Enterprise AI is failing on garbage data, not on models

[music] [music] Ladies and gentlemen, welcome back to another episode of Beyond the Bubble podcast. As you guys know, we're getting knee deep into AI. Uh, and we're of the opinion that obviously uh the narrative around AI is great, but it's also very very bubbly and it might pop at any given moment. But as uh I've always mentioned, I think the underlying technology is absolutely phenomenal. I think there is uh you know what we're the changes that we're going to see in our society in our economy uh are going to be so foundational uh that it might completely change the way that we do a lot of things particularly in business as well. Uh obviously what matters there now is uh what are some of these people who are doing the unsexy more uh you know uh productive more powerful work but work that can actually uh you know make that long-term impact even if that might not sound like something oh as if this new tool is going to completely uh destroy Hollywood sort of a deal. Um, one of those folks here uh is uh this gentleman who uh co-founded Clario, which is a tool. Personally, I was a data analyst by profession. I worked a couple of years at Terodata, which was a sort of legacy, you know, ETL uh data integration uh business uh you know uh intelligence company. So I know how powerful, how uh huge this industry is. One of the problems that we used to face at the time uh was that we were all about structured data. you know, unstructured data was our kryptonite. Uh we hated people who would have these tons and tons of data lakes full of all this crappy unstructured data. Um and yet now in the age of AI, it is you know data is everywhere. It is being created everywhere. It is being processed everywhere and it is not only there is definitely a possibility of taking all of that data and working on it, but it is obviously costly as well. We're talking a lot about token costs these days and uh this particular gentleman has a very very exciting startup that not only you know classifies your data, structures it, cleans it but also saves a lot of your token costs particularly for enterprises who are kneedeep very very into AI but the cost is killing them. Today with us we have Ysef Khan who is the uh co-founder of Clario and I believe also the CEO uh running the ship over there. you. So, thank you so much for taking the time out and joining us. >> Thank Thank you for making the time. You've built a phenomenal platform and uh you know the kind of industry requires forums uh and conversations like this and so really grateful to you for being able to sort of drive that train uh and keep a lot of activity and you know founders like myself being able to be given a voice. So, thank thank you for making time and thank you for the opportunity. >> Absolutely. you. So, I'm going to just start off uh trying to get a bit of background of what you've been up to, my understanding is you were in the investing space uh just a little while ago. So, what actually brings you from that space to building a startup right now? What's your uh near past background and tell me a bit about Clario? Uh I I gave a bit bit of an intro. I just want you >> I appreciate that. Thank you. Actually, you're wearing our corporate color, so we'll probably do a a blog post with you eventually as well. So thank thank you very much for that. No, so just a little bit of background. I'm just a kid from Karachi who's trying to make it in the world. So I was I was born in Pakistan. I grew up all over the world from countries like Egypt, you know, Nairobi, Kenya and the UK for several years and then moved to San Francisco Bay area just over a decade ago. Uh and for a large part of my career, I was an operator as a CIO uh overseeing it and also as an intimate at multiple companies. And so uh my responsibility has really been to be able to drive uh technology transformation at multiple companies but largely uh overseeing IT and security teams combination of business applications data platforms you know driving a number of initiatives and uh you know was part of the executive teams at companies like Qualis and Pure Storage from 600 employees to about 4,000 and and IPO and then worked at early stage company called Move Works which was recently acquired by Service Now uh and then had a brief stand as CIO at Automation Anywhere and then prior to founding Clario with my co-founder Madivora who's a CTO of the company uh I was a venture investor as a general partner at a Silicon Valley VC firm called Ridge Ventures where invested in early stage uh enterprise and AI companies typically at seed and series A uh and then last year I made the decision that it's a very exciting time in the industry uh I you know have a vast network of CIOS I was a CI IO of course and understand IT uh organizations very very well. Um and in all of my research I discovered that the failure of AI projects happening in enterprises was caused by a number of reasons and one of those reasons uh was data. Um and when I started to look into this I thought to myself if I was a CIO in 2026 and my CEO and my executive team my board had said we need to drive more AI where would I fail? Um, and I've tried to solve the problem that Clario is solving over the course of 15 years as a CIO. Um, and so I discovered that the only way to try and solve this is to literally try and do it myself. And that resulted in my decision in co-ounding a company. Uh, and uh, and here we are. >> That makes sense. I just want to quickly understand when did you join uh, you know, formal workforce back in the day right after university? What uh, what year was that? Oh, it was even before that