Elastic-CEO: 'Zo brengen wij zoeken met AI op orde binnen elke organisatie'

- Sorry, sorry, sorry. - Yes, it went wrong. Emotional customer information is sent via your private e-mail. And now you have a damage claim in your business inbox. The customer asks you because of our impunity to speak. - Sorry. - Sorry, it doesn't work. Hizcox does. With the most complete protection against cyber and tech risks. Go to hiscox.nl or ask your insurance advisor to our business insurance company, ICT. The big tech show is made possible by Orange Cyber Defense. From strategy to execution, your cybersecurity in trusted hands. Welcome to the big tech show, "Tech verandert onze wereld". Here you hear how, today from the press conference of Elastic, the American Dutch party that ensures that organizations within large quantities of data can still search and analyze and get results. Without that kind of software, a lot of institutions would drown in their huge data. I'm Joe van Buurik and with me, Ben van der Burg. Hello, Joe, good that you're here. By the way, you can hear us explain what's going on this week in Tech in our ex podcast. You can hear that from every Thursday. So do that for the next big tech show in your favorite podcast app. Now we're going to talk about Elastic and their software platform. To keep the data visible and to search and get results that are quite different from the past. Meanwhile, Elastic has downloaded the most vector database for companies worldwide. And they just announced how to build their own AI agent. All reasons to talk to the top man of Elastic. So we'll be switching to English. SSC, welcome to our studio, the CEO of Elastic, Ash Kulkarni. Well, great to be here. Thank you for having me. Yeah, thank you. Thank you for coming over to the Netherlands. Actually, the roots of your company are here. We'll touch on that later. I should mention Ben interviewed your colleague, Jeroen Berkerkamp, in our other podcast, The Technolog, earlier this year. How does Elastic now position itself in the technological landscape? So think of Elastic as being all about search and being all about AI. You know, one of the most challenging type of data to analyze and do anything with is unstructured, messy information. There are lots of technologies that deal well with structured data, lots of databases. But when you're dealing with documents, when you're dealing with, you know, machine generated data like log files, those are very hard to analyze. Those are very hard to find meaning and correlations in. That is what Elastic is designed to do. And so in the world of AI, when you need to connect your internal information with large language models for these large language models to be able to actually work within your enterprise context, that is where Elastic is now very widely used. And because of that, we are seeing success all around the world, not just in search and AI, but also in observability and security. Yeah, but it is, in my opinion, a fundamental change because when you have the database in 2010, 2011, it was a database structure. It was another way of structuring a database. And now with AI, it's more a context kind of thing. You look in context, "Okay, this, nah, this sounds really good with this part and this with this part." So you need to make a change in the database structure. How did you manage that to make that AI change? The biggest difference was how we searched. So in the past, it was all about searching based on text matches, right? So, you know, you looked for textual patterns in the data. Again, in unstructured messy data. It was an algorithm called BM25, that was the algorithm that was used for that kind of matching. With AI, now you can apply newer techniques like vector search, like hybrid search, to all kinds of data. And you can search not just in text match form, but you can search using these modern AI based vector techniques. And that's now making it possible for you to work with all kinds of data, with images, with video. We have customers that are using us to analyze and find things in video and so on. And it's that context again, that's so exciting. But was there an organic change in your company to search in such a way? Or did organic, so you didn't have to do-- It was all organic, it was all organic. So, you know, for us, the notion has always been to build on a single platform, build new capabilities into a single platform, so our customers can automatically start to get value from these new capabilities. So in the same platform that had the ability to do text search, we added a vector database functionality. And that was the big innovation that made us relevant in the AI space. So we did that about five years ago. You know, being open source is a wonderful thing because we get creative ideas from not just within the company, but from even our community. And that's always been a big source of our strength. So you mentioned this already, the two words, observability and security. How do those two go hand in hand? Because it's one thing to just be able to find anything you need within your company data. But you need to protect that as well. How do you make sure that both, you know, gets up in the right way? The most fundamental common thing in all of these areas that we play in is that the data itself is very messy. You know, you think about a log file. You know, one of our co-founders always jokes that a log file is a message that every developer writes to his or her future self. Right, yes. Because they know that someday they're going to have to debug their own program and so they put a message in a log file. But these log files for that reason are very bespoke. They are different by each application. So trying to find the right kind of meaning in it for root cause analysis to make sure that you understand why a system is not performing well. That's observability. But that same data can be used for finding threat patterns. That same data can give you an indication of is somebody using that application, that database, that system for something that's anomalous, that's abnormal and that's