We're just delighted to have Confluent joining us at the Citizens JMP Technology Conference here in San Francisco. We're even more delighted that founder and CEO Jay Kreps was able to come and join us. The tall gentleman sitting to my left. How tall are you anyways? I'm 6'5". Excellent. Excellent. So I'm going to ask Jay how's business, then we're going to talk a little bit about his background, and then we'll do some more sort of standard questions. Let me get the how's business out of the way. How's business, Jay? Business is pretty good. Pretty good. Yeah. It's pretty good. Yeah. You know, for any of these cloud services, definitely the last year plus, there's been, I would say, a lot of just optimization and focus on cost and fewer new software projects. But we called this out in our earnings that it seems like that's stabilized a bit. I wouldn't call it a reversion back to where we were a few years ago, but there's some stability in that. And then the area that we're in remains really hot. It's an area of investment. Technology around streaming and real-time data is even more relevant than it was a few years back. And the rise of AI and the use cases around that kind of helped push that along even further. So this remains a big focus for our customers, and that's a great place for us to be. Yeah. And we're one of them. So two sessions before this, I hosted. Marc was in there. I hosted our CIO, Chief Analytics Officer, and Head of Consumer Tech. Right? And you guys came up. Well, Kafka came up. And I was a little hesitant. I'm like, so when you say Kafka, do you guys use Confluent? Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. They're like, yeah, yeah, yeah. We got our name in there. Yeah. That's right. That's right. OK. Where are you from originally? I was born in North Carolina, but I grew up actually north of the Bay Area in Sonoma County. Really? Yeah. Any wine? Was there anything? You know, it was probably more cow country at that point than wine. I think Napa started with the wine country, and it was just kind of becoming that as I was leaving. Yeah. And then you were just telling me. My knowledge of wine is kind of woefully poor. I don't drink it at all. Given where I'm from. My tequila knowledge is excellent. My wine knowledge is zero. I don't think I came away with that either. So I don't know. It was ill-spent youth, I guess. And then so we were just talking earlier. So Jay went to Santa Cruz undergrad. Were you majoring in computer science or math? Yeah. Majoring in computer science. But this I did not know. Here's a new tidbit. He was in the PhD program for machine learning at Santa. Yeah. That's right. That's right. Yeah. I was originally going to stay and do research. And then I realized that at least at that time, there wasn't as many breakthroughs in machine learning, but there was really interesting stuff happening out with the internet and Google and all these companies that were using data and had data. And so it was an exciting time to kind of get out in the industry. Yeah. It was basically just a bunch of linear algebra? Well, you know, at that time, actually, there was a whole array. This is 2005, 2006? Yeah. Yeah. That's right. There was a whole array of techniques. Interestingly, the stuff that generative AI is built on, the neural networks, had become very popular and then had become very unpopular because they were too hard to work with. And there had been a whole set of kind of more mathematical models and kernel embeddings and support vector machines, all these other techniques built on kind of other paradigms almost. And sadly, I guess, between all of that, there wasn't a lot of progress on predictive accuracy. Right? So there was a lot of development, but it wasn't really getting better. Yeah. And then meanwhile, some of the companies that had more data were just applying more data to problems and getting better results. And so I think that was a bit of the mindset shift kind of started around then. Probably in the next five years was when you saw companies kind of going back to neural networks and trying out some of these really interesting things out of Stanford where they're learning to fly helicopters and stuff that was not the kind of problem you could have taken on before. And the use of GPUs started around that. It was interesting times. Some of the very early academic work was kind of just after I got out of it. Yeah. So it was a funny thing. You can always go back to. At the time, I had a kind of strong belief that for machine learning and AI to really evolve, it needed a stronger theoretical grounding. So I was like, hey, you know, in physics, there was all these people who had all these little hacks. Then Newton came and kind of gave kind of all those hacks kind of went away. They weren't worth much. And so I thought, well, either you're the Newton in this area or you're kind of one of the people whose stuff gets replaced. Interestingly, the field advanced quite significantly with no change in the theoretical understanding of what learning is. It actually was just more computation of the same hacks. So it turns out it's hard to call these things, I guess, when you start out. So Tom Siebel was up here earlier. Do you know this guy, Stephen Wolfram? Yeah. Do you think about? Yeah. Yeah. He's a mathematician. Yeah. Mathematica. If you had to learn Mathematica in college. Oh, is that his? Is that Is that too? Yeah. Oh, really? I didn't know. So it's 101 pages long. I'm on page 2. All right. Yeah. I've already learned something incredibly awful, which I didn't understand what temperature was before this. Yeah. So anyways, what is ChatGPT doing? I had another thought I wanted to go after on this point. Oh. Totally off topic. I can do it anyways. Did you read the Elon Musk letter, OpenAI, or the complaint? I didn't read the complaint. I heard the kind of headline summary. Yeah. What are your thoughts on the headline summary? You know, I'm actually not sure. A lot of these I think this is around the kind of open sourcing. Should it be a private company? Should it be open development? I think it's super interesting. I think we just don't know. I mean, the way technology plays out usually takes a couple decades for it to really get fully adopted. And so there's a lot of thought about risks and adoption and all the pace. I think it's very hard to call some of that stuff. I'm just a dumb technologist, so I'm just watching it. I think we'll find out. Yeah. So maybe on the other side of the story, that maybe there is some benefit to not immediately open sourcing everything. Is that? You know, I think we just don't know. There's definitely technologies that were dangerous that we did manage to kind of smother. And refining uranium is kind of the example everybody would go to. And I think accurately, that was probably good that that was held back. I think with AI stuff, I don't know that that's a great analogy. But I think we have to really see how this stuff gets used before we know. All right. OK. So when did you join LinkedIn? 