actually. U so my first job uh what it's worth was selling chocolates, imported chocolates at school when I was 13 years old. Uh I think the job I did prior to that was being my dad's assistant uh uh after school, showing up at his office and filing papers. Um so I've been working for as long as I can remember. Um, and I've worked in all sorts of jobs, but I remember at I paid I, you know, I worked uh in I had three jobs whilst I was at college paying for college tuition at King's College. Uh, one was being a uh working in a video rental store when video tapes and rental. Uh, for those who uh uh don't know about that, it's a it's a it's a piece of equipment that's where tapes and that's how this was pre-Netflix. [laughter] I have to remind people what this is sometimes. It's astounding from a generational standpoint. I worked for an import export company uh to understand and understand trading. Um I've helped manage a theater uh at an event conference venue. I've been I've I've ripped out chewing gum from carpet floors. Uh I've run a bar even though I don't drink. Uh so I think they trusted me to do that. Um so I've had long experience. My first career in tech was moving working for internet startup um uh in my second year of college. Uh it was in 99 and it was a company that just got funded. It was the first time I'd heard of the term venture capital. You know, the internet was uh all the rage and this company wanted somebody to help them move boxes into their office for a day. So that's all I and I showed up and I was like, "What is this place? Why are people so happy and what's going on here?" And uh then I heard the term venture capital. I heard that this company was only 6 months old or a year old or something like that. And all of a sudden there were 200 people and I said, "How does this happen? This is magic. And why are you people so happy? I'd like to be part of this." Uh and that got me going as a a career in tech. And so um and then I just I just kept one one day turned into a week, one week turned into an entire summer. And then I had a full-time contract um to join as an IT project manager ultimately. So I'd worked in IT and technology for the longest time. >> Makes sense. So um it's fascinating because your career in IT sort of was like an accident. Your your education was in >> I'm the accidental C I'm the accidental CIO. I say this as the subject of my blog post. So you're absolutely right. I I've accidentally landed into this. I I don't know how but here I am. Yeah. And it also coincided with uh with the sort of peak of the market right so 99 you joined uh 2000 mid 2000 was when that sort of position ended according to the uh your LinkedIn I'm curious were you at the level that you were exposed to this sort of pump and dump that happened at the turn of the tide for that particular technological revolution and the reason why I mentioned that is because a lot of people are now uh you know comparing what what's happening with AI with what sort of happened with with with the internet early days of internet right and contextually it's very easy to explain the difference between how Wall Street views a certain technology and how actually that technology functions and then creates value over time because right now there are two extreme positions either AI is so amazing it's going to change the world and make everybody job free uh and and and full of uh prosperity or it is going to be dystopian and it is going to make everybody job less um when the truth might be somewhere in the middle and I'm curious to understand if there is any correlation to the sentiment to as to what you experienced in the first 5 years of your career. >> It's it's it's a very very good question in terms of parallel and like the way you you framed that actually. So um there is similarity. So let's look let's talk about similarities. Are we in a bubble? 100% we're in a bubble. Uh I think the difference now this has been a much longer bubble than the than the previous one internet. Uh second is there uh was that when that bubble burst um it was really really detrimental for an entire industry. It literally you know the stories I heard before I moved to Silicon Valley about how bad it was and how drastic it was was real. Um and but in this case I just believe that it is a much bigger industry. It's a much faster industry. It's a much richer industry as a general rule. And I think it's a more exciting industry. So I think the risk could be adequately managed. Do I think that there is a a valuation inflation in a bunch of areas? Yes, I do. I think some I think there's plenty of that. But I think that's been the case in any market momentum in a specific industry, whether it was uh crypto or other areas in some way, shape, or form. So I I think the the beauty of the free market is that it is a free market that ultimately equilibrium starts to bring some correction because of forces that sort of come into place. The exception to that has been the 2008 financial crisis when the government had to step in at a national and then a global level with its checkbook. Um and so heaven forbid we come into that situation ever again. I don't wish that on on on for us or any country for that matter. But what I will say is I think um I think it's somewhere in between when it comes to the impact on jobs in others. So what we're seeing in AI definitively is that there are some uh the the it's not about the jobs being lost. It's about how much the job is being changed. Um and at the very foundational level that is a productivity uplift. your ability to be able to, you know, create written content, edit it, then be able to distribute it, uh, is a lot