some upstream data processing challenges, you're going to degrade the AI experience. You can easily turn what is a super smart agent into a chaos machine that makes terrible decisions and damages customer experiences. So if enterprises that are looking to tap into the power of GenAI, they really have to rethink their data architectures. It's not about static, clunky, point-to-point, transforms anymore or pipeline. You have to treat your data like it's a living, breathing network of streams that's dynamic, real-time and reliable. When you think about it, that's actually pretty obvious. GenAI has almost nothing in common with the traditional analytical data stack built around reports and batch decision-making. GenAI is not dashboards. GenAI is real-time experiences. These aren't reports, they are actually applications. These GenAI apps are causing a Cambrian explosion of applications in the enterprise and these applications can be built more quickly and are more powerful than any of us could have ever imagined 2 years ago. And most of these AI projects have been stuck in the pilot phase not because it's hard to build one of these experiences, but it's because it's hard to deliver it at reliably, safely, at scale because you can't feed it with reliable, safe, real-time data. So put simply, every AI problem is actually a data problem. And that's what we're solving for. How is Confluent solving for the continuous seamless movement of data wherever it's needed to power any use case including GenAI, while our product strategy is based around 4 key pillars: streaming, connecting, processing and governing. Let's dive into the first product pillar streaming. The amount of data in the world continues to explode. In 2020, about 64 zetabytes of data was produced globally. But in 2025, that number is predicted to almost triple to 182 zettabytes of data. But honestly, all of that data is nothing more than noise unless organizations can understand, manage and put it to work. And today, organizations rely on a bunch of different technologies to manage and move their data. It goes all the way from massive Cron Jobs, FTP files, ETL and ELT, enterprise service buses, change data capture, it just goes on and on. And that fragmented estate of piecemeal solutions represents the total addressable market for streaming. Kafka revolutionized the way that organizations mobilize their data and it pioneered the concept of streaming. And Kafka is essentially a powerful unified system for moving, managing and transmitting large volumes of data in real time across both the operational and analytical estates. I'd like to imagine it as a bit like a global high-speed telecommunications network. It's literally always there whenever it's needed to push data from where it was created to wherever it is needed instantly to power a new customer experience, deliver some new insights drive some new AI agent. But you actually just heard me argue that the goal here is to change how 182 zettabytes of data is moved and use. Can that goal be achieved with a one-size-fits-all approach. Kafka as the monolithic foundation. Well, I mean, obviously, that's actually a little bit of a rhetorical question because I would argue that streaming is the right way to move all of that data, but the data varies in its value and its volume in the amount of it and how important it is. So you don't just need streams to move the data. You need the right streaming offer to meet the needs of that specific type of data. And that's where we have been building Kora. With Kora, we took the Kafka protocol and behind it, we put a global serverless, elastic modern cloud service. And the result is a service that comes in many different configurations that deliver exactly the right features, performance and cost to meet the needs of any of those specific types of data workloads. Now today, that shows up to our customers as a bunch of different clusters that they can choose from. Each one of them exist to be better, faster, more capable and more cost effective for that type of workload than any other approach. So for example, we recently introduced freight. Freight is designed for fire hose type workloads that aren't particularly latency sensitive, think logging and telemetry. And that offer delivers unbeatable volume performance for that type of data. So what you need to hear is that these offerings aren't alternatives or substitutes for each other. Each one of them is purpose-built for subsegments of the 182 zettabytes that I talked about, and each one of them unlocks streaming for more use cases, more of that data. So customers don't just pick one, they mix and match them to the needs of their specific data workloads, and then they mesh them together to make them operate as one continuous end-to-end system. So streaming is really foundational and it's obvious that others in the world will be looking for a piece of this market. And that's especially true because now it's kind of obvious that streaming is a better substitute for many things that are currently done using batch. You get a more flexible, more real-time foundation that is more built for the needs of modern business, including AI. But those streaming competitors typically are either a micro batch architecture that's dressed up to look like streaming or they're a Kafka like reimplementation that isn't as broad or as deep as the real thing. And so when customers give those streaming offerings a try, they generally really quickly realize that those other vendors can't bring the same scale, reliability, performance and cost effectiveness everywhere that streaming is needed, which is on-prem, in the cloud and even between the clouds. So that's why enterprises like Automation Anywhere, Colliers International, SecurityScorecard, Grafana Labs, L'Oreal, iFood, Character AI and many more organizations of every shape and every size choose Confluent over Kafka-like offerings. So for instance, SecurityScorecard switched from MSK to Confluent, changing over an infrastructure that was really finicky and required constant fine-tuning and knob-turning for a seamless operational experience. In doing so, they reduced the operational burden by over 80% and save money, $125,000 to be clear. Grafana Labs chose Confluent's WarpStream for the cost-effective, scalable, high-performance offering across multiple availability zones and L'Oreal moved away from Azure Event Hubs after they experience scalability and reliability challenges. And they didn't just replace their streaming platform with our technology. They're also building out a whole new next-generation architecture that includes stream processing and governance as well. And streaming is just the beginning. Let me talk a little bit about some of our other product pillars. Let's start with Connect. The customer's organization today, they use a patchwork of different tools to solve for their data connectivity needs. In the operational estate tools like MuleSoft or Informatica come with a bunch of data connectors for application integration. But then they usually actually do it all over again over in the analytical estate where tools like Airbyte, Stich or Fivetran come with their own sets of data connectors to those same systems which are used for ETL and batch extraction? And that duplication is our opportunity as we deliver a unified solution across both the operational and analytical estates. As the number of systems and applications multipliers, the demand for real-time data increases, the data connectivity market represents a multiple billion dollar opportunity for us as a business. And that's why we built the industry's most comprehensive ecosystem of zero-code connectors. They span operational systems, analytical systems and SaaS platforms. We have over 120 different connectors more than 70 of which are fully managed, and they make it super easy to pull data into Confluent or push data out. And you don't even need to know anything about Kafka. It's all zero click -- sorry, zero code point and click. What we're doing is we're eliminating the complexity. We're making it easier than ever to work with data in real time. And today, over 50% of Confluent's customers already use Connect. So Connect is our second most adopted product beyond Kafka itself. But to keep pace with the explosion of new data systems and applications, we're also working really closely with our technology partners and helping them build and evolve their own integrations with Confluent. In fact, our new Connect with Confluent program already has over 50 new integrations that span the data ecosystem, including platforms such as AWS Lambda, Google BigQuery, Elastic, MongoDB and SAP. Now those connect with Confluent integrations actually appear directly in the user interface of those partner applications. So it makes it super simple to move data with Confluent. We've already made some pretty remarkable progress on all of this. By the end of 2024, we had over 900 customers taking advantage of the Connect with Confluent integrations and we're passing over 160 terabytes of data every day through the overall portfolio. And as the demand for generic applications continues to increase, we're also expanding our Connect integration strategy to plug directly into the modern AI data stack. So we have partnerships with major vector database vendors, including MongoDB, Elastic and Snowflake and partnerships with GenAI ML Ops engines for RAG, including Vectara; and emerging collaborations with LangChain and Anthropic. Now those integrations make it seamless and easy for customers to feed their AI tools with real-time data and they get -- they get next level performance and scale from those AI applications they can build quickly. But connecting with data is just the first step. What really matters is making it rich and useful and that's where our stream processing strategy with Flink comes in. For AI to be really effective, businesses need to know what is happening right now, not yesterday, not an hour ago in real time. For example, if you're in airline, you need to know which seats are available at what price points, the weather at the destination and a frequent flyer status to you in order to give them a new itinerary or to make them a personalized offer. But it's not just about knowing information in real time. You also need to have historical context in real time, too. So even though your customer preferences or your purchasing history aren't created in real time, you need them in real time at the moment the AI agent makes the decision. But organizations really struggle to bring all of their data together and to bear on all of their problems because it's unbelievably siloed. In the operational estate, the applications do tend to move data between each other but they do that using a ton of custom code or application integration tools like Informatica or TIBCO. And what that means is that none of the operational applications can see all of the data. And if you want to add new data sets, you have to spend a ton of manual effort and cost. Now over in the analytical estate, you do have a much broader view of all of the data in the enterprise. But the data is established by using ETL or ELT to suck the data out of the operational estate and then move it in large batches and transform it using tools like DBT or Matillion, or Spark. And what that means is that the data in the analytical estate is only as recent as the slowest of the inbound batch files and it's only as good as the worst of the files that landed in the Data Lake. So how can organizations get at all of the data, but do it reliably and do it quickly. Well, Flink, together with Kafka is actually bringing the very best of the analytical estate into the operational estate. And in doing so, it's removing the need for a lot of those fragmented processing tools. And that's what I think has seen such rapid adoption. Really innovative companies, companies like Netflix, Uber, LinkedIn, Stripe, Shopify, Disney+ and many more have adopted Flink as a way to break down those silos and seamlessly move data anywhere it's needed across their operational and analytical estates. Really important use cases like ad targeting, personalized recommendations, dynamic pricing, real-time logistics, they all require high throughput, low latency processing of data at scale from across the organization. And Flink has made those types of workloads, not just possible but approachable and affordable to. Now Flink is really powerful for a few reasons. It's one unified engine that supports processing, both in continuous streams and in batch. It supports complex applications and transformations by seamlessly maintaining state. It can operate at high scale with low latency and with low cost. And that means that developers can work with their data and shape their data as it is being produced. They can do that in batch or in streaming. They can reduce the need for duplication of that processing logic downstream and they can drive greater reuse of this real-time data across the operational and analytical domains. What you need to hear is that previously, things that could only be done in batch can now be done in streaming and the result is not just more valuable, it's actually less expensive too. That's why Flink is a cornerstone of our product strategy. Over the past year, we've had remarkable progress. We've introduced Flink as part of our on-prem offering, and there's a fully managed scale to 0 service in our cloud, removing all of the operational burden from Flink and then in customers focus on their use cases, what they're building rather than working with data. For instance, Zazzle, a global marketplace for custom designs is using cutting-edge technology to connect creators, makers and customers. Now they're powering real-time personalization of designs at scale. And they adopted Confluent Cloud's Flink to optimize their largest data pipeline. And the end result was reduced storage costs, improved efficiency and better recommendations to drive revenue. And on top of that solid foundation, they're now looking to use Flink to drive even greater personalization. And we're still actually just getting started with Link. We want to bring the power of Link to every use case everywhere and into the hands of every developer. So we're continuing to make it better. First, we recognize that not every developer wants to work with SQL. So we're making it super easy to use Flink from Python and from Java. Second, we've already made machine learning models, first-class citizens in Flink, which means that any Flink workflow can trivially invoke your own corporate model or major models from OpenAI, Azure OpenAI, AWS Bedrock, Google, Vertex and many more. And that means developers can now be reliable, scalable AI applications faster than you could ever have imagined. And with Federated Search, we're making it easier to bring in additional context from across the ecosystem, including data from big query, Snowflake, Databricks, and MongoDB so that the results are richer and there are fewer hallucinations or errors. So how does Flink actually help bring AI to life? Well, let's take a look at that in the context of Airy. Airy makes AI agents that can work with real-time data, think, tell me about my stores, sales by category and alert me when I'm projected to exceed my projections -- sorry my goals. Now these AI agents don't just answer questions, they monitor, they automate and they evolve into complete business workflows. And Airy powers their data pipeline using Confluent. Connectors pull fresh data from sources like MongoDB, Postgres and Salesforce. Then they use Flink AI model inference to calculate embeddings for unstructured data, and then they can store that information into their vector databases, including MongoDB, and then they use Flink for all sorts of enrichment activities, filtering, joining, aggregating to bring the very best data at inference time to all those smart AI models. Now the bottom line is that Confluent is bridging real-time streaming data and AI. We're making real-time intelligence real and we're slashing latency, cost and overhead. So let's talk about our fourth product pillar, governance. Now to be clear, governance is not optional. In the world of real-time data, trust is absolutely everything. If data is unreliable in real time, then downstream systems will break and they can do so very quickly. They can end up impacting your customers' experience, they could degrade your operations. They could even harm your brand. So governance is about ensuring that your data is accurate, it's reliable, it's secure and it's always ready for action. But you might wonder how can you enable organizations to manage, govern, trust, and reuse data when that data is always on the move in real time. After that, it starts with data contracts. Data contracts are really simple, they are an agreement between the producers of data and the consumers of that data. So a producer can define the shape of the data, they can define the different fields of the data, they can add business metadata that describes how that data fits into the broader data ecosystem. They can define an SLA for that data and they can define data quality rules that describe exactly how their data will be presented and how it will conform with your broader organizational standards. And with that contract in place, consumers can leverage that data and do so safely. But we've got to be honest, the world is constantly changing. So these contracts are reliable, but they're not static. They are evolvable. Upstream producers can change the structure of the data that they are consuming and downstream consumers don't even need to know that anything has changed. They won't be affected unless they want to take advantage of some of the new information. So for example, if I'm an upstream producer, and I decide to add a field or decide to rearrange those fields or add new business, metadata. I'm not going to affect any downstream application, analytical report or AI agent that's using that data unless those consumers are changed specifically to take advantage of the new data. And that means that data contracts led teams move quickly, but do so safely and they can build AI apps with fast, trustworthy real-time data. So you just heard me say that governance is ultimately the foundation for data reuse, and data reuse is the foundation for agility. Data producers share high-quality data, data consumers can find access and leverage data that they can trust. And there's lots of different people that can take advantage of that. Application developers can build new event-driven experiences or applications using it. A data scientist can build new AI, ML models or run inference using that data. Data analysts can take advantage of it to get new insights or power new reports. And the data platform team presents unbeatable capabilities to all of those constituencies preloaded with the data that they need to be successful. So I couldn't be more excited about all of the capabilities we just walked you through, but we do have one more important thing to talk about. When I kicked off my section, you heard me talk about the fact that the giant mass is 2 separate masses, 1 in the operational estate and 1 in the analytical estate. Well, the good news is that streaming has become the de facto standard for data movement in the operational estate. Thanks to innovation pioneered by Kafka, streaming is the universal interchange mechanism that's used between applications, data systems, databases and platforms all seamlessly. But as Jay mentioned earlier, when you shift over to the analytical estate, it's been a very different story, and it's been filled with inefficiencies. When I introduced Flink earlier in this talk, you heard me say that a lot of the data that arrives in the analytical estate is actually from the operational estate and has pulled into a bunch of ELT or ETL pipelines that suck it out of the source systems and they then land it in the Data Lake as a bunch of files. Now to take those raw ungoverned files and actually make them useful for some specific use case, teams typically load them, they cleanse them, they process them and they reprocess them until they get something that they can use. Now that approach of taking these raw files, duplicating the files, processing them and reworking with them over and over again has a ton of consequences. You end up with data that is stale and unreliable. You have skyrocketing compute costs because you're constantly reprocessing the same data. You have data quality and governance challenges and you're actually mounting up piles of technical debt from one of this black box batch processing code. But you heard me say earlier that the DSP provides a unified foundation so that organizations can have rich, reliable data everywhere they need it. And in fact, in many cases, it's actually Kafka that's feeding all of the data that is arriving in the analytical estate. But analytical state typically thinks about the world in terms of tables -- and until recently, those tables have been tightly coupled to the processing engine that was working with them. So if I had a table in Redshift, I couldn't take that exact same table and use it in Trino or Snowflake, for example. And so this mismatch in the operational domain of streaming, the analytical domain of tables and the tight coupling between processing engines and tables has meant that even though the data was coming from Kafka, the only way to get it into the analytical estate was to dump it is files and then suck it in through imports. And in the process, we lost the governance, the speed and the reliability advantages that we already had from the streaming domain. And you heard Jay talk about the fact that over the past few years, we've seen the emergence of these open table formats, Iceberg and Delta. And what they mean is that they mean that the table can be independent of the processing engine that's taking advantage of it. And this has absolutely unlocked our vision to make data connected and accessible everywhere it's needed across both estates. And that's why we've got Table Flow. Table Flow ultimately evolves Kafka storage so that the data in a Kafka topic can also be used as an Iceberg or Delta table. That means the exact same real-time, reliable, reusable data that is in the Kafka topic, appears in the analytical tool set with no work. You can use it in your data lakes, in your data warehouses in all of your BI tools. And as a result, no more clunky ELT, ETL or batches, no more unreliable data, no more endless data prep trying to make the data work and make the data makes sense. You get effortless data flow from the operational estate into the data warehouse into the lake ecosystem beyond. You save time, you cut costs and let teams actually focus on what matters, building new capabilities rather than data wrangling. So that's the DSP. It's a complete platform. It unlocks the value of data for applications, analytics and for AI. Now I know that it could seem actually super complicated, but it actually dramatically simplifies the lives of our customers dramatically. And even if you don't believe me, we'd like to show you. Later on, Mike Agnich is going to show you how you can use this platform to build an AI agent in less than 5 minutes. And you'll be able to do it even if you don't consider yourself tactical, we might even do some audience participation, anybody happy to come up and do it with Mike? Okay. All right. All right. So the DSP makes life so much simpler for our customers because the existing data infrastructure is so messy. It's packed with a ton of different tools. You have some tools that are solving individual problems. For example, message queues or governance tools like Collibra, that are point solutions. You have some solutions that are a little bit broader. You have tools like, Informatica or Fivetran that solve for some portion of the connect process and govern problems. But whether or not these solutions are wide or they are narrow, none of them is complete. And none of them can cross the chasm between the operational and analytical domain to deliver a unified foundation. So organizations are stuck with this chaotic tangled data infrastructure. They have a mishmash of overlapping concerns and technologies, fractured ownership, incomplete governance and repetitive processing. Now with that setup, enterprises are piling on technical debt. Innovation is super hard, and they actually end up paying for the same capabilities over and over again. That is where we come in. We're not just a piece of this puzzle, we are changing the game from the ground up. Our mission has always been to enable seamless data movement everywhere, streaming unified connectivity, governance and processing. And with table flow, we're doing the exact same thing in analytics as well. To stay at the really obvious in case it's not super clear already, streaming isn't a feature of this. Streaming is not a bolt-on to this platform. Streaming is the foundation of all of this unification we've been talking about for about 30 minutes. And that's what's driving our total addressable market to over $100 billion and continuing to grow. It was only a few years ago that we were laser-focused on the Kafka market with a market of about $50 billion. But as we've laid on these new capabilities, connect, process, govern and now table flow, we're systematically breaking down these traditional siloed approaches. We're chipping away at these traditional legacy technologies, and we're redefining what is possible. Now the disruptive nature of this data streaming platform has been recognized by industry analysts as well. In fact, Forrester actually created a whole new wave for streaming data platforms and named us a leader in this new technology wave as well as traditional waves. In fact, Forrester awarded us the highest scores for strategy in the streaming data platform wave, and they called us a streaming force to be reckoned with. So I'd like to wrap up with some final thoughts. This fragmentation of the ecosystem, these disconnected and duplicative tools has made innovation hard and expensive for far too long. And even before the rise of AI, the shift away from fragmented batchy tools to a streaming approach was already happening. And today, the rise of AI has simply accelerated that movement. In this world, a data streaming platform is not an optional thing. It's an absolute necessity. Customers are choosing us over a basic Kafka ingest or a point solution because we deliver unmatched speed, and we give them an end-to-end platform that solves for their data needs today and for their data needs tomorrow. We're boosting developer productivity. We're simplifying operations, we make seamless data integration between the operational and analytical estates a reality. This is something that was previously unimaginable. And that's what really fires me up. We're changing the game. We're changing the world ahead of us. We're disrupting this data infrastructure, and I couldn't be more excited by the unification movement we're driving. And with that, I'd love to hand over to Andrew, who's going to talk you through how we're making this opportunity -- we're capturing this opportunity.