2007. OK. So 2007, you joined LinkedIn. You're a developer for LinkedIn. Right? Where does the beginning of Kafka come from? Yeah. You know, I joined LinkedIn really kind of with a similar charter where I wanted to use data. I'd come out of this kind of machine learning, had learned a lot of use of data. And as I got into the company, that was my intention, was to focus on that. But what I found was before you can really build a lot of fancy algorithms that use data, a lot of the infrastructure that just lets you get it was the first challenge. And for a social network at that time that was growing, that was a pretty big challenge in that the website had to scale. It was getting more complex. You needed more interfaces and products and mobile applications and just bits of software. And then to really use the data was kind of its own challenge. So I ended up working a little bit lower in the stack. How big was LinkedIn in 2007? You know, I think it was around 100-something people. It was pretty small when I joined. Then it grew pretty rapidly from there. So yeah. I ended up working a little lower in the stack than I had intended in some of these infrastructure areas and just building out kind of some next-generation distributed databases for the live serving, some of the back-end kind of data lake analytics capabilities. What I noticed was, hey, we have all these great technologies for storing data, kind of data at rest. Yeah, but how it flows between systems and how we can kind of act on it as it happens, we have almost nothing there. Yeah, and it was a little bit weird to me that some of the most sophisticated things we did with data, we would do in these big batch processes, where it's like, oh, at the end of the day, this big scheduled job runs and processes all the data and spits out some new results. And it just seemed like a little bit of something out of the era of mainframes. You kind of run the big batch job, and when it's done, it spits out the result. Whereas if you think about what LinkedIn is, it's a very digital business that's continually interacting with people and continuously generating data. So why wouldn't you process that data and react to it as it happens, where you could still benefit the users that are on the site rather than hoping they come back sometime in the future? What would be a really good specific LinkedIn type of example? Yeah. Some of the stuff I worked on, if you've ever been on the site and it kind of predicts people you should connect to. So I helped build the first version of that that was kind of based on real machine learning techniques to make those recommendations. Some of the people who have viewed your profile, this is people who've got a little ego. They really like this one. Like, oh, look at everybody who's viewing me. All the profiles that are related, a lot of the data-driven features, that was kind of where we started was, hey, how can you make something that's richer, that's more interactive? And a lot of that really puts heavy demands on the use of data. I think that became particularly relevant because I think for a lot of companies, that digital interaction went from something that was a little bit on the side to something that's really kind of a first-class citizen. They all had to do similar things. Yeah. And then you had a couple so how did you come out of LinkedIn and start Confluent? Well, internally, we created this layer, Kafka. It was all about harnessing real-time data. And at the time, we thought this was really revolutionary. And we open-sourced it, and it caught on in Silicon Valley. Some of the tech companies really started to use it, the Ubers and Airbnbs and Pinterests, all the kind of tech companies that were at that time relatively small but growing rapidly and needed to harness data in similar ways. And so we knew that it was a really big deal for that type of company. But we weren't quite sure if it made sense in the rest of the world or not. And then it just came about that we started to get contacted by just random organizations, from big banks to media companies to insurance companies, that were all looking at similar problems and had kind of somehow stumbled into this open source. Then they had this long list of new features that they needed and help that they needed. We were like, OK, realistically, there's this very big transition that's going to happen in the world from moving from kind of batch use of stored data to something that's continuous and real-time. That's like as big a transition as any use of data that's happened. We're kind of right at the heart of that transition. We're not going to accomplish that sitting in the back room of some social network in our spare time with just an open source project. There really needs to be a company that will invest in this, that will kind of build out software products and cloud services that will make this accessible to this type of company if it's going to really work. And so then we really decided to go after that full-time and left and started Confluent. The thesis was exactly that. That's effectively what Confluent's done since then is really build a suite of products around this kind of real-time data, real-time processing that are based around Kafka, which has become the standard in this area. Yeah. And where is it going from here? Where is this business