faster than it ever was before. So the problem that we look to solve is ultimately around unstructured data. So the volume of unstructured data is exploding at a rapid pace and part of a lot of part of that is being generated by AI. So I think that's one happening. I think the innovations may my hope and my optimistic view is the volume of innovation will over um will will kind of counter any major detrimental effect and the reason for that way I see it is new industries are being formed and existing industries which have been very nent from a technological standpoint are being completely trans are being completely upleveled as well as a result of AI. If you look at legal tech, legal tech hasn't really been a thing for up until maybe five plus years ago when now you're seeing multi-billion dollar unicorns and entire industries and and an industry which has been has been there for centuries now literally absorbing AI taking it as a way to be able to uplevel and to be able to drive it. Do I think the use of AI and medical research will benefit society? 100%. Does that mean that we will have a healthier workforce? That's my hope, of course. Um so I see that and then of course you have entirely new industries where you know whether it's services or consumer or financial services new you know new products are being created and as a result that will create economic opportunity that's that's the optimistic piece on the on the other aspect then yes there are fundamentally there are a bunch of operational jobs because of the nature of how improved the AI gets every single day with the LM models it's it's breathtaking to see and what that means is that initially will start with 5 or 10% of your job then it will go to 20%. And then it will get to 30 or 40%. And sometimes more and so if you think about the operational jobs which require analysis ultimately that it starts to become automated or threatened in some way shape or form. Um I still believe that there's a a window a huge window of opportunity. Um I think there will be correction. I think I'm I'm just a believer in moderation everything. I I think too much of anything is not good for anybody. Um so I do I think there's a bubble. Do I think there's a lot of um you know companies trying to do things where the promise won't be real? Yeah. I mean that's unfortunately a part of the startup ecosystem. You tend to see that. So I think it will level out. I'm tend to be more optimistic uh simply because having seen the technology I think this is a generational shift uh and very different to where it was in 2008. >> No that makes sense. I think um I actually read a couple of interesting pieces recently that one of them spoke about as a VC that gentleman was saying you know why I'm rooting for a crash um considering you know obviously most people who have money on the line don't necessarily want a crash right now um you know they want >> I don't know who that was but I will tell you they're probably doing it because they have a short position on something so >> but his his thesis was his entire thesis was you know he was talking about how crashes are like you know nature's way of uh seeking efficiency. Uh crashes are nature's way of ensuring uh that they that you know it rewards those who are doing genuinely foundational work in the in the grand scheme of things because you know when there is a bubble more often than not then the ponzies and the and the Frankensteines of of of any industry sort of start to grow right the cancers if you may call it. um if you want to truly find value then you have to be conservative fiscally and otherwise and and and you know crashes tend to do that. Um, so there was that piece. Then there was another piece that was correlating uh, you know, what's happening with AI and robotics. And to be fair, it's it's a range of industries now that are certainly very exciting. Again, I feel like the the last 20 years were very standard. You know, you had these big seven tech players. Anyone who would even pop up would be acquired by them. And it was it was nice financially, but it was boring. Um, and I feel like a young, you know, my young self who was 20 years ago was exciting because every day there was a new technology coming out, right? There we're in that time, lots of new industries. But, you know, the other piece said every time over the last 150 years during multiple sort of industrial revolutions, you know, steam engine and railway and so on and so forth, bubbles have always formed. It's an it's a necessary uh way of how society responds to things. and they sort of talked about a dip and a rise right so um whenever such technologies come in financially and even with jobs there's always a dip because the first uh major layer of workforce whether that was you know people riding on horses driving carriages so on and so forth typewriters for that matter >> are always the first one to get disrupted and that um layer gets automated and there is a lot of value generated Then there's a transition time of reskilling and repositioning that talent into this new world where every person is able to achieve 10x productivity or whatever and that's when you know the society truly begins to uh benefit from that particular technology. Now all of this background in the uh the reason why I gave it was I want to get a sense of you with your VC hat on. You know VCs are strategists. they they generally have a thesis and they're betting on that thesis day in day out, right? Um there is one thing of what the public gets to hear and then there is a lot of other things that the VC that's going on in the VC's mind or what's happening really in closed doors behind in in Silicon