Well, good afternoon, everyone. My name is Andrew Sellers. I lead Confluent Technology Strategy Group. My team consists of distributed systems researchers, our customer and product advisory boards and our global field CTOs. Our field CTOs engage all level of persona, but they really focus on technical executives. They help understand and communicate the business value that DSP brings as well as identify patterns across industries. And so as part of that work, I've had the opportunity to talk to over 130 of our customers over the last 18 months about their aspirations for artificial intelligence and how the data streaming platform can help enable them. So I've been doing data streaming for a little while now, but I've been doing AI in my entire career. It's what I studied in graduate school, and I've spent most of my time since as either a machine learning researcher or someone who commercialized AI-enabled technologies. And in that time, AI really felt like this thing for the last 20 years that was maybe 3 to 4 years away. It was just on the horizon, just on the cusp of being relevant. Obviously, this moment is completely different. I have to say, especially in the last 6 months, there's been this just absolute transformation. Our customers are going from science projects to actually realizing applications in production with real return on investment. As someone who's spent his career doing not AI for AI's sake but for AI to solve problems, to create value, I got to say there's nowhere I'd rather be than Confluent right now. Let me give you a little intuition as to why. Jay spoke about this. The way that AI used to work is it was very purpose built. And so I had to employ a legion of PhDs and data science and statistics. And they would have to do things like feature engineering and model training and model deployment and bias characterization. And they had to do that for every use case. Well, now with Generative AI, the models are inherently reusable. But we've all used ChatGPT. And so in a sense, the data engineering challenges haven't really left. I can't really use ChatGPT to run an enterprise application because it doesn't know anything about my business. It lacks domain-specific information. And so really, the data engineering challenges, they didn't really go away. They just shifted. They became more real time than ever before because in order to deliver the reactive sophisticated experiences that our internal knowledge workers and our external customers have come to expect. I've got to be able to do all that in real time. And now we're entering this exciting new way that we talked about of AI of Agentic and so agents are -- like those Generative AI workloads, but the Generative AI workloads were very prescriptive. I had a very defined description and flow of how data came in, what data I needed access to, how the generative AI would be used to make some kind of induction or learning? Well, now with agents, it's completely different. Agents have to react to their environment. So if anything, the data has to be even more contextualized. It's got to be more real time. It all has to be accessible. And so that's why we're seeing the adoption of the DSP for these use cases, accelerating in a way they never have before. Let's walk through a bit of an example. And so for those that aren't familiar, in the Career Sales ladder, one of the entry points is the sales development representative. They're people they do prospecting. And so what they'll do is they'll get some kind of lead list. They'll go out and do some research, so they'll gather some heterogeneous data, look at a LinkedIn profile, read an article, maybe see some proprietary information. And then they make some kind of evaluation. What's the right way to engage this prospect? Should I call them on the phone, sure I write them an e-mail. If I write them an e-mail, what does the e-mail say? Well, it turns out with our customers, this is a fantastic use case for AI. A lot of them actually do this because and the intuition is again with SGR kind of work, it's a numbers game. 95% is good enough if it gives me access to an order of magnitude of more market. And so the intuition between why this would be multiple agents and not one big agent is because agents are better when they specialize, a lot like people, a single kind of model doing a single kind of thing. So I can have an agent that does the evaluation and research lead work, another one that decides how to engage and another one that crafts an e-mail maybe and sends it. Pretty cool, pretty intuitive, right? Well, the challenge is, is that these agents are complex systems. And if I have multiple agents, they're even more complex steel. So standing up agents now, there's a lot of frameworks out there I can use, things like Bedrock or Agentforce, GPTs, langGraph, CrewAI. It's not quite a commodity yet, but it's getting close. However, to solve real problems, what our customers see is they need to use these multi-agent systems. And it's really challenging to choreograph all of this, to orchestrate their actions, to make them communicate with one another to coordinate their activities and to get domain-specific information from the enterprise. So they make the best decisions that they can. Well, it turns out this is a problem we've solved before. At Confluent, we helped revolutionize the way that services were operated in the enterprise with making micro services a reality. So about 10 years ago, we had these big services that solve the business problems. And they were really hard to deploy. They're really hard to reason about it took entire teams of people to run them. And if I want to make some kind of change, I'd coordinate with all those people and it was really tough. And so as there's movement to decompose that work into smaller micro services. But it turned out -- it only really solved kind of part of the problem because now instead of having all this in trust service communication, now I needed those micro services to talk to one another. And micro services never really became a feature that were accessible to enterprises until they became event-driven. And this has been our business over the last 10 years for 5,000 customers, is how we do the same principles and patterns and solutions that we've done to make micro services work, they all apply to agents as well, and that's what our customers have found. So the future is event-driven agents and the agents work just like micro services. Agents or micro services without guardrails. So to use the DSP to get well contextualized information to make great decisions and the patterns by which they communicate. And these are well established in the state-of-the-art. Those can all be implemented as event-driven systems. And so in fact, if you want to think about how we as Confluent is positioned in the AI ecosystem, we are a bridge between the foundation models, the data management platforms, the things that help you operate your LOMs and the agent frameworks themselves. So we're not going to sell you an LLM, but we're going to take that LLM and contextualize it with really great information. We're not an agent framework, but we'll take that agent no matter where it's built and will help it communicate with the systems in your enterprise and with each other, even if they were built on different frameworks. It allows you to be AI tool agnostic. These things become modular components that can be substituted because we all know the pace of innovation here, it's only accelerating. Of those 130 customer conversations, I'm so pleased that especially recently, we now have dozens that are in production, doing remarkable things with AI that are actually changing their businesses. We have a few featured up here. We have many more you can read about on the website. And we're getting adding by more every month. So all of this is possible today. We're going to work over the next few years in our road map to make this even more accessible, but it's already pretty easy today. And in fact, to show you how I'd like to invite Mike Agnich to the stage, please, our DSP GM to just show you how quickly this can be done.

Thanks, Andrew. I think Sean, maybe early promise that I would teach all of you how to build Agentic AI in 5 minutes. My goals, I guess, are a little more modest than that. But I am excited to show you what we're doing with event-driven multi-agent development and give you a bit of a demo. To do this, we've created an imaginary company. River Retailers. And River is -- has been a confluent customer for a while. They are already using us as their central nervous system for their operational data, for their micro services and applications. They're already using us to ingest data into their lake in their warehouse. But now they want to do something new. They want to leverage the new capabilities in our DSP and really build a Agentic AI to solve some of their stickiest, trickiest customer problems. And to do this, we're going to talk about the age-old problem of shopping cart abandonment. River is going to leverage Confluent's DSP to connect existing systems to a set of multiple AI agents, which are in turn integrated with their LLM or model of choice all orchestrated within our product. This is easy with Confluent because agents are just micro services and micro services are kind of our thing. Our multi-model system is going to leverage 3 agents. First, we built an agent that is our engagement, engagement agent. It's aware of how all of our customers want to be spoken to, have they signed up for our e-mails or SMSs. Are they currently on the site so that we could put a real-time experience in front of them right now. The second agent is our Insights agent. It's constantly reading reviews, it's aware of new products that are being launched. Think of it as playing a role of a merchandiser. The last agent is great at generating relevant content. It's integrated with our recommendation system, and our CMS. So that's always ready to put the right content in front of the customer. It's also aware of things like what discounts we have available to give. In all cases, these models are connected through Flink. Let me show you a little bit more about that second agent and the merchandising one and how it's built. First of all, I don't worry, there's not going to be a quiz on this. So if you can't read it all, that's fine. But what we're doing here in Flink is we've created a reference to a Generative AI model. All AI models are now first-class citizens in Flink. They're like tables or functions. In this case, the AI model I'm using is from open AI. In the description, you can see the instructions that I'm giving it. This is the task we want the model to do, the problem we wanted to solve. So now I'm going to run this model to analyze the product reviews and identify the #1 feature about the product cited by customers who have purchased it. The identified product feature is saved into another topic with the product review information. This agent will now constantly update itself as more products are launched and reviews are created. Because our Flink offering is cloud native and skills to 0, there's nothing for customers to manage on the infrastructure side, no provisioning, no new contract or commitment that has to be made, it's auto scaling from day 1 and fully integrated with streaming. And so now here is the results. In this case, our customers are still on the site and our engagement, our engagement agent knows it. Going to get this started. I think we are out of sync now. Okay. So the engagement agent knows that our customers are still on the site. So we can render a chat window with a personalized message. It highlighted what customers loved about the product, what they left behind. It referenced to previously purchased positively reviewed item. It showed other products that the customer might wanted to buy, and it offered a special discount based on what our customer, how they like to buy and the current -- the size of the cart that was abandoned. This isn't just AI to build AI. It's actually agents communicating with other agents in real time driving engagement and revenue. Okay. So let's take a step back for a minute. At the risk of pissing off my boss, this is a Jay, 8.5 years ago explaining the power of streaming. I'm showing this because making micro services work at scale isn't new for us. This is our original use case. It's really Confluent's home turf. Agents are micro services, specifically, they're stateful micro services with a brain behind them that's an LLM. Agents will communicate the same way micro services have always communicated. Streaming is the obvious protocol for this Agentic AI communication and Confluent's DSP provides the missing pieces. First, a ready-made ecosystem of connectors that connect to every critical system; second, stream processing to prepare and shape the data in real time. Third, governance so that every enterprise around the world can trust it and finally, native integrations into models of all sizes and existing in all locations. So one more thing. These applications and agents aren't just operating. They're also creating more data. That critical data needs to keep flowing into lakes and warehouses. If micro services was the first major use case of streaming and Confluent, great ingestion into the analytics estate has emerged as the second. In the near future, as others have mentioned, we're releasing table flow that natively integrates and materializes streams into the open standards of Iceberg and Delta and unifying with the critical catalogs of Glue and Unity and Polaris. Historically, this has been a huge pain. But now this has become easy. This demo is about table flow. It's unifying the analytics state, and we're making it as easy as first, choosing a topic, second choosing a destination. In this case, we've picked Databricks and Unity. And third, enabling table flow. You can see that with just a few clicks, our data is materializing in Databricks. Their users will simply see tables showing up in the interfaces they're already using. The fact that these tables are actually streams is invisible to users. This is ingestion done right, and it's the final step of unifying operational and analytical domains of data, and it's done in partnership with industry leaders and the vendors that our customers love. Table flow is unified with the DSP in the same way that Flink is unified. It's serverless, it's one click away. You don't have to pre-provision or presize anything and it just works. So I think it would be easy to see those last 3 or 4 slides and kind of think how cool is that really? What I'd say is and I think Ali mentioned this, too. If I was sitting up here with our champions on the left and the operational space, who are running streaming at scale. And on the right, they would have their friends who are on the analytics estate. To them, this creates applause. When we show them these 3 or 4 -- 3 or 4 screens, they know that this is replacing dozens or hundreds of tasks and pipelines and sometimes vendors. It's an extremely painful process today to get your data in the language, something like SAP speaks and to make it immediately usable in the language that the data analytics world speaks. It's about type casting and merging and all kinds of these data prep tasks that today, we've really left up to our customers to do. What I just showed you in that demo was doing all of that. It wasn't just moving the data. It's also doing all of those tasks. All in a few clicks and all done automatically. So thanks for your time. I hope this helps demonstrate how our investments are helping make AI real for our customers, and I appreciate everyone's attention. I think from here, we go -- we're going to go to break. And I think Erica Schultz will take it when we come back. So thanks very much. [Break]

Everyone, please take your seats. We're going to begin momentarily. Ladies and gentlemen, please welcome President of Field Operations, Erica Schultz.