how you actually detect cyber threats within your organization. So one data, but different algorithms that serve observability and security. And that's the reason why our platform approach has been seeing a lot of success. Can you go a bit more in depth here? Because it's not just then about generating results in a search, but it's also about bolstering your security. Can you give some concrete examples of how that translates into practice? Absolutely, so we do a lot of work with public sector companies and public sector agencies, you know, the Dutch Ministry of Defense right here. They're your clients. They're a big customer of ours. We work with KPN and we do a lot of work with Dutch MOD. And they use us for security, observability and search. They have built an agent application that they call DEF GPT. Yes, I've heard of it. And that's built on elastic, but they also use us for securing their perimeters. Because our technology works in air-gapped environments. So it can work on private cloud, it can work on sovereign cloud, it can, you know, the data always stays within your environment. And by detecting threats the way we are able to, in your systems, in your data, we are then able to quickly stop the threat from moving around and doing damage. Doing damage like, you know, encryption for ransomware or exfiltration of data. We have the technology built in into the platform to stop those attacks. And today, for that reason, security accounts for over 25% of our business globally. Right. Yes, yesterday I spoke about, in my techlog, my other podcast about Palantir and they said, and there was said that they also work for the police and for the Ministry of Defense. How do you compare yourself with so with Palantir and elastic? We are the open company that allows you to do the same kinds of things that they are able to do. We are able to find relationships in data that can be used for all kinds of investigation purposes, that can be used for all kinds of detection logic, whether it be for cybersecurity, whether it be for fraud analysis, whether it be for protecting your borders, and also, you know, making decisions on how you do investigations in police situations and so on. So we are widely used by agencies all around the world for that. Yeah, and Palantir has a market cap of 400 billion. Your market cap is a little less than 10 billion. How you compare that? That's the opportunity ahead of us. Yeah, but that's a big open source. It has more value than closed source. How do you see that comparison? You know, that's the real opportunity because open source gives you a massive install base. But monetizing that install base takes time. Whereas a closed source approach to things allows you to go after fewer, larger customers much faster. Think of open source as a long game, if you will. You first make sure that you have a very large customer base and you slowly start to convert more and more of them. So I feel that our approach gives us the opportunity to be around in business for the next 30, 40 years, well past my retirement, but you can build a generational company on the back of open source. So that to me is the exciting part. Yeah, and we should stress you have some pretty high tier clients. I mean, you mentioned Dutch government and all, but we should mention eBay, Uber, Netflix. Very, so over 55% of the Fortune 500 are customers of ours. Right, OK. So because of the service you provide, but as Ben just alluded to, talk about how you generate your moat from there. Yeah, so the biggest moat that we have, even though we are an open source, we add a lot of capabilities that are only in our quote unquote paid tier. So all of our code is available in our public GitHub repo, but we depend on, you know, the IP laws and so on to make sure that nobody copies the stuff that is monetizable, that we charge for. And our moat is the fact that we are good at dealing with unstructured data better than anybody else out there. That's been our biggest. How come? How come? Why are you bearing this? Because that's where we started. That's where we started. We took this approach that when you're dealing with unstructured data, you need to have the notion of relevance, just like what Google did for data that's on the Internet, right? So when you're searching for stuff on the Internet, what matters is not a deterministic answer, but the most relevant top 10 links. That's what we started with internally for data that's within the enterprise. Now, how that's changed with AI is you no longer give top 10 links, but you give an answer. And even within the enterprise, that's our core strength because we work with every large language model player out there. We use their technology, whether it's open AI, whether it's Gemini, we are model agnostic, but you can take our relevance capability, marry it with a large language model through what is called context engineering, and now allow organizations like Dache Modi to build DEF GPT. Similarly, we have other customers that are building all kinds of agents that provide value and answers on top of their own data. So we are the retrieval engine. We are the context engine that makes chat GPT like experiences possible within the enterprise. So what are the buttons you control to make sure the clients are happy with the results? Because if there's one thing that is a very much a topic with, if I can call it semantic results, is that there's the nuisance of hallucinations or any correlations that don't match. Literally last week, we had a developer in here who designed an AI voting assistance, which are all the rage in the Netherlands because we had elections last week. But he was like, "No, it can be done right. You just need to watch this and that. What can you do as elastic to make sure there's no disappointment when a company or any client switches to the new way of searching?" That's a brilliant question because at the end of the day, these large language models are not deterministic. They're probabilistic. So by design, they are guaranteed to have some variability in different times that it tries to give the