Valley. A lot of VCs I would argue right now would be very conservative. They' be like this is a bubble. We don't know where this goes. going out there, setting up a startup, raising that funds. Um, considering that you might have in your thesis the idea you can raise money right now, but that the model of the last 20 years may not be the model of the next 10 years, right? Um, you may have to be a lot more conservative. You may have to be um a lot more nifty and it's not going to be as sexy as it's been the last 10 years. What was going on in your mind? what was that thesis uh with you particularly coming into this space and when you're looking at your startup I'm going to talk about the technicals of it in just a little bit but I want to get a sense of the financial the business side um of of of raising doing this having a projection of where this might go um and actually genuinely being conservative and making sure that you actually um sustain during this turbulent dip. >> Yeah. So I'll say a few things. So look I think fun is yes uh as the VC industry has grown I was a general partner at ridge uh ventures for you know five plus years um one thing that stood out for me is I think what does work in the industry is level of focus uh focus in investing in a particular area particular industry set of type of businesses building a thesis uh I think being narrow uh in in understanding and being having a depth uh of domain intelligence adds resonates uh both with founders who get your business right as well as with LPs who invest that fund who because you built a thesis based on on deep level of research. The second part which every VC will admit to is there's a lot of fundamental luck in this business. It's just it's the nature of it, right? It's the the and but the the thing that really counts is the level of relationships and what you're able to build in terms of credibility and strategy. So your to your points the strategy ultimately is not just about well we want to invest in this area. It's about how that strategy get be get gets put to work. How are you going to be able to connect into the right founders? How are you able to sort of get into the right level of deals to be able to to to get those positions basically. So I think that's that's a large part of how I how it how it how it's done. Now in terms of your question about well you know how do you think about how whether they're conservative or not. I think some of the best investors are the ones who are most patient. I think venture as a general rule is one where you have to have a longerterm horizon. You know, most companies, you know, I was I was lucky to be part of a company called Pure Storage where uh I joined them in 2015 and they went public in 2015. Uh but only after like seven or eight years after formation. Uh so the fact that they were able to do that um I don't think many companies have been able to that is not the norm. like now companies are staying private for a lot longer. Uh but also the bar to going public is definitively a lot higher. Um but you see a lot of the M&As that are typically happening but so as a general rule if you are a venture investor you have got to invest with a longerterm horizon and that's factored in in terms of how a VC fund is typically structured. When you think about being conservative I think that comes down to just being focused more than it is anything else. I think you will get excited about a particular company or a particular industry and you will do whatever you can to be able to make sure you can get an investment into that company uh based on the fact that you know you built a thesis on it and you earn the right. I think most people forget you have to earn the right to invest in a company. It's not something that just comes up and you're like okay let me just invest like you have to a founder like now who myself wants to be able to if I'm if I if I want to give up a certain level of equity and raise a certain capital I have that's limited it's not unlimited not everybody so VCs have to ultimately be able to earn the right to be able to invest um and I think um in some cases yes some VCs are going to shy away because they believe that certain areas are inflated or overvalued or don't see a large return. Sometimes it's just as simple as I don't see a sizable venture return happening in this market. That's okay. Just because it's a good product doesn't mean it's a good venture backable idea that should be invested by a VC. That doesn't mean to say that it's not a good idea or a good product or even a good company. It's just not the best model for a VC uh industry for example. So yeah, and an example is this AI consulting services. Wouldn't you be wonderful to be an AI consultant? He says, but that's not a typical venture backable company because it's services. The margins are different type differentiation is not there. It's difficult to are we seeing now AI powered like tech enabled services. So there's a there's there's you know there's a next it's morphing into something that could be venturebacked whether it's rollups that are basically being done or AI powered services. So the industry is morphing. So I think the conservative is the one that shows up in saying we have a thesis. We're going to stick to this focus and we're not going to we're not going to get distracted from it. And the good news is it's such a big industry now. It is so much more exciting because of the avenues that have been created with AI across all of these industries from restaurant tech to property tech to fintech and then enterprise software and then verticals. I mean it's it's so so I think it's a it's a great opportunity to build