Thank you. It's great to be here and great to see all of you. Thanks so much for joining us. Before I dive into the content, I just wanted to follow up on Jay's announcement earlier with a couple of comments. First and foremost, it's just been such an honor to have been part of this leadership team in Confluent and to have been part of this incredible chapter of Confluent's great story these last 5.5 years. So I'm forever grateful to Jay and to the leadership team to have been able to partner with you all. And second, I want to make sure it's really clear that we are committed to a very thoughtful transition. I am very much in seat until my successor is identified and we've onboarded that person. It's really important to us that we have a very thoughtful and planful transition. So as far as I'm concerned, it's business as usual. We have forecast calls tomorrow. We have a Q1 to deliver in many future quarters to set up for success. So I want to make sure that's clear. Okay. So let's dive in. I'm going to talk a little bit about our next-generation go-to-market model and how we're set up for success in Confluent's -- next Act. What I'd like to do first is kind of reflect on our 10-year journey as a company, and we think about the 10-year journey and where we're headed in this framework of Acts. And I'll talk about a couple of key milestones and the bets we're making in each Act. Since the company was founded in 2014 in those first 5 years, we reached $100 million in revenue and hundreds of customers, largely through monetizing open source with our Confluent platform offering. And this defines our Act 1. Confluent Cloud after we launched that, it really fueled this incredible land-and-expand, and it helped us accelerate customer acquisition by fiscal '19, over half of our customers were Confluent Cloud customers. And last year, Confluent Cloud surpassed 50% of our revenue base, an important milestone. So this is our ongoing Act 2. Building on this momentum, we have begun our Act 3, which is really about this transition from being a product company to a multiproduct platform company with our industry-leading data streaming platform. Milestones such as 5,800 customers, 194 of them spending over $1 million a year with us, strong customer adds in 2024, so that we continue that land-and-expand cycle. Those are all a testament to how Confluent is becoming a central nervous system for our customers. And we've achieved a powerful combination of growth, profitability and scale as we've surpassed the $1 billion run rate, and we've become full year profitable for the first time last year. And along the way, we've made some very intentional go-to-market bets aligned with our objectives in each of these Acts in order to fuel the growth and deliver efficiency. So let me talk a little bit about those. So in Act 1 is we are taking Confluent platform to market. We were very focused on becoming one of the greatest open source monetization engines. And to do that, we focused on key verticals that had a lot of Kafka penetration like financial services and public sector. We rolled out professional services offerings to help those early adopter customers adopt Confluent and we selectively enter new countries. In our Act 2 as Confluent Cloud was rapidly growing. We leaned into our CSP partnerships, our ISV partnerships, the digital native segment, and we introduced a product-led growth offering. We saw rapid global expansion. We got into a number of new markets. And then, of course, over the last couple of years, as we've been talking about, we executed this consumption transformation. Now we find ourselves in our Act 3, where we're building on the previous 2 Acts, and as we transition from a product company to a multiproduct platform company, the key bets we're making include a focus on technology executives, and I'll talk more about that, our global system integrator partnerships, specialization in our field force and more verticalization. And in a few minutes, our Senior Vice President and Global Head of Sales, Ryan Mac Ban, will join me on stage, and he'll double-click on some of these key bets for Act 3. So let's talk about the consumption transformation. I know that's been a hot topic that we've all communicated on over the last year plus. It really started almost 3 years ago and the original intent, the intent has always been to better align customer value with field incentives with Confluent's success. And in 2024, we took a step change really in terms of how we incentivize the field around consumption. The earning power in the compensation plan was around consumption more so than around booking typical subscription deals. And this consumption transformation has really paid off. We are super excited about where we are. When we reflect on some of the outcomes, it certainly benefited us in terms of new customer acquisition, being able to bring customers in early just by allowing them to start consuming. We saw 2x growth in year-over-year customer net adds in FY '24. And we're able to point that machine at the highest propensity prospects in the market. We created a custom curated list that we call the Confluent 2000, and we were pleased that we penetrated several hundred on that list and welcome those customers to Confluent. So we're really pleased about the acquisition engine that the consumption model has helped us build and then after we land, of course, we want to focus on expanding. We can -- now with all of the new products in the data streaming platform, our customers can experiment and can really -- we can offer a frictionless adoption with this consumption model. There's no contracting process in between customer adopting a new product, for example. So it enables us to truly land and expand, and we see those outcomes in the growth of our larger customers -- and for example, we're super proud at the end of 2024 to count several customers over $10 million in annual revenue with us. So we're excited about what the consumption transformation has enabled us to do. In '24, it was a lot of work to execute all of these changes. You can see some of the operational items that we took on along the bottom of the slide here. And what's exciting is not only these outcomes, but the fact that this model, it gives us the right foundation for this next leg of the journey. It's the right model as we go from product to platform. So we're really excited to have this strong foundation underneath us. In fact, a consumption-based go-to-market model is where much of the industry is headed. Many of the new Agentic AI solutions are leveraging a consumption business model. And so we're pleased to have this transformation behind us and to be operating in this model. And we -- as we were going through the transformation, we were fortunate to be in contact with many of our peers like Databricks and MongoDB and Snowflake and Elastic but the reality was we are all pioneering to some degree. And even those partners and those industry leaders were only months or quarters ahead of us. So we are all pioneering together. There was no proven playbook. As we look at this next transformation of going from product to platform, it's a much more well-trodden path in the industry. We all know many companies who have executed this transformation and we're fortunate here at Confluent to have deep expertise on our leadership team of leaders who have been at companies who have successfully executed this transformation to a platform offering. Companies like Oracle, LinkedIn, Palo Alto Networks, Salesforce, VMware, UiPath, Cisco. So we're able to draw on that experience as we execute this next transformation. So I've talked so far about our internal readiness for this shift, the consumption transformation and the experience across our executive team. What I think we're most excited about, and you've heard a little bit about it this morning, is the market readiness and a demand for what we're bringing to market. Over the last several years, we've really been evangelizing the place of data streaming and often through open source migrations. That's what we've been talking to our customers about. And now that we have the full data streaming platform, we are able to lead bigger architectural conversations with our customers. Conversations about bigger technology and business impact in their organizations. If 2 years ago, we were just talking about kind of connecting and moving data. Now we're looking at the whole data landscape, and you heard some of that this morning. We're talking about the systems upstream and downstream and real business impact. So this message of shift left, and Shaun talked about it this morning, it's really resonating with our technology executives. And this is the conversation we're having day in, day out. I love this quote from one of our largest customers, the Head of Data at a Fortune 50 financial institution who -- and Mike talked about the chaos of these environments, this individual said we have 40,000 or more than 40,000 ETL jobs today, it's chaos. So shifting data governance left would be transformational for our organization. So the market demand for DSP framed with this shift left value prop is really resonating in the market and specifically with technology executives. As I mentioned, this message really resonates with tech exacts. And so the way that we're leveraging that is we are building on our existing relationships with our champions in the accounts, the data streaming practitioners to reach the next level of scope. And we're finding ourselves in more and more conversations with CTOs, CIOs, Chief Data Officers, Heads of AI, which is really exciting to have this bigger conversation. Of course, we want to do that as often as possible together with our global system integrator partners because those partners have long-standing trusted relationships with these executives. And so where we can, we're going arm-in-arm with our GSIs and other partners that have a seat at the table. And Ryan will talk a little bit more about that. What we're finding is this shift left framing and value prop, it's resonating and then it's opening up new conversations with more leaders in the organization. And so we're finding that it's opening up conversations second and third and fourth conversations with analytic practitioners or data engineers. And we're getting into different parts of the organization so that we have a more multi-threaded approach in the account. And so we're getting into new areas that maybe we haven't been in before, and we're finding this is a really effective way to uncover new opportunities and new use cases within existing customers that maybe we weren't aware of before. A great example of this I met earlier this week with a CIO of a top 20 bank in the U.S., a super regional bank who's been a customer of ours for a number of years. They started on Confluent platform. They then migrated to Confluent Cloud several years ago. And we have about 30 use cases in production with this customer. And they're a very happy customer. They spend more than $1 million a year with us. And the CIO said to me, they have 30 use cases in production and another 30 in the pipeline. And he said to me a couple of years ago when we started working together, the need for real-time data was really only in the minority of my use cases. Today, I need real-time data in all of my use cases from fraud detection to marketing. So we need Confluent data streaming platform at the center. And then, of course, our discussions around AI and one of our key follow-ups is a deeper dive of a briefing on Confluent's role in their AI strategy that is just an accelerant on what was already a really exciting future for the data streaming platform. So that's one example. And these are the types of conversations we're finding ourselves in. So just in that example, we're finding so many customers are identifying more and more use cases where real-time reusable, reliable data is essential. And of course, facing down the barrel of GenAI and Agentic AI and realizing that the role that, that data plays is just key to making it all work. So we feel like we're in the right moment to have a strategic seat at the table with the technology executives. And as Jay and Andrew discussed earlier, the creation of AI models in their application is truly reinforcing every company's need for the data streaming platform. Companies are just starting to make the leap from that early interest into real deployed use cases that are making a difference in their customers or employees' lives. And to do this, they are recognizing the need to unify these operational and analytical estates as we've been talking about. And what I will tell you is that AI is a primary topic in so many of our customer engagements. At least 50% of our executive briefings where we bring customers in for a few hours with our executives include AI, and that's kind of increasing every week. And it was one of our top topics during current, our customer event. So AI needs a platform that can bring trusted enterprise data to bear and coordinate actions into the operations of the business in real time. And that's what we're bringing to market. So we're approaching the opportunity in the market in a couple of different ways. We think about the different segments of opportunity. We talked about this a little bit earlier, but we think about the infrastructure providers and then the application providers and then, of course, what I might call traditional enterprise adoption. And in 2024, we landed several players in GenAI, including major infrastructure players like OpenAI. And while these customers have created the AI wave in the industry, just as much as they have a need for Kafka and data streaming sort of the application builders and there's more and more of these every week, every month. And so we've made great inroads through a very targeted strategy with the application builders. In '24 customers like Dialpad and Cursor AI are building AI applications that are underpinned by Confluent's data streaming platform. And those successes will scale as more enterprises adopt those solutions. We do have AI use cases in a number of our enterprise customers. Andrew referenced that earlier. But really, the growth in the kind of mainstream adoption in the large enterprises, it will take time, but we think this is the biggest category of opportunity. So we're pleased to have a seat at the table. But this is a little bit about how we think about the market segments. We had a fantastic conversation between Jay and Ali earlier about the partnership between Confluent and Databricks. We're so excited about it. As you can tell, our customers are just -- our joint customers are so excited. And our SI community is very excited. And so 11 SIs have agreed to combine their Confluent and Databricks practices to capitalize on this announcement and our plans to deliver products together. So we're super excited about that. These are SIs that cover all theaters, and we have ongoing conversations with some of the larger SIs, including Accenture, Deloitte, Ernst & Young, who are -- we've been partnering with for a long time, but we're really now talking about, okay, how can we capitalize on this partnership. I love this quote from the CTO at KPMG, who says, by integrating Confluent's cloud-native platform with Databricks Lakehouse, businesses build real-time AI apps, driving faster insights and better decisions. So it's just a little bit like our CEO said on stage this morning, what took you so long, but our customers and our partners are super excited about this. So all of this adds up to an expanding TAM for the DSP. The combination of these best-in-class products means a really significant opportunity for our customers and for us. And we see kind of a flywheel effect as customers adopt different components of the platform. It's leading to more Kafka adoption and adoption of other products. So we're really excited about that and the expanding opportunity. And as we've established the relevance of DSP in the last few years, we've made major inroads with a number of our long-time customers, long-time key Kafka customers. One of those is Wix, the website builder, if you're familiar with them. They've been a customer of ours since 2020, a great Kafka customer, and they're using our platform for many use cases. Most recently, they've also implemented Flink for their feature store that processes billions of events per day, to power a range of data-driven experiences from real-time personalization to machine learning model inferences. So we're super excited to be working with them and a number of other customers who are excited about these components of the platform. We have more progress to make to deliver on the full vision for our customers. We truly are in early days, but we foresee each of the DSP components to have the potential to become large businesses themselves in their own right for us over time. So we're excited about that. And at this point, I'd like to welcome Ryan Mac Ban to the stage, who will talk about how we're evolving our go-to-market motion to maximize this opportunity with the data streaming platform. So Ryan, come on up.