same answer. The way we control that is, again, through what is called context engineering. So we give the most relevant information to the language model to answer the question that's being asked of it. And then we provide the prompt that says, "Do not make stuff up. Only use the information that's in these documents that we are providing you to answer the question." And language models follow those kinds of prompts. What is important is the accuracy and the relevance of the data that you're providing it. And that's our core strength. At the end of the day, our whole value comes in and looking through all of your internal data and saying, "For this particular question, these documents are the most relevant." And another thing that we are very good at is being able to show you where did this information come from. So the data provenance, if a language model is giving you an answer, you want to know where did you get this answer from? What did you look at? And if you're using elastic, we can quickly show you that because we can show you what were the documents that were used in creating that answer by that language model. And that's why we get used so frequently because we are able to connect the dots between the source of information and the answer that was created. And is this then guarantee 100% accuracy in the response? As long as the data, so it's going to be accurate in terms of being related to the data within your organization. So the data needs to be sorted out well. Exactly. The data needs to be correct, right? So if you have, it's garbage in, garbage out at the end of the day. So if your internal data is faulty, then you're never going to get perfect answers. Yes, which model do you like the most at the moment? So we have discovered that we get different quality of results based on the use cases. For our security use cases, we tend to use Anthropics Claude model as the default. For other kinds of use cases like translation related use cases, we find that Gemini and OpenAI tend to give the best results. We find that open source models like Mistral and Lama are really, really good, except for the most complex use cases. So our customers will often use, they'll switch between different models based on cost and other attributes to get the best outcome. But then you're always dependent on those models, when the token price will increase, you have less margin. Well, the way I look at it is, one thing that I'm very confident of is the price of dealing with tokens is going to keep going down, right? So you look at the investment, exactly, look at the investment that's going on, I think that's going to keep going down. But that's why also we're betting on open source. So I'm a big believer that open source is just like it brought down the price of all kinds of things. Database costs, operating system costs and so on over the years, that open source is also going to bring down the costs of models. Okay, open, yes, so open source models, open weights, but you can still be squeezed because you mentioned Gemini and Tropic. So you can be squeezed between those parties. At the end of the day, when we look at the overall mix of the spend for a solution, you know, I would say that typically the majority of the cost is spent on the model inference, right? That's a that's a given, like without a doubt, you'll probably be spending 80 to 90% of your cost on the model. Whereas, you know, 20% of the cost will be spent on elastic or something like that. But that's okay, because these are huge markets. These are massive markets where 20% can mean a market opportunity worth hundreds of billions of dollars over time. That's how we continue to talk to Ash from Elastic after this message. Since the beginning of 1900, we have been focusing on the technology in front of a store. With a focus on complex energy demand pieces and industrial automation, we make a nice morning possible with our customers. Look at how we approach this at www.batemburg.nl. Batemburg Technique, smarter focus, brighter tomorrow. How attractive is the Netherlands for companies and entrepreneurs? Is Den Haag really going to solve the housing problem now and will America or China play the game in the world tournament? That and even more are heard in daily courses, the daily podcast of the FD. Within a quarter, we talk to you about everything that is played in the financial-economic world. Every working day, a new day course at www.fd.nl or in your podcast app. The day course will be made possible by AB & AMRO. Your plan? Our expertise. Preferred banking. And now we will continue our conversation with Ash from Elastic. What's the use of it when you're really proud of it? You thought, "Whoa, did we can get this out of it?" I'll give you an example that I think is probably one of the most satisfying from a human perspective. And I'll give you an example of real monetary value. From a human perspective, like one that I am most proud of, and this was something that somebody from the Department of Homeland Security in the United States presented at one of our events, one of our user conferences. There's an agency called Homeland Security Investigations that's part of the Department of Homeland Security in the United States. And one of their functions is to prevent human trafficking. And when you're dealing with investigating human trafficking situations, just think through, you know, you've watched enough CSI, you've watched enough crime dramas on TV. It typically starts with somebody calling a hotline and saying, "Hey, I saw some young person being, you know, picking what seemed like against their will at this bus station, at this airport, whatever." And then when that hotline comes in, then the investigation kicks off, somebody's looking at CCTV camera footage, somebody's looking at calls. Once you identify a person of interest, you have to look at their social media, you have to dig through so much information. They built an application where they're doing semantic search across all of these different sources of information and correlating them. They did that on Elastic. They claimed that they were able to increase what they call their "solve rate," the speed with which they're able to do an investigation by 40 times. Imagine that, right? 