a company. It's a great time to build a company. I think the capital is there. I think the imagination is there. The structures is there and we're operating in a global market. And so um from that standpoint I remain optimistic about what can and cannot be done. But at the same time with every market bubble there's always going to be a few things that are just not going to work. And that's just a natural part of it and I don't shy away from it. >> That makes sense. Um what's your core thesis here uh in terms of uh you know what Clario does? What's the problem? And >> why do you think that uh that needs a solution now? Like what got you started? >> Uh I mean I'll just be blunt and honest. The bottom line is the world is just being the volume of garbage data that's being created uh in across enterprises and companies is absolutely ridiculous. Uh I've you know to me I find it offensive. uh because I can't believe how much is being wasted. Um so to give you some context, I've been trying to solve this problem for 15 years, Mama, right? Like I was a CIO of several companies uh and all I would try to do is to try and say answer some basic questions. Can I clean my file system? Right? For example, you probably don't clean the photos on your phone, right? You're visual first individual. You you're in the that's fine. But with the greatest of respect, there's a bunch of stuff there probably should be deleted. probably doesn't volume need to be back to be backed up cost is growing. I have three terabytes of content that I need. I I do need it, >> right? I I know you don't need it, dude. I've been paying for a subscription on another cloud file system for 15 years. I don't even know what the hell's in there. Okay, but it's ridiculous for enterprises. And the reason I basically thought about this is I have been trying to solve this problem 15 years initially way before AI. Just clean infrastructure. I wasn't This is not rocket science. This is like, can you find the files that are over 10 years old that nobody has touched? I mean, it's not that hard, right? Guess what? I couldn't I couldn't get an answer to that question, right? Then I said, could you just show me the files that have been duplicated 15 times over from seven years ago? >> Can't answer that question. Think about how much stuff it just adds up. The file name version_26_mazam_final do not change undersc_final final. What's going on, man? What what is this right? It's ridiculous. Like get get a hold of this situation. So So then this was the genesis of the problem. I mean I was joking. I was going to call this company garbage.ai, but none of the investors wanted to invest in a company called garbage.ai. I then tried to call it catra.ai and none of the desi VCs wanted to invest in a company called catra.ai. So they call it Clario which is clarity of infrastructure and operations. Clarium. So the genesis of the problem is can we clean a file system and a content system. The reason you want to clean a content system is because you want to build a conversational AI project for example for customer service or automating an engineering project based on confluence articles. Right? But a garbage in is garbage out. So, for example, if somebody talks to an agent or a bot and says, "I would like to set up email on my phone and the article shows up, please set it up on your BlackBerry phone because the article hasn't been updated in the knowledge base from 2008." >> Like, what's going what you know? So, now you've spent token costs, >> right? Now, you've now you've now you've got people who use that bot or agent who don't trust it, so they won't use it again. And now you've got this problem just just create becomes a big >> problem and they're going to blame the AI because they're going to be like, "Oh, it's hallucinating. It's terrible." But the fact >> they're going to blame me as a CIO, dude. That's what that's going to happen. They're going to call me up and they're say, "What the hell's going on here? This company, you you told us you're going to build an AI project and roll it out across the company and it fell apart." So that's what's going to happen, right? And in customer service projects, it's going to be speak to an agent. The point is very simple. Unstructured data is 80% of a company's file system first and foremost. Okay? Or a content system. Number two, they are generating more unstructured data than ever before. So if you and I wanted to create a document, pre-Chat GBT, we would probably wait for about a week to basically do it. In fact, we would avoid having to write a document because it would be so hard and cumbersome. File, new, here it is, type it up, go back, have a cup of coffee, come back, etc. Now you go to Chad GPT, you go to Claude and you say, "Give me a 20page document about the wonderful podcast that Mozamal has created. Put some photos in there and tell me it will create it in 5 minutes." Now imagine that millions of those types of content being created in companies being uploaded into Google drives and Microsoft SharePoint articles and Confluence and Microsoft one drives and box folders and then the impact of that in the longer term. Number one, when you want to search for content in a company, what are you going to see? You're going to see garbage. Yes or no? Second, you're going to now say, I would like to do an AI project internally. So let's take a simple case, right? If you were CEO of a company with Zill and you wanted to automate new hire on boarding because you don't want to hire any people ops people or HR people to start with. You say if you're a new employee of my company, talk to this bot that's going to help you with all the