Thank you. So I'm a little over 90 days into the role. It's so great to be here. I see some familiar faces having done investor callbacks in a previous life. But I thought it just might take a minute for those of you that are new or that I'm new, too. Ryan Mac Ban, I am leading global sales here. They're, again, 90 days into the role. Background, spent time at Cisco in an acquisition coming in with Mario, Luca, Prem and Soni, who created multiple billion-dollar lines of business, spent time there across, obviously, hardware, software and services, then was part of creating a market around SDN. I know that sounds funny today, but was with Martin Casado and Nicira and the acquisition of VMware, where I led software-defined data center, $4 billion business, 1,200 people growing 20% and 6 acquisitions and then recently from UiPath, where started pre-IPO, $600 million and then wrote it to $1.5 billion and was asked to actually bring online everything outside of core RPA. So now getting into it. As more and more customers look to bring on our data streaming capabilities, we're going to continue to evolve our go-to-market. As Erica touched on, our shift left architecture and messaging is resonating with our executives in our accounts. And I will tell you that we're going to continue to double down on this high-impact engagement. We're reaching these executives where they're at. We're participating in 60-plus global executive conferences, summits and roundtables and I'm asking each of our AEs all the way through my leadership team to leverage our executive briefing centers that are available throughout the globe. With our DSP and the value now that we can offer to executives within our accounts, we've actually earned the right to deepen and strengthen our relationships, and we're meeting our executives where they're at, at the right time, at the right place. Second, we're now a platform company. This is the right time to introduce specialization, sales specialization as well as technical specialization and being able to deliver these resources at the fingertips of our sellers across each region in the globe and to our partners. As you heard from Erica, many companies have gone through this. I've had the pleasure of actually doing it 3 times. So whether it's at VMware, at Cisco, UiPath, ServiceNow, even CrowdStrike with multiple modules, multiple companies have been on this journey before and have done it successfully. I'll give you an example. We actually had one of our account teams meet with one of the largest U.S. air carriers. I happen to be Concierge Key. I will leave it at that. They asked us if we could actually stream real-time crew scheduling with that of irregular flight operations. Our specialists then were engaged, went through and qualified the opportunity. The answer is yes. Within 3 weeks, we were able to, within our system engineering team of specialists, go in and migrate that carrier to a modern Confluent Cloud stream processing platform within 3 weeks. This is just one example. We're going to continue to drive best practices from these type of engagements out there in the field and not only share it more broadly internally, but then also with our partners as well. And third, you heard Erica talk about our continuous investment in verticals. And I've been through this one before as well. We want to create vertical-centric plays that solve vertically centric challenges with our data streaming platform. We want to build on the success that we've experienced with our financial services team. And I think you've seen the number of banking customers and financial services customers we have to date. We've now invested more in our public sector team and see that as an opportunity going forward. And we've also added telco and high-tech in this year. I make a joke about it, but if I had another 30 days in the role, I would have instantiated health and life sciences as 30% of the world's data actually comes from that vertical alone. So we'll be looking at that going forward. And we will continue to really evaluate a handful of verticals and continue to strengthen this in our go-to-market approach. Look, we're asking also our partners, as you heard from Erica as well, to drive our data streaming platform, especially in that SI community. These partners are already leading digital transformation efforts within our accounts. Now we have the ability with our data streaming platform to help them unlock further opportunities. And I've been around the agentic game for quite some time. The one piece that was missing was real-time data streaming. That's exactly what we're able to do. But partners like Nest, they've developed Flink stream processing capabilities and expertise, and they're leveraging our data streaming platform, and we're driving adoption and awareness and doing that with them. We also have what you heard today from Ali and then also from Jay, our Confluent Databricks partnership. We're now working with them to put together joint go-to-market value propositions around Delta tables and Tableflow, as you heard, really bringing together that operational and analytical -- or analytical -- or operational estate and analytical estate as one. It starts by sharing our shared installed base, and we're doing that today. And look, we also now have solved problems with data mess for less. One of the top 10 North American banks, which we partnered with GoodLabs, was struggling to unlock, and this is not new because the mainframe has been around for quite some time, the siloed and stranded data that sits in the mainframe to actually make it available as data as a product to then take advantage and to reduce actual financial crimes. And if you think about that and all of those massive transformations with our global system integrators as well, Accenture, EY, Deloitte, that have been stuck, that have been waiting for something like this to be available to unlock those massive global transformations. We also believe in the utilization of professional services, whether it's delivered by us or by our partners, we know that it leads to faster business outcomes for our customers. We also know that 80% of our customers get more value when they leverage our professional services. And on average, they consume 100% more than without our professional services engaged. We're going to continue to drive this. We're also going to continue to drive this through our partner to help our customers realize the value of our data streaming platform. I'm now going to turn it back over to Erica. It's nice to see some of the new people and some of the familiar faces. Thank you.

Okay. Thank you, Ryan. All right. Now I'm going to shift from -- we talked about the one growth vector that is taking our data streaming platform to market. I want to talk about our other growth vector of continuing to soak up the world's Kafka. So this was a big part of our Act 1 and Act 2, but there are 150,000 organizations out there using open source Kafka. We have 5,800 customers roughly. So there's still a wealth of opportunity out there. So let me talk about some of our strategies. So first, acquisition of new logos remains extremely important. And I talked about that a little bit earlier. We have created kind of a custom curated high propensity list that we're very focused on. We refer to it as the Confluent 2000. And over the last year plus, we've invested in a global team that -- whose job it is to go penetrate landing in those customers. Think any Fortune 500 customers that we don't have, plus new GenAI companies, large digital natives and of course, top companies around the globe. In FY '24, we are able to acquire hundreds of these accounts. And again, our consumption-based model enabled us to do that without a lot of friction. So our goal is continue to land those companies and then, of course, be maniacally focused on winning the first workload and getting that into production and getting that consumption flywheel going. So that leads us to the next growth vector. I want to talk a little bit about key markets -- we are -- our goal is to be very intentional about highest propensity markets in terms of where we put our resources. And so let me talk a little bit about that. We do continue to see greater demand in our digital native segment. That's been a big grower for us over the last couple of years. Customers like [ Block ], eBay, Clari and many others are long-standing customers. And more recently, we've landed one of the top e-commerce businesses in the U.S. and many food subscription and food delivery companies, a global monitoring and security platform and a number of others. So digital native continues to be an area of opportunity and focus for us. Also building on the success of our longtime strong suit, which is financial services and this more recent strong suit of digital native, we've increased investment in our U.S. public sector focus as well as the telco and media sector. Within public sector, that was an early vertical for us. We continue to see great fit and great use cases there. We have use cases across all sectors, including DoD and civilian agencies leveraging both CP and a little bit of cloud. And then, of course, the use cases are quite replicable globally. So for example, the United States Postal Service, which definitely has a problem of scale, they deliver 129 million pieces of mail daily. They have 232,000 vehicles and 10 million users to their site daily. And they rely on us to connect legacy and modern systems to enable operations at scale. So that's a great use case and again, very replicable globally. In telco, 10 of the top 10 global telecommunications providers already rely on Confluent. And yet we think we're early in the opportunity. There's room for expansion in those accounts. And there are a number of accounts that we have yet to land and expand. So that's an area of focus for us as well. And then finally, in terms of geographic priorities, we see high propensity and exciting opportunity in the India market. So here, we've seen over the last several years, great success in digital native and financial services and other verticals. And we've struck a strategic partnership with [ Jio ] of the Reliance Group, who is running 3 major use cases with us, an MSK conversion for Internet usage data. They're running Confluent Cloud on Jio Cloud and [ Jio Star ] is now streaming the IPL for the cricket fans in the room for 80 million users, leveraging Confluent. So that's an area of focus for us as well. So we're excited about these growth bets in very specific markets. We've talked a little bit about some of our bets with strategic partners throughout. Let me just complete the picture a little bit. We spoke about our SIs and our Databricks partnership. We're also executing on joint go-to-market activities with other ISVs like MongoDB, Elastic, Snowflake. As Ryan mentioned, the goal is to share installed base account information, get our teams together, have a very clear joint value prop for the customer and go to market. With the CSPs, our relationships are truly one of symbiosis. We continue to work very closely with AWS and GCP and Microsoft. And our strategy with them, just as a reminder, all of their sellers are compensated. They retire quota when they sell Confluent. Their customers can purchase Confluent through their private marketplaces and retire against their enterprise agreements with the CSPs. So there's a lot of value to the sellers and to our joint customers to partner together. So we're constantly looking for areas of joint innovation together on the product side. We're looking to optimize the marketplace experience for our customers. And then similar to what we've talked about with Databricks and others, bring our field teams together with joint go-to-market strategies around the globe. So that remains very active. And then finally, we're also investing in the OEM and MSP model to be able to scale by tapping into new markets and pockets of existing open source. Our recently announced partnership with Saudi Cloud Computing organization -- Corporation in Saudi Arabia is a great example of this reach, and we'll hear more about that in just a little bit. Finally, the last big growth vector as it relates to soaking up all the world's Kafka is really about making Kafka more accessible. And Shaun spoke about this earlier. We know that large enterprises run a large variety of workloads. Some are mission-critical. Others are more flexible maybe in terms of their availability or latency needs. And in the past, some customers have viewed Confluent versus open source Kafka through the narrow lens and for some lower-value workloads, our dedicated offering wasn't the right fit. So in 2024, we've really expanded our product portfolio. And now we have a Kafka offering for every workload and kind of every need from our customers. And we also adjusted our list prices. So this has truly opened the door to more conversations, more workloads and ultimately, more opportunities to deliver value. Sean referred to this already, but the addition of freight clusters and the enterprise SKU and WarpStream into our Kafka portfolio, it really gives us a complete suite of offerings. And we're seeing a number of customers kind of mix and match and add on these new offerings to create that streaming mesh that Shaun referred to. So we're really excited about how these new offerings are taking us to new workloads and in some cases, totally new customers who -- for whom these offerings are a better fit. So we're really excited about this portfolio, and we feel like we have a great balance of capabilities, cost and scale. So in closing, I think I do want to spend a moment on -- as we're focused on these growth strategies and the go-to-market bets, it is always important to us to deliver go-to-market operating leverage at scale. And I just want to highlight a couple of data points here. Our land-and-expand model, as you've heard me talk about, it really drives leverage as expansion takes hold and the consumption model really removes that friction from the expansion motion for our customers. We saw 15% year-over-year growth in revenue per sales and marketing head count. So we're continuing to get more productive and efficient in the organization. Our gross retention rate continues to be over 90%. So our customer base is very sticky, which means a lot to the business. And then we've seen 25 points of improvement in terms of the sales and marketing as a percentage of revenue, non-GAAP since 2021. So significant margin improvement. And we're not done yet. We see efficiency drivers ahead such as continued land and expand and retain motions with the consumption models that continues to take hold. These strategic partnerships and the partner ecosystem overall, we believe, can bring us leverage and efficiency, the cross-selling of the platform and the DSP offerings. And then as we talked about the prioritization on the highest propensity, highest value verticals and market segments. So in closing, the company is well positioned for our Act 3. Our go-to-market machine and the bets we're making continue to be in service to the vision that we've had since the inception of the company, soaking up the world's Kafka and now delivering incremental value with our data streaming platform. And the milestones and achievements that we've had in the first 10 years are exciting truly because they've now put us in this position of strength. They've built this foundation for us as we embark on our Act 3, and we look ahead to driving continued durable and efficient growth at scale in this next act. So thank you very much. I'd like to welcome Stephanie to the stage. I forgot the intro.

Good to see all of you today. As we continue through today's agenda, I want to go back for a moment to what Jay said at the beginning of the day, which we're really here for a couple of things. One, to deep dive on data streaming; two, to understand the mission criticality of data streaming for AI; and three, to really understand Confluent's position to lead that. That's what we're -- this is all about here today. We've done a lot around the product. We've done a lot around go-to-market conversations. We mixed in some customers throughout, but I think I'm luckiest because I get to spend time right now with the customers on stage. And we know at the end of the day that it's their success that will drive our success as an organization. So together, we're going to spend the next hour with 2 of our customers, one at a time and answer those very 3 questions and things that we've been talking about all day and hear directly from them. So I'm pleased to welcome to stage first. We have with us the Saudi Cloud Computing Company. And joining us, they're also known as SCCC. Joining us from SCCC is Talal AlBakr, CEO. So please join me in welcoming him on stage. All right. Welcome. Let's have a seat.

Hello, everybody.

Hello. Let's get the energy up here. Thank you. Very, very much. He's joining us here from Saudi, obviously. So you've come a long way, and we appreciate it really very, very much. I think the first and most important thing is we've been running around saying SCCC for a while now. And really grounding everyone here in the room, how it came about. This was a venture, how does this come about? And what is the mission of the organization today?

That's a beautiful question. We've been trying to localize cloud in Saudi since 2017. We initially through STC, we built something called STC Cloud based on OpenStack, and we try to really enable that element. And we came to the conclusion that we needed more of a hyperscaler capability. And thus, we started saying, okay, our approach is probably not to go with OpenStack, but localize different technologies from outside of Saudi. And we've been doing a lot of those partnerships to really localize different technologies. Initially, we started out -- we wanted to work with the American hyperscalers. There was delays. We kicked off something with Alibaba Cloud, and then we started to pivot and bring in other elements and other solutions. And the momentum there was that we really wanted to build something localized to the Saudi market addresses the needs of the market specifically and allows us to grow that ecosystem in a positive manner.

Fantastic. So obviously, to be part of that you're building this, data streaming came on to the forefront here as part of this data architecture. Maybe talk about the role of data streaming as you're building out this technology innovation.

Yes. So I remember I used open source Kafka in 2018. I think we had an AVL solution where we were doing vehicle locations, and we had fleet management solutions that used it. And I remember, from a technology point of view, it was extremely good, but from a manageability point of view, it wasn't really easy for us, and the team was not competent enough. But as we started seeing Confluent coming into play and the data streaming element play a more pivotal role in AI, AVLs, fleet management, all of those things and having a full managed platform, things started to improve and our outlook on how we could utilize Confluent became even more proficient. And we saw a real use case where we can really enable customer solutions and offload a lot of their headaches and help them and enable them for the future of what we're going to do with AI, and all the different ambitions that the country has for AI in that regard.

Yes. As we spoke earlier about it, just the importance of real time to be the backbone under all of this. We also spent quite a bit of time talking about your vision around AI and the role of data streaming within that. Can you talk and elaborate more around your AI initiatives as it relates to data streaming?

Yes, yes. So right now, as part of what's happening in Saudi Arabia, the AI is becoming extremely exciting, and there is a huge investment. There are investments that have been announced. There are investments that will be announced soon, and we're seeing a lot of momentum. And a lot of -- and the beauty around that is we're not just working on use cases, there is top-down enablement. Right now, almost every ministry has to have an AI use case in place. So you see the government is really embracing it. A lot of things that are happening right now is a lot of -- there's a privatization effort to move all of these government entities also into private sector companies and to be self-sufficient. So we're seeing a lot of play for AI to enable more enhancements to enable -- to streamline their operation in a better manner to reduce costs and things like that. And as we're seeing different sectors and different things that are happening, and like Ryan mentioned, we're seeing health care going into these privatization elements, and they have huge requirements for data and interconnectivity. We're seeing AI going and playing a more pivotal role in what we're hoping to achieve. And right now, from our perspective, we've invested heavily in a lot of GPUs. We're buying a lot of GPUs. We're working with customers on their use cases. And then we go and how do we feed that into those use cases? How do we do the training element? How do we feed data for inferencing? That's where we're seeing a huge play for Confluent and the data streaming element and enabling us to really give real tangible data to these inferencing models and these training models so that we can come up with use cases. In Saudi, we also have an LLM, [ local LLM called LM ] that was developed by the Saudi Data and AI Authority. And that's one of the things that we're very proud of. Why? Because it's specifically detailed to the Arabic language. And it's one of the few Arabic language LLMs. And then the beauty about it is it works in multiple dialects which is very unique and very interesting because as you see, the Middle East has multiple dialects within the different countries, and that platform is able to do multimodal access, be it written or be it verbal and be able to interpret that in a very good manner. And we're very excited about that.