headed? Don't worry about not this year, next year, not this whole time frame. Just where can this thing end up? Yeah. Well, from when we started the company, we had a very clear picture of what we thought was going to happen with software. You could see this as an evolution from maybe the early days of software where you really have these kind of disjoint applications that you've adopted that really kind of act mostly independently. They don't really interact with each other that much. That was certainly true at a point in time. Then you could imagine there's more and more bits of software. Now the requirements for software, it has to integrate. Big parts of the customer experience, big parts of how products and services are built and deployed and managed out in the world, the real drivetrain of a business is now all run through bits of software that all have to act in concert. Our view is that the role for this kind of streaming technology is really to take on. It's almost like the central nervous system in an animal or something that connects it all, that kind of brings the real-time impulses of what a business is doing, that lets you act on it. Much as the central nervous system is in an animal, that's an incredibly important platform. We believe that whoever gets that, that's probably the most strategic position around data that you can add to over time. Our view is every company is going to end up like that architecturally. You can see that happening out in the world now. Our goal as an organization is really to build the product that enables that and capitalize on all the opportunities around it. You can see that progress out in the world. In virtually every industry, there's been pretty rapid progress in the most sophisticated companies towards this kind of real-time architecture. There's been adoption really up and down in the not-so-sophisticated companies who are kind of starting on that path. The world is indeed moving in that direction. Awesome. All right. So it was two quarters ago that you guys had a hiccup, right? Yep. That was your first hiccup, right, as a public company for sure. As a public company. Was it your first hiccup as a CEO? No. No? No. I mean, look, Confluent is a young company that grew quickly in a domain that is evolving rapidly. We went through a bunch of big changes from really aggressively building out a cloud product while we were very early as a company and running both in parallel, a whole bunch of hard things that we did. So yeah, it was by no means the first difficult thing, not to mention the fact that the whole sector we were in went through a bit of an adjustment. Everybody. Yeah. Everybody. So tell us what happened two quarters ago. What did it feel like as the quarter went? So there used to be this company called i2. This is going way back when. Yeah. It had this founder and this guy. The founder's name is this guy, Sanjeev Sidhu. He told me once, he said, "Walrave n, you always ask the wrong question." He goes, "When a company misses, you always ask the wrong question." I go, "What do I ask?" He goes, "You ask why they missed. Yeah. He said, I go, "Well, isn't that the right way?" He goes, "You should ask them, when did you know? Because often when you know the answer to when did you know, you no longer need to ask why. Yeah. So two quarters ago as you were going through and by the way, very nice recovery from Confluent for anyone who hasn't looked at the chart. OK, so it's all good. But when you had the hiccup, when did you know? Yeah. Well, I think the stock reaction was largely relative to our guidance, not our results. Yeah. But the guidance comes off of what you booked in the market. That's right. That's right. There were several things that happened in parallel. It's been a difficult period of time to operate, particularly in the kind of cloud infrastructure space overall. Yeah, one of the things we felt was the customer behavior had changed fairly significantly from, say, a 2021 environment where there was a lot of buy-ahead type behavior to something where people were really conservative and kind of still adopting but more hesitant in their forward commitments. So one of the decisions we made was really accelerate the internal orientation of our go-to-market to consumption. Externally, we've charged customers on a consumption basis for our cloud product for a number of years well before we went public. But internally, like a lot of other cloud companies, our go-to-market was really oriented around kind of booking upfront commitments from customers for that pay-as-you-go usage. The peer set of companies that we would be similar to, whether it's the different offerings in the cloud providers or the Snowflakes and MongoDBs of the world, they had all made this transition a year or so ahead of us with different kind of pace. But for any company that has both a software business and a cloud business, you can't kind of jump right to the end state because your software business, you are indeed booking subscriptions upfront. But one of the things we felt was, yeah, this is not really where the market is now. And it's certainly not where it's going. Customers really appreciate the flexibility. And you want to have your go-to-market team really driving the adoption of new use cases, not just trying to book ahead everything that the customer might do in the next three years. That's a relatively big change internally. It may not sound like it, but you're kind of changing your compensation model. You're changing how you think about pipeline. We had done approximately 20% of that change already. But we wanted to do the rest of it really on an accelerated time frame rate. We knew that would have some impact. But we felt like over even a reasonable time period, it would end up being positive for us. That did mean a little bit lower guidance for the first half of the year. I think the reaction was partially that lower guidance. I think it was partially just, well, what does it really mean? Yeah. Oh, no. Yeah, that's what we're always trying to figure out. And so I think as we came into Q1 and reported our kind of Q4 results, I think there was a pretty solid recovery, which I think was more or less