questions. If your documents are from 2020, it's going to bring up COVID vaccination requirements, isn't it? Because you didn't clean it up. It's if you are going to automate a legal pro uh uh process. If you put in 20 versions of a document like an NDA, it's going to bring up conditions and it's going to bring up, you know, wrong answers for stuff and you're never going to use it again. What have you done in the process? You've now had another failed AI project. You now spent thousands of dollars on token costs. And the reason that token cost becomes a problem is because when you put garbage into trying to do an AI project, retrieval becomes noisy. you're able to process. Think about you cleaning a house, right? If you knew that two rooms in your house were dirty, but you still tried to sort of clean them. A waste of time, waste of money, waste of resources, waste of energy. That's what's happening with garbage data that's sort of coming into place. So, our goal is very simple. We're the Marie Condo of enterprise data. We want to spark joy in file systems and content systems all over the world. We integrate with one drive, Google drive, confluence, box, sharepoint. We understand these file and content systems inside out. We scan for the metadata in the file system. We're able to understand how you can clean it. We'll be able to batch a whole bunch of recommendations in terms of actions that you want to take and then be able to sort of clean that up and you can have a much cleaner infrastructure and a better AI outcome. That's the way we think. That's what we're basically doing. So that's that's what Clario does and that's the genesis. Does that make sense? >> That makes a lot of sense. Um I'm curious on on on one end I understand the whole um you know there's a lot of garbage there's a lot of uh redundancy there's a lot of obsolete data and you want to clean that up archive it if you must delete it if you must um but then there is the whole concept of I mean you're dealing with unstructured data now >> I'm curious whether >> because right now your particularly the idea that you can create so much data um from these LLMs which everybody is doing personally I'm doing as well you know I come up with a thought I made 20 page documents I come back it's it's probably three pages down so I'm like okay let's do it again so [laughter] now I made 40 documents >> yeah totally yeah why not >> and a lot of this content is now being created a lot of you know on enterprise it's going to be at scale um but then there's a lot of useful content being created as well that's also unstructured and random and all the all over the place Right. >> Sure. Technically speaking, once you're actually in and you are in the infrastructure and in the position of dealing with a lot of this unstructured data, I'm curious if the eventuality also includes you being able to provide consistency to this unstructured data because a lot of let's say in my understanding the next generation of AI is also going to require you being able to um utilize your unstructured data and generate insights from that and to potentially make your AI better. Um, you know, Palanteer came out with this thing yesterday about sovereign AI and how you shouldn't be putting your data out to uh these frontier models and build your own sort of XYZ. Um, so I'm curious whether this is just about removing redundancies or whether on some level this also optimizes your ex entire unstructured data lake if you must call it and position it for better utilization and insight generation for the future as well. >> Right, great question. Let me just be more foundational about this. You need to clean your system. >> You need to clean the system first. It's as simple and straightforward as that. Okay. So, let me give you some examples. We have discovered terabytes of Access database files being shared on one drive. Can you open up access database anymore? No, it's not. It's an unsupported file format. Okay. There's literally volumes of this not just being stored, it's being backed up and archived. So, unsupported file formats shouldn't show up in search results, shouldn't be backed up, shouldn't be archived. Now let's move to content. Does it make sense for your automation of customer service agents or an internal engineering automation to have to to to be created with knowledge articles that belong to a product you no longer sell or you no longer support? >> Does that does that make any sense at all? No. So why do those articles still exist for what? For some historical case. If it is then move it out of the system. That's all we're trying to do. So the first thing that we're trying to do is to clean your system. When you clean a system, you help organize it. So where we come in from an AI perspective is the ability for us to then proactively keep it organized and keep it clean. For example, dduplication. Now 17 versions of a document from 17 years ago have no place in your file or your content system. Okay? a knowledgebased article that has been sitting or a set of articles that have been sitting around that nobody has accessed by the way they've literally been sitting around for ages have no place in your system. Okay. So we are focusing on the foundational piece. We're six-month old company. We raised a $6 million seed round last year led by Preface Ventures, participation from Bridge Ventures, Moment, High Sage, Rain Capital, and Foster VC and a number and transform VC. Our goal was very simple. We said CIOS need to have trust in their data. They need to go to their executive partners, their CMO, the chief legal officer, the CRO, and say you can run much more automation. You can drive