It's fantastic. You basically are saying there's an AI mandate across all of these industries and sectors to get started. And you're believing in the fundamental that you cannot do this without real-time underlying on it. So just the magnitude of the role real-time can play within it is pretty compelling. You could have gone about this a couple of different ways on why Confluent? Why the partnership with Confluent?

Well, again, we've had previous experience with Kafka. And I personally suffered. So when Confluent came into play and the enhancements that came as opposed to the open source Kafka, we really like that. Now the beauty about what's happening right now in Saudi as well is we're seeing huge initiatives across multiple sectors. We have the Asia Football Cup will be in Saudi in 2027. We have NEOM Winter Asian Olympics will be in 2029. We have the Expo 2030. We have also the World Cup, the Football World Cup will be in 2034. And a lot of these elements are really pushing us to come up with real-time use cases for AI, for all of these different things. So I had a meeting with several ministries and again, the ambition that they want is extremely, extremely big. And what they want to achieve is extremely ambitious, and it requires real-time data that currently, the setup of what we have today is not able to deliver. So having Confluent in the backbone in the middle of that element, really interconnecting everything, interconnecting multiple systems, be it local, be it on-prem, be it in the cloud or be it all over the place and pulling that data was something that's crucial to all of the AI demand that we're going to be developing with these customers. And when we sit with them, a lot of them, we've already onboarded into our cloud, and we start talking about what they need to access. And we see their data is not enough. We need to really get data from third-party elements from other government elements and get that data processing and streaming element into play is becoming more and more critical.

That's incredible and the ambition that you guys have around it and realizing that, seeing just the data that you're needing to bring together, as you said, wherever the data exists on-prem in the cloud and then not only bringing that together, but your ability to just process it, govern it and do all of that with us. We're very excited about the partnership. We've talked a lot about the importance of real-time data to power AI in these applications. Maybe looking ahead, can you share just your take on which industries you think will get the fastest uptake of this? And what are some of those use cases that you're most excited about?

Wonderful. Now as I mentioned maybe previously, sports is picking up in Saudi. We have global events being sponsored. We are also investing heavily in really bringing global players into the football league, the Saudi Football League. We're working on that. But also one very important element is we coined the phrase giga projects. So initially, it was mega projects. Now we're talking about giga projects. NEOM is probably the most well-known. We have the [ Ria Gate ]. We have [indiscernible]. We have Red Sea Global, all of these different projects that have partly huge scale and that need multiple elements to manage. And there is huge demand. And when we start looking at the private sector there, we're thinking, okay, that's where the potential business will come from and the potential need. What we're surprised with today is also the government entities. So we sat down with people in the Ministry of Tourism that we're extremely excited about what AI concepts they could do to sort of enhance a tourists visit and experience when they come to Saudi. And they're coming up with these extremely unique use cases that they've been working with us on. And then we sit with Ministry of Transportation. We recently launched the metro in Saudi. And so they are more interested in seeing how do we really leverage that and public transportation in a wider scale. And it's just so exciting. I was telling somebody almost 4 months ago, we announced King Salman Airport. I was going to the airport right now. Currently, the airport in Riyadh is called King Khalid International Airport. That's going to be a terminal within the King Salman International Airport, and they've already begun construction. So things are moving at an amazingly fast rate. Investment is heavy. And the beauty about it, all levels of the country, be it private or public, are really bought into the AI element, and we're going to see huge use cases in the coming years.

It's -- the opportunities are endless. Maybe share with this group what success looks like in that 2030 vision for those who are not familiar with Saudi's 2030 vision, like what the growth will look like on that over the next 5 years?

Yes. So we put an ambitious plan to really transform the country. It started maybe around 6 years ago so that in 2030, we reached specific KPIs on how we really diversify our income and our economy and not be fully focused on oil, have other streams of income. Tourism is something very critical. Technology is something very critical. We already launched a company called [ ALA ] which is building factories within Saudi, which is something 6 years ago, you never heard about. And it's really evolving the whole ecosystem, the whole socioeconomical fabric of Saudi has evolved in a really amazing manner. We're attracting foreign investment. We have a lot of companies that have their regional headquarters in Saudi, and they're really benefiting from that. And we're really trying to drive the initiative on how to really grow the Middle East from Saudi being Saudi, the heart of the Middle East and really leverage that. Huge investments in AI, huge investments in data centers and connectivity in a lot of different vectors, and we're seeing huge growth and ambition around that. And 2030 will really tap into those elements. One of the beautiful things I think that we're doing is right now, we had a KPI for women empowerment. They already achieved the 2030 KPI, and now they have a newer KPI to achieve, which is something very amazing. Also tourism expectations, we had a KPI for tourism. The Ministry of Tourism last year said we already achieved our 2030 KPI. Now his Royal Highness, the King and his Royal Highness Prince Mohammed have provided even new goals for the ministry to achieve. So it's exciting times. The whole energy within Saudi is very beautiful and something that's very contagious and you really get excited about this potential that you're going to see and tap into. And in the last 6, 7 years, we've already seen things change drastically in a very positive manner, and we're very excited about where things are going to be in 2030.

It's truly fantastic. It's admirable. And I can tell you, as our leaders come back from being there in Saudi, each one of them, my turn will come, and we'll get that opportunity. But our leaders who have been over there and coming back just are -- it's contagious, energy and the ambition on how you guys are going about this. So it's truly impressive. We are absolutely thrilled to have you as this managed service partner and be the first there in Saudi on data streaming. And I just want to thank you, and can we give a round of applause. Thank you very much. So when I opened, I realized that I probably said what I dreamt for versus reality. I think I said an hour altogether, and we only have 30 minutes altogether. That was my wishful thinking. But what I can tell you so that I'm not a liar is that both of our customers, Talal and who you'll meet next are actually staying and we will be here through the rest of the day and here at the cocktail reception as well. So hopefully, like when we put all the time together, I'm not lying, it's an hour that you can spend time together and get questions answered with them. Okay. So we're going to shift gears now, and I am pleased to welcome -- we have with us Affirm Holdings, I don't know if all you know, but think buy now, pay later. They are totally transforming in fintech and creating a whole world of better than credit cards. And to join me on stage, we have Shyam Mani, come on up.

Thank you so much. Glad to be here.

Very appreciative to have you here. Well, let's start with you. Maybe tell folks a little bit about you and your role at Affirm.

Yes, sure. So I've been at Affirm about 6.5 years, and I'm the Director of our data and storage services team. We're responsible for literally all of Affirm's data, starting with our production MySQL databases all the way through our batch and streaming systems, our data lake, Snowflake, analytics, all of that. So it's a big responsibility, lots of data, lots of exciting things that happen. My previous to Affirm, I was at Mozilla for about a decade. So a lot of data center level work and everything has moved over to the cloud now, which is very exciting.

It's very good. I want to highlight when we were talking the surface area that you cover on data, it's all data. All day long, we've been talking about here what's going on in the operational side, what's going on in the analytical side, you represent trying to bridge the 2 because you're responsible and looking across all of the data in the organization. So I think you're just in particular, well poised to be here for this discussion.

Yes. And it's very exciting to see how much things have changed and how much it's becoming easier for customers like us to do more because of players like you.

Awesome. Well, maybe let's spend a minute. I did my probably poor elevator pitch on Affirm in the opening, but I can say you guys are brilliant. You're everywhere with every purchase, at least that I'm trying to make as a consumer. It's right there when you're at checkout. But I would think it would be better for the group to hear from you about Affirm as an organization and your business priorities.

Yes, sounds good. I'll take a stab at that. So I think for those of you who don't know about us, Affirm is a payment network. We're primarily there to empower our consumers and help our merchants grow. Consumers basically use us to pay over time. And the 3 key factors there that we offer them are flexibility, control and transparency. I think transparency is a huge part of how we founded the business. We don't take any late fees, no hidden fees. And because of that, our incentives are very aligned with our consumers. We have to make sure that the credit that we offer them can be repaid. And so we're very focused on our underwriting models to make sure that's possible. And then overall, as a business, I think over the last couple of years, we focused on growth, both via repeat users. That's about 90% of our users are repeat users, and we want to keep that going, continue to grow that. And so yes, that's a little bit about Affirm.

That's fantastic just to give us that backdrop. There's multiple ways, obviously, that you could think about the importance of real time and data streaming and financial services, but really talk about the role that it plays within Affirm.

Yes. And I think just as I mentioned, it's a good segue because underwriting is so dependent on data. And data is our backbone. We basically use a lot of data to assess risk in real time and make a really quick decision, sometimes as fast as 15 seconds. And so there are 3 key things I'd like to touch on here. One is sort of high-quality fresh data, right? This is super critical to be -- for us to be able to operate the business safely. And then we use streaming in both front-end and back-end services to sort of create a feedback loop for us to ensure that our loan processing, underwriting as well as fraud detection are working -- really performant. The second thing that I want to quickly touch on is streaming for an analytics use case. For about the last 5 years, we've had a sub 5-minute SLA that takes production data to our analytics systems via streaming, especially for fraud operations. This is super critical because, again, if there's a fraudulent transaction, the longer time it takes for us to discover that, there's impact on a consumer, there's an impact on a merchant. So -- and the last thing I'll touch on is sort of Black Friday, Cyber Monday. That's a -- it's a huge critical time for our business. And your Confluent's high availability is -- comes really, really handy for us in these situations. We're talking more than 4 9s of availability, and that gives us the confidence that we will be up for our consumers during that time.

It's interesting when we were talking about some of the specific use cases that it solves for you. Underwriting was one, but you also had talked a bit about fraud detection. Can you elaborate a bit on that as a use case as well?

Yes, sure. I'll actually touch on the underwriting one first, and then we'll go to fraud. I think a little bit of a story here, right? Like if you think of traditional systems or traditional computing, we're usually used to doing things in a sequence. User enters, logs into Affirm, goes through a step of series of processes and then checks out in the end. Now you introduce something like streaming where right when the user enters, if they log in, I have some information about the user that I can then put into a stream. And by the time they come into checkout, there's been a bunch of processes that have happened in parallel to enable a seamless experience for both us and the consumer. I think that's really critical, and that's sort of the future we're going to definitely look more...

And does that close that lapse of time that you had that you can now paralyze it versus it being sort of linear...

That's definitely one of the goals. We've seen some early improvements. We're still thinking about investing more in that area. And then in terms of fraud, exactly the same thing, right? Like how quickly can we -- and as somebody else was actually talking about real-time data and like everybody wants real-time data. This is actually a really great use case where it actually makes a difference. And so having that sub-5-minute SLA, which getting that information, getting it into the right people, our fraud ops team can analyze, make a decision, that makes a big difference.

It's fantastic. What additional -- I know we probably talked about a dozen different use cases. What additional ones can you share with the audience? I feel like it helps to really contextualize the value of data streaming in the business.

Yes. I think one other thing that I can think of is as an engineering organization, we hold ourselves to a super high bar. And so one of the initiatives we had over the -- I would say, probably the last year was trying to understand how performant our endpoints are because sometimes the more performant they are, the better they experience. And again, our observability team used Confluent Kafka and Flink to actually set up like a real-time endpoint monitoring system. So all our engineering teams can understand what their latency -- what their SLAs are, how they're performing, their uptime and so can go ahead and improve stuff. That was really helpful.

That's incredible. Just giving the audience the opportunity to understand the breadth of -- from servicing the application process and the underwriting all the way to an observability use case in the organization. Obviously, you have to measure what you're doing here and understand the value and the impact of it in the organization. Can you share some of the metrics that are performance improvements that are coming out of this in a quantified way for this group?

Yes, sure. I think -- again, I think I've emphasized on BFCM quite a bit. So I'll start there. So for example, last BFCM, I think we handled over somewhere between just across that weekend, over 10 billion events with Kafka with a beat $10 billion. And our uptime at Kafka is 100%, right? We basically saw no loss in service. And on average, every month, Kafka handles tens of billions of messages for us. And to just give you an idea, like the loss on that is like incredibly tiny. So maybe basically 1 in 4 trillion messages is lost. And so I think the key thing for us is it allows my teams to focus on what they have to do, not have to worry about whether Kafka is going to miss something or not. So that's a good level of comfort we take that.

You've set me up for this because you just started down the path, but maybe what was the full kind of list of reasons to say why we're not going to do this with the open source Kafka, why we're going to do this with Confluent. If you can just net that out.

Yes. I think this is basically comes down to, I think, 4 sort of things, right? One is scaling. I think we had a lot of trouble scaling open source Kafka. Scaling is expensive, especially during peak events and Black Friday, Cyber Monday, that was a bit hard. Then we definitely saw some issues with the right latency, and therefore, that impacted our applications. And so we were like this is not going to scale long term. The version of Kafka we were using was only 3 9s. And so not something we could live with. We wanted higher availability and then also limited integration option. S -- wasn't compatible with Flink or Kafka Connect, which are both things we use today. And so I think for us, primarily, the way we function at Affirm is we look for partnerships, right? And I would say you guys have been great partners, not just because I'm on the stage, it's actually true. And you guys have been great, and that's a big thing. And then for us, the other 2 things from a business perspective are improving resilience. We talked a little bit about the uptime metrics. That's super critical. And also a fully managed serverless solution. That means it's less operational overhead on my team. Like I said, we can focus on things that matter to us. So I think those are the key things.

That's fantastic. You started to mention a little bit about Flink and stream processing. So maybe talk about you're doing some of that now, but for this group who's probably very interested in our earlier days here on that, maybe you can talk about what's next at Affirm with Confluent and data streaming.

Yes. I think about a year, almost 1.5 years back, we decided to pick Iceberg as our choice for our data lake. And I think for some of you who've been around, I consider this as the blue day sort of the table format wars. If you remember the Blu-ray versus HD DVD fight, Iceberg is Blu-ray right now. I think everyone is getting behind Iceberg. There's a lot of support. You folks are working closely with Iceberg, which is great. And so the next couple of things we're definitely working on or would love to explore is Tableflow. I think it's for efficiency, basically cutting down on an extra hop here and there. If you can just process right on the stream with Flink and put it into your data lake, you save a hop. And then one of the earlier presentations talked about managed connectors. I think that's a huge -- that could be a huge time saver for us. And so that's another thing we're definitely going to be exploring in the near future.