what we said would happen was what happened. Yeah. So how are the salespeople handling being paid on consumption? Yeah, that's gone well. I mean, it was a little bit rougher, I think, for some of the infrastructure companies that were doing that. Do we have any salespeople in the audience, by the way? Yeah, they all come to these. No, no. I invite them. I invite them. Oh, yeah. I invite my distribution list. So usually you get a couple. Anyone want to—no, no one wants to do it. Ross Kavinsky, is that you back there? Oh, right there. Well, how do you. One of the best software sales recruiters in the industry. All right. Yeah. So do people care? Is this come up? What's that? Do candidates go like, oh, I don't want to be on a plant where I'm paid on consumption? No. It all depends on what it all calculates out to. Yeah. Will I make money? That's the question, not what the metric is. Yeah, I think this is an evolution that's happened in our space. And I think it makes sense from the customer's point of view. Ultimately, the thing that's valuable to them is these new use cases and applications. And it's actually what makes sense to Confluent. Our revenue is, again, driven by these applications. And so it's a little bit more complicated to drive your go-to-market off that for a whole set of reasons. It can fluctuate hour to hour and day to day. But it ultimately lines up the company results, the valuable thing for the customer, and your go-to-market apparatus. So you are indeed driving these new use cases, helping unblock things, making sure that comes successfully to production. It's really different for a salesperson. I mean, it's great. But it's a really different. It's a little bit different. I think one of the things that's happened was, as a lot of companies did this, I do think the larger set of enterprise reps saw that change. Some probably fled those companies. But I think the ones who stayed, if you were at MongoDB through this, you did OK. You did just fine. You did OK, right? And I think that's true for our team as well. So I think we benefited from two facts. One, a number of companies had already done this. Two, if you look at the behavior of customers in 2023, this was very pronounced. Customers did not want to buy ahead. It doesn't mean that there were no new use cases happening. It doesn't mean that there were no deals to be had. But let's go around the organization and round up everything that might happen over the next three years and commit to it upfront. That was a hard motion to execute. And so I think as a result, the message was probably a little easier for us than it was for some of the peer companies who went earlier. Nonetheless, it's a significant change to make. We feel like it's gone well so far as we've rolled this out and switched over our systems. Obviously, this doesn't impact anything to our customers. It's not like they're paying us differently or anything in our offering would show up differently. But it definitely changes the motion that we execute. All right. What's the most important thing for you to get right over the next year? You know, I think that that consumption transformation is very important. The second thing that we're very focused on is the expansion of our offering and the usage of the non-Kafka components. So Kafka was kind of the first thing we started with. This is that kind of raw stream of data. But the connectors that plug it into your organization, the capabilities of governing that data, and then the ability to process it in real time, we have something called Flink, which is part of our offering that will just be going GA this quarter. And that set of components, we think each have really huge potential. And they're all kind of earlier on that S-curve ramp than our Kafka offering is. When we think about, hey, what's important for us to execute this year, that consumption transformation and then the adoption by our customer base of those additional components and starting to see that ramp, those are two of the most critical things. Yeah. What's going to be the biggest challenge that you have to overcome in the adoption of Flink? I think there's a fair amount of customer demand. A lot of it for Flink comes down to just any new cloud service. There's a lot to get right. A lot of the iceberg is beneath the water in these cloud infrastructure offerings to really get something that works at scale across every cloud, that checks all the boxes. We've done that with our Kafka offering. I think we're on a very good trajectory with what we've done for Flink. But there's obviously a lot of work to do. So this quarter, what are we talking about? We're kind of getting there. Yeah, that's right. You want to know the results ahead of time. Is that the? All right. We have 44 seconds. Any questions from the audience? Please. Are there situations where your clients might value they might value a sampling of the data rather than all the data that's passing through? And therefore, they might prioritize specific applications over larger volume applications? Yeah. It depends a little bit on the type of use case. A smaller portion of business would be kind of analyzing data where maybe you could downsample and just look at a fraction. But a lot of the application use cases are just software that runs the business. Like in a payment system, you can't just emit 80% of the payments and still serve the customer the right way. And so yeah, there's less of that type of optimization. I guess our use cases tend to be these production use cases that come out relatively well optimized. The thing that tends to drive us either to accelerate or not in the market is more the pace of new software application development. If that slows a bit, that's a headwind for us. If that accelerates, that's a tailwind. There's typically less ability to kind of dynamically turn on and off individual applications. You build some app. It does what it does. By and large, there may be some way to optimize it. But you probably would have done that in development if you could. All right, Jay. We're honored to have you here. Thanks so much. We really appreciate it. Yes. My pleasure. It's great. Great to talk. It's great to talk to you.