more AI projects than ever before because your data is in a much better set than it is before. The the harsh reality is people have not cleaned up data for a long time because it's very hard. In the introduction that you made was very apt. It's very boring. I'm very boring dude. Like trust me, it's totally fine by me. But the reality is it's a huge market and it's a huge problem. And so all we think we need to do is to start this change management this behavior. I need to it's a news flashash for some of your uh some of your listeners. Let me put this the cloud is actually built on hardware. There's literally storage and data centers everywhere and this needs to be backed up and stored. So, Mama, when you were talking about earlier about your like Apple cloud, that wasn't there several years ago. The reason that was there now is because companies like Apple and Google are managing hundreds of exabytes of our photos and our videos in a visual first generation and they need to keep it, retain it, back it up and archive it. And I can tell you for a fact that data centers aren't being fast enough. I've been hearing that we're thinking about building data centers in space. Someone needs to pay for that cost. And so my viewpoint is is that if you are running an AI project as an enterprise, you need to be able to be conscious of token costs. Token costs can be reduced by reducing the level of garbage that goes in. You should have a cleaner infrastructure by making sure that the stuff that is not there and not necessary should no longer be there. There's no there's no excuse for it. The people who say I may need it one day, I'm totally fine with those people. But unfortunately, it's been 10 years. So I'm not sure what one day you're waiting for. Is that the day you retire and say I may need it that day? I don't know. So the point I'm basically making is like if order for AI to be successful in a number of there are a number of things that need to be done right. Data is one of those. I think it's a huge opportunity to be able to have a cleaner unstructured data set that could be maintained clean which improves the level of trust for executive leaders to say we would like to drive more process. We would like to make sure that this can be driven by AI and we could train it. We could make sure that it's inferred and processing the stuff that we trust. That's the way to think about it. Does that make sense? >> H that makes a lot of sense. Um I'm curious though about the economics of of your business right now. Um and what I mean by that is there are there are multiple ways to sell a product particularly in today's day and age. I would argue that a lot of consumer and I know you're you're focused on enterprise. I'll come to that just just a moment. But a lot of consumer products right now have a lot of valuation and a lot of revenue largely because there's a FOMO in the market and a lot of people are there right now because they're they're just worried that if they're not they're going to lose out on something. And I think a lot of these companies are going to get hit by the bubble bursting because once the FOMO passes and I would argue normally when the bubble bursts, it's not just FOMO going back to normal, it actually goes the other way where the same folks who are omoing become extremely harsh and say, you know, this is all terrible. You know, we're never going to come back to this again. And and that's the tough part where if you un unless you have some economics baked into the system, you're not going to be able to survive. Um >> and so what I mean uh in terms of enterprise is whatever you're selling, whatever the cost of that selling is, is that amount less than the efficiency gain or the cost saved? And if that is less, then it's baked into the economics because it doesn't matter about what the narrative is. It doesn't matter about the bubble. All that matters is today or post crash, I go in and I say, "You're already spending $200. Let me take $100 and save you $100 every month." It's a no-brainer, right? run me through the economics for the customers here in terms of what you cost um you know integrating you into any enterprise solution. >> Okay. So first and foremost is again we operate in in three particular areas. We can have some um hard ROI in in in a couple of them and you know much more sort of project based that you'll basically you'll see um in a much more longer term. So let's start on the hard stuff. it you know if you think about token costs as a general rule right just let's just do some let's just do some basics of it if you have to process hund you know 100 let's just say 100,000 documents and articles >> that ultimately the token cost for 100,000 articles of processing is vastly different than processing 70,000 articles or 30,000 right so you have to think so our improvement is improving rag number one and reducing the rev of compute and processing and token cost that you would typically have to do, right? That's so that is just just very clear from our standpoint and what we're basically seeing from folks. So that's one. The second is much more downstream, but the outcome based, right? So think of it from an outcome base. If you feed it incorrect, like the here's the thing that we need to sort of get people to to wake up to much more, right? bad data resulted in sort of bad outputs. But with AI, what's happened is it basically gives you you can [clears throat] literally come up with really bad outputs more of the time simply because of your ability to direct it in the way you can. I could go into Chad GPT and ex and ask it to give me the linkage of how you miss have caused the financial crisis of 2008 