Fantastic. Well, I want to thank you again for joining us today and coming up here on stage and talking to all these folks. And I want to remind everyone that he will be joining us later too for the reception. So thank you so much for joining us and for being such a wonderful customer.

Thank you.

All right. While they move, I will move out of the way. Next up, Rohan is going to come up stage and be talking to you all next. So come on up.

Thank you, Stephanie. And before I start, a big thank you to Shyam and Talal. This was awesome, and thank you for making it all the way. All right. How are we doing? I'm going to take us home, and it's great to see you all. As you've heard from speakers today, Confluent is on a mission to set data in motion and be the central nervous system of data within an organization. And as we aggressively pursue this mission, we also want to build a world-class technology company. And in my mind, what separates the best technology companies in the world from the rest is their ability to drive durable and efficient growth at scale. In the last 4 years since we've gone public, our subscription revenue growth rates have increased over 3.5x. Our cloud revenue run rate has increased over 10x. When you look at our operating margins, we've improved our operating margins on the non-GAAP side by over 40 percentage points, and we've improved our free cash flow margins by over 35 percentage points. Today, we're a $1 billion-plus revenue run rate, non-GAAP profitable, free cash flow generating software franchise. And before I jump into the materials, I want to reflect on the decade we've had. Erica briefly touched on it, and I want to underscore a couple of points. When I reflect on the last decade, we've had these 2 distinct waves of growth. The first wave of growth has been all about creating this new category of data streaming. It was built on the heels of open source traction, and we commercialized that through Confluent Platform. Along the way, we built a business that was $100 million in revenue run rate, and we were one of the fastest to actually do it. Our second wave of growth was fueled by the introduction of Confluent Cloud, essentially taking advantage of the streaming opportunity in the cloud. And during this phase, we grew our cloud business over 25x from a $20 million run rate to over $550 million. Our cloud revenue mix increased from approximately 10% to over 50% in these 5 years. And when I take a step back, Act 2 was all about solidifying our leadership position in this data streaming category with Confluent Cloud and Confluent Platform. And along the way, we also achieved $1 billion in revenue run rate. And again, we were one of the fastest companies to actually do it. Today, the leadership team articulated the incredibly large opportunity that we have in front of us and more importantly, how we intend to achieve it. We expect all these drivers will set the foundation for our third wave of growth, which is Act 3. We believe that Act 3 will be all about Confluent being the data streaming platform of choice in the AI era. And as we enter into this next phase of growth, you can be rest assured that our business fundamentals around driving durable growth and driving efficient growth will be at the core of it. So let's start with some of the drivers of durable growth. I'd like to start off with we have a large and expanding TAM. The future of technology today is being shaped by 3 unstoppable forces: cloud, data and AI. Gartner is forecasting cloud growth to be at a compounded annual growth rate of 20% over the next few years. When you look at global data creation by organizations, it's expected to exceed 394 zettabytes. And I looked it up, 1 zettabyte of data is equivalent to 250 billion DVDs. And IDC is projecting AI spend by 2028 to be north of $750 billion. These secular tailwinds are actually supporting the really large TAM that you heard Shaun and Erica talk about. Our TAM has nearly doubled from the time of our IPO from approximately $50 billion to over $100 billion today. And the increased expansion in TAM is on the heels of expansion of our product portfolio, not only on the streaming side, but also in DSP with connect, process and govern. Erica touched on it earlier. We expect these DSP businesses to be large independent businesses for us. And when I take a step back and I look at it, our complete data streaming platform is uniquely positioned to take advantage not only of these secular tailwinds, but also this large TAM that we have in front of us. The second driver of durable growth is all around our leadership position in the streaming category. You've seen this chart before. More than 150,000 organizations are currently using open source Kafka, and we're monetizing approximately 5% of it. Yet we've built a $1 billion-plus business just monetizing approximately 5% of it. In my mind, we're just scratching the surface, and there is this incredible amount of land and expand opportunity ahead of us. First, when you think about land, again, we have 150,000 organizations using open source Kafka. And we're basically breaking up -- we are soaking up the world's Kafka with our offerings. When you think about 18 months back, we just had 3 pricing and packaging optionalities, basic standard dedicated on the cloud side and Confluent platform on the on-prem side. Fast forward 18 months, we've added newer offerings, which include enterprise clusters, freight clusters and work stream. Ultimately, our goal -- objective is to make sure our customers can come to us with not a few, but all their use cases. And they need to be at a meaningful ROI and an attractive total cost of ownership. And that's the objective as we look ahead. When I think about our leadership position in streaming, it is just evident with our outstanding win rates. When you look at commercial vendors that we are competing with, including CSPs and small startups, our win rates exceeded 90%. Against open source Kafka, we are delivering 10x better performance, 30x better elasticity, 10x better reliability while reducing TCO by up to 60%. For high throughput relaxed latency workloads, freight clusters can drive up to 90% savings for our customers. We are the clear leader in streaming, and we are winning on all fronts with our faster, better and cheaper technology. The third driver of durable growth is our multi-platform data streaming platform, multiproduct data streaming platform. I touched on this earlier. When I -- when we did a lot of work on the streaming side with our pricing and packaging options to soak up the world's Kafka. We've also significantly expanded our product portfolio to include DSP, which includes connect, process, govern and the soon-to-be GA Tableflow. When I think about our multiproduct suite, it's a one-stop shop for all real-time applications and AI workloads. And it is the Switzerland for deployment models. It is a Switzerland for cloud models and it is a Switzerland for all data destinations. I want to share a little bit of context on our DSP momentum on-prem. This is an important cohort for all of us to understand for 2 reasons. First, in the world of AI, we do expect our on-prem workloads to benefit as much as we expect our cloud workloads to benefit. Second, on-prem customers for us are the most mature customers. And their DSP adoption is a good indicator of DSP cloud adoption. So when you look at Confluent Platform connect momentum, we exited 2024 with a run rate of over $100 million, which is over 20% of the Confluent Platform business. This is a strong indicator of the potential we can expect from our Cloud Connect business. With 1/4 of GA for CP, Confluent Platform Flink, we've built a pipeline, which is north of $20 million. Overall, when I zoom out, we're seeing really good momentum on DSP for our on-prem products. Our customers are in the early stages of adopting DSP in the cloud. This is essentially a large opportunity for us that the management team, all of us have shared with you before. The early data shows that increased product adoption drives higher NRR and higher lifetime spend. Today, only 6% of our cloud customers use all our DSP products. And when you compare them with customers that are just using streaming, the net retention rates of this 6% customers is 35 points higher and their lifetime spend is 18x higher. Ultimately, when you think about DSP, it not only drives customer success, it also drives customers stickiness, which in turn will drive long-term growth for us. I want to share an example of a retail cloud customer. This customer spends over $1 million with us. In its first 2 years with us, they increased their spend 10x, primarily on the heels of streaming. In year 3, that is in the last 12 months, this customer has leaned into Flink, and they have more than tripled their spend. We'll be -- we look forward to sharing more such examples with you on DSP customer success stories. The last pillar of durable growth comes from our land and expand momentum within our customer base. When you think about our customer cohorts, all 3 of our customer cohorts are showing really good momentum over the last 6 years, that's as shown in this chart. Our total customers have increased to approximately 5,800. That is we've added 5,000 customers in this time frame. Our 100,000 customers have expanded and now we exited fiscal year '24 with 1,381 customers. This is an important cohort for us because this cohort contributes approximately 90% of our company ARR. And if you look at history, we've done a really good job of progressing these customers -- total customers to 100,000 plus customers. And then when you look at million-dollar-plus customers, we exited fiscal year '24 with 194. And taken together, 100,000-plus and million dollar plus customers, they are, when you look at as a percentage of total customers, all the pure count we're clearly in best-in-class category. When you think about the opportunity, the land and expand opportunity is very large. Obviously, the first part on the land side comes from monetization of Open Source Kafka, 5% penetration. You've heard from Shaun and Erica talk about it with our pricing and packaging optionalities that we are providing our customers. So that's the land opportunity. On the expand side, obviously, we've got a strong track record of moving our total customers to 100,000 and moving our 100,000 customers to $1 million. That's another growth vector for us as we look ahead. If I take a step back and zoom out and look at our customers from a cross-sectional perspective, over 42% of Fortune 500 companies are confluent customers today. When you look at different industry verticals, all top companies in each of these verticals are Confluent customers. And the bar to be a top 10, a top 20 or a top 100 customer at Confluent keeps going higher and higher. Today, to be a top 10 customer, you need to be spending at least $7 million with us, which is 5x higher what it used to be in fiscal '18. Today, to be a top 100 customer, you need to spend $1.5 million plus with us, which is 8x higher what it used to be in fiscal year '18. Also, the top 100 customer cohort contributes -- the NRR for this cohort is greater than 130%. Let's dive in and understand our top 20 customer base a little better. Our top 20 customer base is very well diversified. They come from 9 unique industries Almost 50% of these customers are hybrid, which means they are using both Confluent Platform and Confluent Cloud. On an average, these customers are using 2 products of our DSP, albeit very early days. And when you look at the unit economics, they've grown 25x since inception, and the bar to be a top 20 customer is $5 million today. So overall, when you look at the growth drivers we have, we're very excited around all of these drivers, and we are excited to execute on it. Next, I want to talk about something which is equally important. That is, as we look ahead, how are we building a business model to deliver profitable growth. 2024 for us was a consequential year. This was the first year where we were non-GAAP profitable as a company. And when you look at it in comparison to some of our other peers, we did it in 10 years. I would put us in the upper quartile, top decile of companies to actually get there. Consistency matters. And over the last 3 years, we've been consistently delivering efficiency. We've delivered over 40 points of efficiency in the last 3 years. And we've kind of looked at a wider range of software infrastructure companies. Again, we are right up there in the top decile. When I think about operating efficiency, when I think about leverage, first thing that comes to my mind is, are you delivering an efficient product to your customers? Our subscription gross margin profile is best-in-class, and we are operating at above 80%. When I take a step back and think about what are the drivers, there are 2 drivers. First, our Confluent Platform gross margins are software type, and we've maintained them at high levels, and we've been very consistent around it. Second, our Confluent Cloud gross margins, we are basically getting a lot of economies of scale in cloud as we are building out our cloud platform. And we're also driving incredible amount of operating discipline within the organization. Looking ahead, as we get more DSP, these products tend to be multi-tenant. It's going to be a benefit to our gross margin profile. Moving on to operating side of the world, the operating leverage. Each of our business lines, be it R&D, sales and marketing or G&A, we've delivered consistent margin improvement and leverage over the last 4 years. Starting with R&D, we've improved R&D as a percentage of revenue in the last 4 years by 6 percentage points. And this essentially goes back to the point I made. It's coming from the economies of scale that we get from building out this platform that we are building. From an investment standpoint, you've heard both Jay and myself talk about it. One of the core focus areas for us is we need to continue this engine of innovation. And for that, we need to be allocating resources for future bets. And that's something we will continue to do as we think about R&D. Sales and marketing has been a highlight. I know I've had a lot of conversations with all of you over the last few years around sales and marketing efficiency. I can tell you, over the last 4 years, we've improved that by 25 percentage points. And part of that is our land and expand model, and part of it is while early days, DSP adoption. You heard Ryan talk about it. As we shift from a single product company to a multiproduct platform, we will be investing in specialization and sales specialization. So that's going to be a focus for us from an investment perspective as we look ahead. On the G&A side, we've improved our G&A leverage by 5 points. Although the number is small, this is a 33% improvement in productivity. And we've done that by leveraging lower-cost regions. From an investment perspective in G&A, we do expect to invest in AI and automation to make sure we are driving broader leverage for the company going forward. Now I want to discuss where this opportunity actually plays out with our target model. The first thing around target model is our capital allocation strategy. This is not new news. This is very consistent with what we've had before. The first focus area is all around driving efficient organic growth. We want to invest in the business, but we want to invest in the business with that sharp ROI mindset. We want to make sure we are investing and we are making sure we are getting the returns for our investment. The second is having a disciplined approach to M&A. As I think about it, when I think about M&A for us, it's all about 2 things: it's accelerating our product road map and getting access to newer markets. And you basically do that by bringing in best-in-class technologies and world-class teams. And that's been our blueprint and that will continue to be how we think about it. And finally, returning capital to shareholders. As our business grows and we generate cash flows, we envision returning capital to shareholders. So this is broad brush what our capital allocation philosophy looks like. From an overall next 12 to 36 months, what are the key drivers of growth that I think through? There are 4 in my mind. First, we did speak about the streaming opportunity, and that's going to be a big land opportunity as well as an expand opportunity with over 150,000 organizations using Kafka, and our penetration is approximately 5%. Second, we're making this transition from a single product company to a platform, and DSP is a key driver of growth for us in the future. We exited Q4 with a DSP mix of approximately 13%, and we expect that to grow as we go through the next couple of years. And the math of it is all of the DSP products are in their earlier stages of their growth curve. So essentially, they are growing faster than the average company. So that's going to be a driver of growth for us as we look ahead. AI adoption, you've heard from multiple presenters today, be it agentic AI or be it generative AI, we will play a very important role in that ecosystem. When mainstream adoption of AI happens, we will play a role in that ecosystem. And finally, on partner ecosystem. We're getting to a stage where we need our partners to get scale and to amplify our message. Our partners also need us. We are providing clear technology advantage over others out there. And it's getting to a point where it's going to be a good marriage with respect to our partnerships, not only on the GSI side, the RSIs, strategic partnerships, so we are excited about that. So overall, when you take a step back, we have multiple drivers of growth over the next few years. From a margin perspective, you heard me talk about it, we've delivered 40-plus points of operating leverage over the last few years. Our guidance for fiscal year '25 is approximately 6% for operating margin. Our expectation for fiscal year '27 is we'll get operating margins in the range of 12% to 15% and free cash flow -- and long term, operating margins expected to be north of 25%. Free cash flow margins will be in line with the operating margin numbers that I just shared. As a company, the logical next path for us is our path to GAAP profitability. And the biggest difference between non-GAAP profitability and GAAP profitability is share-based compensation. We exited fiscal year '24 with share-based compensation as a percentage of revenue at 41%. Our goal by fiscal year '27 is to bring it down to the mid-20s, approximately 25%, and over the long term, to bring it down to mid-teens. You've heard me talk about it. Share-based compensation is a lagging indicator, the leading indicator is net dilution. Our net dilution in fiscal year '24 was approximately 3%. We expect to bring that down to approximately 2% in fiscal year '27 and below 2% over the long term. And as we do it, we'll get to quarterly GAAP profitability sometime in fiscal year '28. Let me wrap up by giving you some modeling points. I'm not planning to go through all of these in detail. We will be posting this on our website after the program, but I want to touch on one, which is the customer account. Starting Q1 2025, we will be introducing this new metric. We're calling it the core customers. These are customers that spend greater than $20,000 in ARR with us. And this cohort of customers contribute greater than 95% of our ARR. And when you think about this, you heard us talk about high propensity customers. This cohort will be a good representation of high-propensity customers. Going forward, we will be reporting this metric on a quarterly basis, and we will move to reporting our total customer count on an annual basis only. Okay. I want to summarize the discussions we had this afternoon by zooming out and looking at the big picture. Today, every company is a data company, irrespective of the industry they operate in, and how they harness that data differentiates between success and failure. When you couple that with the secular tailwinds of cloud and AI, Confluent has a very large $100 billion-plus market ahead of us. How you take advantage of the market is equally important. And we have the industry's only complete data streaming platform with differentiated and disruptive technology. And finally, we have a world-class team that has a track record of delivering high growth and profitability at scale. So with that, I'd like to thank you all for coming today, and we'll invite the management team over for Q&A.