40 times more effective agents in solving these kinds of problems. To me, that was amazing. And a hundred billion dollars. So it adds up as you look at it across multiple customers. But customer support, we are seeing a tremendous amount of uptake in terms of using AI to actually solve customer support use cases. And this is why you've launched the agent builder recently. That's exactly right. That is an interesting one because we have a lot of discussion that adoption is really difficult. Well, now you say, well, adoption is, you don't say it literally, but that adoption is okay. We are seeing very good adoption, not just okay. We are seeing excellent adoption in these use cases. I think the most important thing is you have to use, you have to think about the right use cases. Like, think about customer support. When you're dealing with customer support, a customer calls you. Your answer isn't going to be the exact same every single time. These aren't deterministic use cases. You are trying to understand what the user is asking for and then try and figure out what's the best answer I need to give this person. What's the best possible next step? These are probabilistic systems by nature. That's where AI works very well. If you try to use AI for something that's very, very deterministic, then you're not going to get a great answer. So customer support I found has taken off really, really well for us. Tell me, please, what kind of customer support in the Netherlands use Elastic because my experience is still really bad with this. We are working hard to do more in the Netherlands, so give me a year and then maybe next year I can tell you more. Right. Well, speaking of the Netherlands, what typical signs of Dutch culture does Elastic still have? Or is it more of an American company? You know, we are still a Dutch company. Our headquarters is in the Netherlands, we are Elastic and we, our IP is all in the Netherlands. Yeah, it takes reasons. Yeah, so not just tax reasons, actually, because we pay taxes in every country that we operate in. Right. But we have more engineers in Europe than we do in any other part of the world. All right. And about the culture, is it the direct code? Can you tell us something? So our culture is really, you know, best represented in what we call our source code. So these are some principles that our founders wrote and we've actually added to them. But probably the one that I'm most proud of, I think represents Elastic the most, is humble, ambitious. That is one of our, you know, six source code items, our values. Right. And we try to be that way. Like, I'm very proud of who we are as a company, but most people who get attracted to Elastic are not folks who are about themselves. Like, I remember you asking me earlier, are you like a tech bro? And my answer was like, I sometimes fear that I might not even realize that I'm becoming one. But we truly believe in that value, that notion of being humble, but being ambitious, right? We truly want to build a generational company. I want to build a company that, you know, my kids can work at. Like it's something that can outlast me, but humility is very important. Another one is, you know, we really believe that we should have time to work the way you want to, right? So we hold ourselves accountable to a very high bar of productivity and so on. But everybody is different. We need time to work our way. So we've always been a distributed first company, right? We have employees in over 43 countries at this point. We don't expect people to be in offices. We have figured out how to be incredibly productive being remote. Some people prefer to go and go for a long run in the morning and they prefer to start their work at 10 a.m. in the morning and then they work till late. That's fine. Others, I prefer to start my work day at 6 a.m. in the morning. And that's OK, too. We are big believers in letting people be how they are, who they are, just making sure that you actually deliver to the outcomes that you have committed to each other. I think those are, you know, I don't know what you would describe as Dutch values. I think those are good values. I think this level of freedom, if I can call like that, definitely applies. I would say that that's great values for your company. But I would argue at the same time you have clients which are using your technology to make sure AI takes over some jobs and some people will lose their jobs. How do you look at that? You know, at the end of the day, I've been in this business for the last 30 years and I'm a techno optimist. When I started my career, the internet was in its early days and we were expecting grocery stores to go out of business. We were expecting bookstores to go out of business. We were expecting half of, you know, known businesses to go out of business. Look, what happened was because of the internet, new business models were created. Some things did go out of business. Right. Bookstores did go out of business, but grocery stores actually thrived more because now you have last mile delivery. That's better. So much changes every time there is a technological shift. But I believe that we human beings are really good at figuring out how to work with new technology to create new kinds of jobs, new kinds of opportunities, new kinds of services. I think AI is going to do the same thing. We will go through a period of having to retrain ourselves, you know, really figure out what are these new kinds of jobs. It's going to it's going to happen. I think some jobs will change. Some jobs might go away. Other new jobs will get created. And I'm I'm a big believer because I've seen it happen now at least three times. Whether it was mobile computing, cloud computing, Internet. And what would you say is then the way to realize this? Because I feel a lot of people who may be, you know, having to deal with a layoff are are struggling to reinvent themselves. It's really important