When you need to ask a question, all right. And so it just takes a little minute here. So I have 2 microphones on the floor. Fahim has one, Tomy has one. When you get the microphone, we ask that you state your name and your firm's name and you limit yourself to 1 question to allow for more folks to participate. With that, let's go to Raimo.

This is Raimo Lenschow from Barclays. Jay, a question for you now that you had Ali on the stage, how do we think about that kind of the world between the 2 of you in terms of kind of meal-time batch, kind of what they kind of try to do with Spark, where is this where you come in? Like how do we have to think about that differentiation going forward?

Yes. Yes. Yes, I think it's a great question. Maybe at the heart of it is like, hey, is there some competitive overlap here. There's clearly some points of cooperation. Yes, there's a little bit of overlap, but not huge. Like at the end of the day, we serve kind of different constituencies. By and large, like Spark is used primarily by data scientists kind of in that analytics ecosystem. The kind of Flink processing is probably more in the operational space by and large. We felt that there was some smaller areas of overlap, but that's probably true between any 2 data companies you would pick. And the amount of synergy was really large. And so ultimately, not a huge concern.

Michael?

Michael Turrin, Wells Fargo. Rohan, I appreciate you stepping through the target model. I think one of the questions we're going to get is just high level how you're thinking about growth and the construct of that model. Are there certain levels you need to preserve to keep to get to those margin targets? And maybe just thinking about mix as your product set matures, how should we think about a large mature customer? And how much of their spend could come from DSP versus core streaming as this all ramps?

Yes, there are a few parts to it. I'll break it down. To start with, when we are guiding operating margin number, we've actually done the game theory around it. So with different levels of growth. So we feel good with our guidance that we provided for operating margin. So that's directly answering your question. On the -- if I take a step back and look at the growth drivers for us, I called out the 4 growth drivers and be it the streaming opportunity, be it DSP, be it the AI or partner ecosystem, these are all multiyear opportunities. This is not just a '25 opportunity, even I would argue that the opportunity gets stronger with every passing year for each of these areas. So that's how we think about growth at a broad level. I mean, from a DSP mix perspective, it's early days. So it will be not right for me to comment on it. But what I can tell you is you saw our top 20 customers. And our top 20 customers, on an average, 2 DSP products are being used by that cohort. These are the highest spenders with us. So it's early days. The potential is really big. So we'll keep you updated as we progress. But that's essentially a little bit of color.

Mike Cikos?

Mike Cikos with Needham. Question because you guys have done a great job as far as acquiring new logos in the last year. I wanted to get a sense with these target customers that you're adding, what has traction been for the broader platform? Are you seeing increased traction across DSP with these new logos? Where are they in their consumption versus historical cohorts?

Yes, it's a great question. So yes, we have a couple of motions. Obviously, we want to expand into the existing customer base that's already using Kafka with DSP. But yes, the nice thing about new customers is a new use case, and I think you get through from scratch and so they are a little bit more likely to start with everything versus kind of lifting and changing something that's already out there. And so yes, those are a great example of that. One of the things that Erica touched on also is the shift left pattern where we're talking about some of the integration into the analytics world. And this is an example where it's a specific use case, and we're landing with really the whole platform, the connectors that acquire data, Kafka, which is acting as the transport, the processing to land it in the right format and Tableflow to manage the integration. And so we're certainly excited about these patterns that actually bring it all together. One of the nice things about having the full set of capabilities is it actually enables these full end-to-end use cases in a much more powerful way. And we think that, that's a big unlock for our team as we build that capability. And you may have more to add to that.

Yes, that's exactly right. Yes.

Jason?

Jason Ader with William Blair. First, congratulations to you, Erica. I wanted to ask you a question about what you've seen, I guess, over the last several years in your role, what had been the main hurdles in getting more customers to convert from the open source Kafka to Confluent? And how have you sort of evolved the playbook to improve that? I guess you talked a little bit about that, but it would be great if you could talk first about like the key hurdles, like 5% in 10 years, like that doesn't sound great to me even though you're $1 billion business, but just if you could talk through that, that would be great.

Sure. I think there's 2 things, and we touched on them today. One is this consumption model. And if we go back to the traditional subscription model, there's a contracting process in front of acquiring a new customer. So what does that mean? So maybe you've done a technical proof-of-concept with the customer. But before you kind of final -- bring them on as an actual customer, you have to predict the production environment and size that and get the customer to agree to how much they're going to spend in the first year or in a multiyear agreement. So there's this contracting process kind of in front of actually bringing them on as a customer. In contrast, with the consumption model, we can tell the customer, you can just start using. You can use -- we have a PLG offering, which is a great way to start. We'll also let customers sign kind of a shell contract with $0 committed just to start using the product. And so what that means is we will catch them earlier in their evaluation cycle, oftentimes even before the POC, but it means that we've removed that friction and that need to predict what usage is going to look like. And so it's made it easier to acquire customers. So that's one is the consumption model. The second big one is now the portfolio of offerings that we have. And so as I referenced, when we were going to market, let's say, with cloud and the dedicated cluster strategy, we were finding there are lots of different workloads and some that were mission-critical, a dedicated cluster was the right model, but others might have less -- they might not have the same low latency requirements or maybe just really high volume, but the data isn't as mission critical. And so we didn't have a great cost-effective offering to meet those needs. And so Shaun and the team has done a fantastic job of expanding the product portfolio and price points. So now with things like WarpStream and Freight, we're acquiring customers in categories that where we weren't winning as much before. So that's fantastic. And then we're also acquiring new workloads where, again, our dedicated offering just wasn't a fit. So those are 2 things I'd highlight.

And if I may just add 2 more points from a slightly different lens. 5% penetration approximately and approximately $1 billion revenue, so the opportunity set is incredibly large. So that's one point. The second point is around the time lines. Open source Kafka is 13 years old. Confluent is a little over a decade old and us selling a fully cloud-native version of the product, give or take, a complete product is 6 years. So when you think about those time lines and with our cloud-native product, we continue to differentiate more and more with open source. That's also an additive in addition to what Erica said.

Sanjit?

Sanjit Singh, Morgan Stanley. As someone that had to stand up warehouses in a past career. I don't envy that and this vision, Jay, that you've laid out, for Confluent is simplifying the data infrastructure and building that bridge for customers between their operational state and the analytical estate, makes a ton of sense. I guess the question is that what's the signal of that vision is actually going to come to fruition? If I look at the ecosystem today, a lot of those vendors as point solutions they're still growing quite nicely, DBTs of the world, [ 5/10 ] of the world. And so is it a function of the open format, iceberg thing, that gaining more adoption, people building more lakehouses? Is that the thing that we should look to? Or is there something else that we should sort of mark-to-market to...

Yes, I think it's a great question. So I should be clear, like I wanted to make the platform analogy between data streaming and data warehousing because I think one of the things that made data warehouse powerful is, hey, it was the place everything came together. And the way that things are -- need to come together now for actually running a business is different. And now does that mean all the old stuff goes away? No, absolutely not, right? There's a whole set of use cases for data warehouse, continues to be valuable, there continue to be products in that space. That said, I don't think it's going to have the same strategic importance as that place where you can actually access all the data. It will have a lot of things, but that will be a projection out of this real-time world. In terms of how you can see it coming about, I think the biggest thing I watch is what's happening in these more advanced customers. This is really turning into a significant platform for them? Is it just some ingredients that they've sprinkled here and there? Or is it something they're really thinking about as a critical part of their data strategy. And I think increasingly, you would see that. And I think it's reflected in the growth of these larger customers. They're paying more because it is valuable because they see it as strategic because they see it as an important part of their data strategy.

Pinjalim?

Pinjalim from JPMorgan. I wanted to ask you a little bit of an odd question, I guess. A lot of the confident products reinforces the usage of some of the other products in the portfolio. And Flink is a good example. You're write a SQL statement, you write into the topic for Kafka. It seems like Tableflow is going to drive a whole bunch of other things together. Have you ever done any kind of work where you kind of figure out that amplification factor X percent of this usage results in Y percent of that usage because that has implications on the growth flywheel as well as probably much more on profitability going forward.

Yes. Yes, it's a great comment. So first of all, just technically, it's 100% true, right? If you run a connector, the whole point is to produce or consume streams of data to Kafka. So you're driving Kafka consumption. If you have a Flink job, the inputs and outputs are streams. Even governance, which doesn't directly generate data, we found with customers that once they can actually have a well-governed system, their ability to actually use that channel broadly across the organization really unlocks. So there are these kind of multiplicative unlocks. We've done some different analysis. I would say it's a bit early. Some of these products, I think, are just getting going. And so the easy thing to do is stand up like maybe one end-to-end use case built with everything, what does that look like. But it does end up being very use case dependent. And you would see this in like data warehouse workloads as well, right? So how much do you spend on your query processing relative to storage? It depends on how complicated the queries are, right? So I do think it's a useful bit of analysis for us to bring forward, but it will, of course, make some assumptions about the customer use cases.

Alex?

Alex Zukin with Wolfe Research. Maybe I'll going to zoom out just a little bit and say, I think a lot of investors are trying to figure out kind of who are the key enablers of the AI data stack. And I think you guys did a wonderful job laying out some use cases and some customer examples. But maybe, Jay, you have, I think, a differentiated vantage point around the evolution of some of these elements. And so maybe to you, how close are we to actually like large-scale enterprise adoption of either agentic GenAI use cases or non-agentic GenAI use cases? Like is that a '25, '26, '27 like when it scales? And how like important or impactful is that as part of the pipeline to like whether that's in the guide? Like is that required to get to the numbers you're targeting? Is that upside? And maybe 1 or 2 that you're particularly excited about that you kind of talked about?

Yes, it's a great question. So first of all, I mean, when we think about guidance or our plan, we almost never assume some discontinuous event. It's like, oh, and then AI is going to happen in Q3 and it's going to go way up. It's just not the way we run things. The -- yes, you see these as a build. Any new enterprise technology adoption is, it's kind of a slower build. I was talking with somebody about this even with data streaming, where it's been a layering of use cases over time. And so in an organization that may have a bunch of traditional software, a bunch of uses of Confluent. It's not like suddenly, they're introducing 1,000 things. They're introducing one thing, right? But that one thing is actually quite important. And that turns into 2 and it turns into 4 and in turns into 8 and that's kind of the nature of growth in these customers. So yes, we're seeing a really great set of these early use cases come through. The agentic stuff, I think is newer. I think that's more in the lab than in the data center as it were, but I think that's where it's going. So I've been watching this space since I was in college. Actually, the machine learning was my area of focus when I was in a PhD program and I'm kind of following it down I'm on the Board of Anthropic and get to see some of these different companies do different things. And I think these are very real and meaningful things being done in companies, and it's not everywhere and it's not a large scale. But if you want to see a use case that is very close to home for tech companies, look at what's happening with coding, where it's -- that one is very advanced in terms of the sea change in productivity for individual engineers that have gotten good with the new tools. And it's not like a 10% thing, it's much more. And so we -- how do we benefit from that? Well we're a software company. Obviously, we're doing coding, so that's great. We also sell to these companies. So Cursor is a customer that's, of course, taking that out -- a customer of ours that's taking that out as a coding product and selling it. And so yes, we see it on both sides. The faster movers are the digital-native companies. So that the people we will see adopt most quickly are these companies building AI solutions, tech companies adopting AI, they will move faster. The more mainstream enterprises will buy those products. And so we'll sell indirectly through them, but they will also build their own things, and that's happening actively now. I would think of it as a layering. It shows up first in that digital-native segment, which tends to be fast and chaotic. It shows up over time in the enterprise segment. I think we're seeing it now. So it's definitely a long-term tailwind. But yes, I wouldn't think about it as a Q2, that's when AI happens. And I think people who have thought that way have been, over the last year, have probably been disappointed broadly if you look at software and you're expecting some kind of inflection point. I don't think you've broadly seen that across software companies. But I do think you're seeing a build of very different use cases and you would hear this in customer conversations, you would see this in just where investment is going. And yes, so that's how I would think about it.

Kashh?

Kash Rangan with Goldman Sachs. Great to see you guys, first of all. And Erica, congratulations. Very few women have let go to market and had senior executive positions in the software industry like you have. And so thank you for that. A question for you, Jay. I heard Shane said only 1 question, so -- and all the smart questions have been asked. I got to think really, really hard about this. DSP sounds quite exciting. And the promise of Confluent as a platform has been to enable maybe a few killer applications. And the good thing about killer applications, you know this better than me that once those are established, they drive the use of the platform, you don't have to do anything. It's like Oracle. And frankly, every major platform that has been around has been driven by a killer application or 2 ERP databases, that's sort of thing, right? So is the right way to think about DSP is that it was on its way to enabling a bunch of killer applications and this thing called AI, came about and just distracted the hell out of everybody, us and everything sort of shifted, corporate priority shifted, all of a sudden became the fear factor, I need to do something in AI. So is it that AI is stealing mindshare from what seems to be an underlying inevitable development of DSP into a potential platform in the future? Or was that not going to happen and AI is actually coming to the rescue and you're going to see new AI use cases that make the promise of DSP come real? So did you need AI? Or is it been a distraction so far?

Yes, it's an interesting way of thinking about it. I mean, I guess, overall, we've been pretty pleased with the growth of DSP. So we talked maybe it was 3 quarters back, it was about 10% of cloud revenue. We talked recently it was about 13%. So it's a nice growth curve for us. So yes, I wouldn't say -- so first of all, it didn't seem like, okay, it was -- people were distracted it wasn't happening. It seems to be happening. The AI adoption is of the full platform. So it drives growth of kind of core Kafka and streaming. It drives growth of the other components. So I would think of that as almost an orthogonal concern. For Confluent, the use cases have been kind of broadly all the pipelines of data between different parts of the business. I would look at even this kind of Tableflow thing as an evolution of that. And then these applications that react or respond in real time. And those can be quite specific to a bank and they're absolutely killer within banks and every bank does set of things around risk and around trading and so on. But they can also be quite broad, right? So use cases around security and stopping the bad things, fraud, that's very cross-cutting. So yes, so that's been the kind of set of killer applications. The newer use cases as they're integrating AI, I think are a continuation of that pattern. Like I talked a little bit about this in the beginning that where we show up is these areas where you are kind of trying to close the loop in software in some sense. And in that sense, I think the AI use cases are definitely an accelerant. And I think that's an exciting thing for us. But -- so yes, I guess I wouldn't see it as necessarily a distraction or a critical save in that I'd say basically that stuff is on a reasonable trajectory and this is a nice tailwind.

Derrick?

Derrick Wood at TD Cowen. Can you give us a sense of how much revenue touches operational estates versus analytical estates? And kind of where you see that mix shift going over time? And then just as you lean more into the analytical side and table flow comes GA, can you just give us a sense on how you're going to weave that into the go-to-market motion?

Yes. Yes. It's a good question. So certainly, the use cases touching the kind of analytical world have been a smaller percentage. We don't have an exact categorization, but I would say probably it's sub-20% of use cases. Gut instinct, but that's kind of where I would put it. The -- in terms of the growth there, yes, we do expect that to grow, right? I think that what we're doing in that area is really interesting. And I think we'll see more adoption. And I think we'll see breadth of adoption, right, where we see deeper usage of multiple parts of the platform and a broader set of data that flows through it. So yes, it's obviously an exciting thing for us.

Awesome.

And was there a... I think I might have missed....

[Bringing tailwind flow to market. And on the sales side of the go to market question.]

Yes. Yes. So I think Erica talks a little bit about shift left, motion. I don't know if you want to talk a little bit about just kind of what we're doing as that enters the world.

Yes, kind of we're able to talk about shift-left and deliver a lot of value today. And then Tableflow like really connects the kind of enhances the value prop of connecting the 2 estates and making that a lot easier. And then, of course, with the Databricks partnership, that is just a great connecting piece there. So we think it just makes it even more real of connecting these 2 states. And we see customers like really lean in and say, oh, this would be incredibly valuable. So I think it's a pretty -- I guess what I'm trying to say is we're -- it's not a different motion. It just enhances the motion that we have today.

Go for it.

Austin Dietz with UBS. Sort of building up the last question, Jay. The conversion of the batch ETL world and the streaming world has always felt like a pretty slow role and you guys have put up a really healthy revenue growth despite that. So just given everything you've done around freight clusters, workstream, storage pricing and then given the need for more real-time data with AI, do you think we're at the point where the pace at which you could start picking up more traditional batch workloads or at least going after them could pick up over the coming year or 2?

Yes. Yes. So there's really 2 things when people talk about batch. There's kind of batch data delivery and there's batch data processing, right? And both happens. So batch data delivery would be like an Informatica or somethings that fills up data warehouses. The -- and there's obviously other products in that space. Batch data processing would be the scheduled set of data warehouse queries that run at the end of the day. And so yes, we're -- we think where there's display spend in both categories. The -- to displace some of the batch data movement, there's a set of capabilities we needed to have. One was a rich ecosystem of connectors, which we've built out. The other was the ability to do powerful transformations with Flink, which we've built out. The last, certainly for the analytics case, is this integration, really good integration into the analytics ecosystems where -- Tableflow. We obviously have that for a while, but Tableflow makes that much better. And so yes, I think that does make us very well positioned for some of those use cases. And the key point for us is when you think about the use cases around data movement, there's been really kind of a fragmentation with a dozen little subcategories of tool, each of which have a little bit of market share. What's happening with streaming is those are collapsing into a bigger market where the same stream of data can go to many different destinations without needing 7 different tools to take it to 7 different places. And that's actually a very powerful thing for customers who find a lot of pain in all this piping and it's a leverage point for us in that once there's a stream in the system, we can kind of resell it to multiple use cases as it were with the customer. And then on the processing side, that comes along naturally with the data. As the data lands in batch, then it's processed for a bunch of batch quarters and get it ready. Some of that is pulling up into the Flink layer as well. So when we talk about shifting left, that's kind of moving some of that upstream so that instead of delivering bad data to multiple places and cleaning it up, you're creating good data and making it available across the different states. So yes, I think we are now very well positioned to take additional share in that space.

Let's go to Will?

Will Power with Baird. As you target the next 95% of Kafka organizations out there, what are kind of the barriers or friction points you're running into in terms of adoption today? And how does that compare to the first 5% that you picked up? And I guess the final piece of that is how important is DSP and that further penetration?

Yes. Yes, I don't know if you want to take that or I can...

Go ahead.

Okay. Yes. I mean the -- it's a great question. So the -- I think Erica really got this right. So when you think about why do people make a change for the people who are using open source already? There's a couple of reasons, right? Part is about kind of cost and TCO, like, okay, is it a better deal? Part is about functionality. And that has 2 parts. One is through this core streaming capability is it better in some way. And the second is the things I need around that, do I get more of it. And so yes, it is actually all of the things you say incentivize customers to move and different customers will be more motivated by different things. There's customers in kind of a cost savings mode who are like, okay, the savings is the thing, that's what I care about. I'm all about the TCO model. That's really what I want to talk about. The DSP stuff is like a nice afterthought, like, great, sure. Once we save the money, there's new things we can do. There's other set of customers that are about their capabilities, and that's really what they need to unlock some use cases they're building. And all of these add up. And so when we think about what's moved some of the larger users of open source, it's very much that you actually need to check multiple boxes. You have to have something that's better. You have to have something as kind of faster and just a better offering, more reliable and then cheaper, right? So it's actually -- it's hard to the question of, hey, why didn't everybody already move? Like why are you only at 5%? That's why. But as you get that, then it's very powerful. And I think we have made a lot of progress on that and that kind of flywheel is turning faster now than it was before, and I'm very excited about what that enables.

Ittai?

Ittai from Oppenheimer. It was very informative and very persuasive. So good luck with the plan here. I guess just following up on your answer right here right now, Jay, -- it seems like you have a lot of confidence. And clearly, also what Erica mentioned before about how the motion going forward trying to convert the open sourcing to pay. It's going to be a little bit smoother easier, grease on the wheels, if you like to call it, right, to drive this forward. I guess, Rohan, why give up on customer account, why not stick with that and show us in the customer account that you can deliver on that? And the second part of the question is you presented very compelling TAM numbers here, which I think the CAGR is about the same as last time, 19%, if our math is right, -- but referring rates over 90%, why aren't you comfortable enough to already also give us mid-20s revenue CAGR growth over the next 3 years as well.

Okay. Great questions. So on the first part, when you look at our customer cohorts and total customers, obviously, we will still be providing updates on total customers. It's just not going to be quarterly, it's going to be annual. And when you think about how these customer cohorts impact performance, I think that's an important correlation you need to drop. For 100,000-plus customers, the contribution to ARR is approximately 90%. The new customer cohort that we will start reporting our core customers, the contribution to ARR is greater than 95%. And as a company, we've been talking about the consumption transformation. As part of the consumption transformation, our focus was to drive high propensity customers. And this quarter is a good reflection of that. So we're not taking away customer metrics, we're actually adding one, and we are providing one that's probably much closer from an organic momentum and monetization perspective, right? So that's -- on your comment around guidance, listen, I think when you think about the growth drivers for us, that doesn't change. And these growth drivers are multi-year growth drivers. So we're excited about, like you mentioned, the streaming opportunity, DSP opportunity, the AI as well as partner ecosystem. And when you're thinking about today, the -- if you look at our peers and other companies, nobody is guiding multiyear. There's a reason for it, primarily because there's just volatility in the economic environment. So it's just not a prudent thing to do, and our guidance philosophy has always been have a prudent approach to guidance. So that's how we think about it. Our growth drivers continue to be the same, and we're excited about the growth drivers we're going to be execute against it.

And on the customer count the -- I think the key thing to understand is like, well, there's a very low threshold. It was like $0.01. And it tends to -- as your kind of marketing tactics change, you're bringing in more PLG stuff just tended to fluctuate, we would, of course, look at these more indicative numbers internally, and it was less interesting. We just felt like, hey, without having a higher number that's a little more indicative, you're just not really getting the right picture of the business. And so that was kind of one of the points where we felt, hey, externally versus what we're looking at internally, we're just not giving enough for people to really understand the true trajectory for customers.

Let's go to Eric. One part question, please.

You saw my wheel spinning in my head there. Shane, I appreciate catching that in front. So I'll just give you a bigger picture question, Eric Heath, KeyBanc. There was one comment made, I forgot who mentioned it, but I thought it was pretty interesting. The comment was you thought Confluent platform would equally benefit from AI. So I thought that was intriguing. So I thought if you could just expand on that, that would be helpful.

Yes, I'm happy to touch on that. So the way -- one of, I think, the misperceptions about Confluent has been that there's kind of a transition where customers are just getting off self-managed software and getting out of their data centers and [ the cloud ]. In fact, what we've seen is Confluent Platform as a business has continued to grow and our cloud business has grown even faster, right? And so the reason for that is data streaming ends up having to span the different parts of the business. When you think about the customers that are adopting AI, it's actually a whole spectrum from very large customers that run their own data centers to small tech startups that are all in the cloud to traditional mainstream enterprises that have a big mixture of things. And so we do expect the benefits to be across. And that's true even when the use case itself is not being built in their data center. And the reason for that is actually AI use cases, as many of these use cases do motivate a lot of data flow. And so we end up having these outposts in customer data centers to collect data out of some of the older systems to power the newer things. So that would be true even today, even you're building some next-generation customer experience in the cloud, but sure enough, back behind it is some old crufty thing sitting in the data center that have a lot of the core data. We end up with a presence in both. I think Rohan talked about the prevalence of hybrid deployments in our large customers, particularly true in the kind of more mainstream enterprise space outside of digital native. And those are both parts of why.

Miller?

Miller Jump, Truist Securities. So you all have said you have the headcount you need to execute on your 2025 plans. But obviously, it sounds like there's a bunch of exciting sales plays that you're pursuing right now. So I'm just curious like how are you thinking about potential need to accelerate headcount growth to attack those plans? And I mean maybe how you think about that in the midterm model as well?

Yes, I'll be happy to take it. When we go through our planning exercise, it actually starts 6 months before the year starts. And that's when we are making capacity decisions for the next year. So entering the year, we typically have the capacity that we need to execute on our plans for the year. And then as we go through the year, as we are executing, we can dial it up, dial it down depending on how things are. So entering 2025, we feel good with the capacity that we have on board. And Ryan and Erica, both of them touched on our focus on specialization, selling the platform, continue to make little tweaks around consumption transformation, which is behind us right now. So that's how we think about it. And from a medium-term perspective, you're right. I mean as we think about 2026, we'll be working very closely with both Erica and Ryan to make sure what that capacity requirement will look like, and we will start sprinkling in investments as we go through the year.

Yun?

Yun Kim, Loop Capital. If you can just expand a bit on the go-to-market around CSPs and GSIs. You talked about technical specialization and the verticalization. Are they going to play a critical part in that buildout of that capacity? How much of your business is in partnership today with them? And I also heard that there's going to be some professional services investment that's going to be happening. How would that impact your relationship with GSIs as you roll that out?

Sure. Yes, I'll take that. So I think part of the question was around in our partnerships with CSPs and GSIs, are they helping bring specialization and verticalization to the market? And the answer is, yes, to a degree. I mean when we -- sometimes we'll look at joint solutions together that we can bring to specific markets, but we'll bring our go-to-market teams together. And we've done this, for example, for a long time in financial services and digital native and just be really focused about the accounts that we target and the offerings that we bring to the market together, both with CSPs and with the GSIs. With the GSIs, they're playing at kind of a different level. So it's a great opportunity for us to partner with them. We're talking to one of our large GSIs today about, okay, how can we go target the Fortune 500 companies that are Kafka users but not yet Confluent customers? How can we go do something there together? And then that's a great example where they are much further ahead than we are on vertical solutions and narrative. And so -- but then we come in with the technology offering and it's kind of a great pairing. So we do see a lot of synergies there with both the CSPs and the GSIs. And oftentimes, we're kind of triangulating together. We'll do something with AWS and one of the GSIs together to go to a target market. And I think there was a second part to your question that I've already forgotten.

Just the ramping of the professional services.

Yes, professional services. So we have long had professional services at the company. I think we started it in the early days because our customers were kind of early adopters, and we to do everything from staff augmentation to resident architects. We've always thought of our professional services as #1 kind of expert services versus we don't endeavor to build a big services practice at the company. So we do want to have experts that can partner directly with customers and partner with the SI community to get scale by enabling that community. And actually, we -- our percent of revenue coming from services has actually been decreasing slightly. And our percent of services delivered by third-party partners has been increasing because we really want to build scale. So we want to retain expertise, the right amount of expertise to Confluent, but we want to scale through investment in the SI community.

Last question. Let's go to Rudy.

Rudy Kessinger, D.A. Davidson. I appreciate you guys putting this together today. Firstly, Rohan, I also appreciate you letting us know in advance that you'll change the customer metric instead of doing it on Q1 when you pull it. So I appreciate that. I want to know on the DSP products. You guys obviously sound very bullish on it. You're saying each of them, you think you'd be its own large independent business. Do you have any ability to land with those products, for example, like landing with Confluent Cloud Flink or platform Flink with a customer who's using open source Kafka or using MSK. Have you guys thought about that, any opportunity to do that?

Yes, we haven't -- it's always a topic of discussion internally about whether we should unbundle the things we're selling and sell the ingredients a la carte, that could be a draw customers in, or keep it together as one unified product. I do think, at the moment, at least, we feel the most important thing is actually bring all these parts together into a coherent platform where everything just works and customers can just solve problems. And that's kind of the overriding concern rather than trying to pick it apart and make each thing work with every other combination. I don't know that, that's always the case forever, but at least for now, we felt like that's really the opportunity to build to success in this area. If you're thinking about pushing out the boundaries of what you do, that's the best way to do it. I think we've particularly seen that when we've looked at what succeeded in the data world. I talked about data warehouses maybe like they're old and going out of fashion. But the reality is, I do think Snowflake did a great job of building an integrated ecosystem. I think Databricks has done an amazing job of going from core Spark to an ecosystem that all works around Spark. I think conversely, the cloud providers have actually struggled a bit to not be just a bundle of 300 different 2 pizza teams that have all brought something to the table, but where the parts don't necessarily fit. They're also interested in that, but I think the recipe and the need from customers is, hey, come solve the larger problem for me. Don't give me 15 ingredients. It's actually -- I want a platform for my real-time data that's going to solve all the different parts from how I get it, how I process it, to how it flows to the transformation. Just come in and do that for me, that's a much more useful thing than trying to unbundle it into half a dozen open source components.

All right. Thank you for all the great questions. This concludes the Q&A and the webcast portion of our Investor Day. For folks on the webcast, you can now disconnect. Thanks again for joining us.