for businesses not to think of this as an opportunity to cut costs. I think it's important for businesses to think about this as ways to accelerate growth by using the talent pool that you have in other ways. That's what we do, right? So we might hire fewer customer support engineers, but we want to make sure that as we are growing, we are expanding, we take some of those same technical people and now start to, you know, train them on how to be great developer evangelists, how to be great at helping our customers expand what they're doing with elastic. It's the skill set that you want to leverage and think about new ways because that's going to drive growth. So rather than thinking about everything as a cost exercise, I think about it in terms of if I have great people, what more can I do with them? How can I get them? How can I motivate them to take us even further as a company? Yeah, that to me is a is a better approach in life. And I think that's, you know, the best companies always do that. Yeah, they don't look at layoffs. They look at way to implement those people in new ways. That's right. What's going to be the biggest challenge for elastic over the next few years, Ash? A complacency. That's the thing I worry about most. Really? Yeah, because at the end of the day, look, the space is moving so fast. It's we are not, you know, with AI, things are not changing every year, every six months, every month. Things are changing every week. Like every week, there's something new that comes out. New models, you know, new capabilities because of reinforcement learning, new approaches to provide context to the problem. So with each of these, we just need to keep, you know, making sure that we are disrupting ourselves, right? And that takes a level of determination on how do you keep reinventing? And if you stop doing that, if you get too happy about your results and, you know, how you're growing, etc. I think that can be bad. So that's what I worry about. Honestly, when the world is moving this fast, you got to move faster. Yeah, but complacency is more a culture, more a mentality kind of thing. Yeah, it is. But it's now we are 4,000 people. We need to make sure that we are embodying that for more people. That's right. That's exactly right. Yeah, they are humble and eager and that's it. And they are always trying to figure out how do I disrupt myself? Don't wait for others to disrupt you, right? That takes, Ben, that takes a special kind of confidence, self-confidence in people. How did Elastic disrupt itself for the last time? The most recent thing that we did was we launched our serverless product a few years, a couple of years ago. And, you know, we used to in the past to get this complaint that Elastic takes a lot of effort as the size of the data store grows. It takes a lot of effort to manage it. And so we were thinking of ways to simplify and, you know, it just felt a lot like it was cosmetic surgery. And so, you know, a few of us, our co-founder shy, our head of products, Ken, I, a bunch of other people sat down and basically asked the question. Fundamentally, if we were to rewrite Elastic Search today, let's say Elastic Search did not exist and we were writing it for the first time today. How would we write it? And shy, who was the original author, is said, you know, when I wrote Elastic Search, we did not have S3. We did not have object storage. That's why I had to build an entire architecture on disk-based systems. And that is the root cause for all the complexity in managing this data store as it grows. OK, now you have S3. If you had to rewrite it, how would you rewrite it? And that became the genesis of our serverless product. So, shy led the effort and, you know, it took us almost two years to come up with the first version. And that was very, very disruptive because we knew that once we came out with that in the next three to four years, the majority of our customers will move from our current architecture to that, which means we have to think through how we enable that migration, how we enable that transition, how we retrain users, a lot. But without that, we would never have been able to create something that is truly like a lovable experience, right? And that, that is, I think we are determined to keep doing that to ourselves every few years. Now you're also a corporation with AWS, of course. Are you thinking about building that yourself in the future? Building our own cloud? Yes. You know, the bigger, I mean, then maybe it's worthwhile. Yes. So, so our big model is we work with all the three hyperscalers and more. But we also make sure that you can run Elastic in your own data centers. Actually, what I am seeing is there is a greater movement towards sovereign clouds. Oh, yeah. There's a greater movement towards, you know. You've been to Brussels to talk about it. That's right. I was, I was meeting with the director general there responsible for directional Viola. And it was, it was amazing to see all the things that are going on in the EU to make sure that there is a better understanding of, you know, how to have more control, more governance, more provenance of the data. And our architecture just naturally enables that. And given the fact that, you know, we are headquartered here. We are a company that has so much of our presence in Europe. You know, we feel that there's a real value in us collaborating with the teams here to do more. Yeah. So, yeah, it's very exciting. Yeah, we're going to hear a lot more from you over the next few years in Europe. Data coming 30, 40 years. Coming 30 to 40 years. It might not be me, but it's my future. Your future me. OK, well, we'll see. Your kids, you mentioned, should be working on Elastic as well. Thank you so much for your views and your time, CEO of Elastic, Ash Kulkarni. That's all for now. Listen to the big show, just as a podcast, because from now on, you'll see our analysis of what's going on in the world of tech. Until the next big show. Bye. The big show is made possible by Orange Cyber Defense. From strategy to execution, your cybersecurity in trusted hands.