and it will give me an answer right it won't say by the way it doesn't out of context it will actually give me that answer right so my point is is that that is definitively part of troubling and fixing that is a data problem it's not all of it but there is a data problem that feeds into that so the other thing is LLMs are very good for everything that's outside of your firewall. It's out in the public domain, but it doesn't understand the nature of what we are building internally. Now, some of it's available on Reddit and our support blog, etc. But a company's taxonomy and its language, it's native to its company. That's how it's understood, right? So, you know, you have to get accurate and you have to understand that the experience of getting an AI project done well, it could it's unforgiving if it go if it goes badly. Now the third piece is a more simple piece. So you mentioned this earlier on, right? Your Apple storage. Okay. So right now if you have to upgrade or pay more every month you would say okay that's annoying. >> But if you only had to do it once a year that's not the worst thing in the world is it? So storage expansion of course just from a basic infrastructure perspective the ROI for us is very simple right? If you if you think of it from a big picture standpoint, think about the objections that people and communities are raising about data centers being built in those areas. Right? So let's just just just just look at this on a foundational level. Data exploding, more infrastructure is required to store that data. And you can argue the storage technology and the infrastructure technology is getting better. You can compress it. But dude, compressing 15 million articles is still compressing 15 million articles. Okay? you know, whether it's here or in space, it's still 15 mill 15 articles of crap. That's the bottom line. Okay. So, not we're already seeing that in RAM as well in terms of the prices where the bottleneck is actually the molecules, right? It's not Yeah. >> So, quadrupling of nan storage on like >> it. So, keep telling me storage is cheap. Storage may be cheap from a commodity standpoint. It's not cheap to manage and and secure and comply and archive and process, right? So, so just from that samp from an infrastructure standpoint when you have scarcity in terms of infrastructure data centers are not being fast enough number one storage nan chips quadrupling in price and then in your data center if you were you know if you're a CIO onrem healthcare financial services a lot of the world still runs on prem for as as most people most people have to realize you have to now reserve space in your infrastructure for Nvidia GPUs which used to be the case when you didn't have AI project you just put in more storage >> so now if you were constrained for resources time and space or otherwise what are you going to do I mean can you just simply say we're going to just add more storage good luck I don't think it's a good practice so I think somewhere down the line we have to come to a conclusion which says three nearly four years post chat G TBT we are not seeing the level of ROI that we should be expecting from this transformational technology I'm not saying there isn't any I'm saying there isn't enough there should be and and from a CIO perspective. So that's the genesis of how I thought about it from a an enterprise use case having been a CIO of multiple companies if you wanted to roll out successfully AI internally. You would run into that problem. You would not have an engineering problem. You would not have an LLM problem because that's been provided to you. You would not have an imagination or process or workflow. You could all design this. You can design all the products you want, but it has to be fed on your data. And if your data is not clean, I'm sorry, it just will not succeed in the longer term. I just know that I've seen this too many times. >> That makes sense. I'm going to divide this now into two parts, right? So there is a world of startups and tech enabled, tech first, you know, enterprises. Then there is a world of corporate America. Um, >> and this is something I experienced last year when I originally came to the US. I actually took a trip around and about. I went to New York. I went to San Francisco, Seattle, Dallas, um really some of these major hubs trying to see the narrative and the and the vibe and the feeling. I'll be honest, I feel like San Francisco right now with where you're based is living in a different dimension completely disconnected from reality. It's a different world. >> And it's a different world. >> And and I feel like and when I was in San Francisco particularly, I think a lot of people I met while I could see the future there 100%. You know, no doubt about it. But I also felt like they're feeding each other so much of that Kool-Aid of what's happening in Silicon Valley and where everybody is so knee deep into this whole revolution um that they're not realizing honestly if I if I'm in Atlanta, even in New York, corporate America is just still living their best life, not really doing much about any of this. You know, >> I I I I think it's a I I want to stop you and correct for a second. I think it's important. It it you're absolutely right, but this itself is a bubble. Okay. I I travel to Minneapolis, Minnesota, and Chicago, Illinois, New York, Boston, Charlotte, North Carolina, you know, Dallas, Texas, and otherwise. So, it is the age of transformation. It is the age of AI but you know entire industries some are much further ahead and I see this across the US but yes the capital structure is largely based here >> you know a large part as a result of that has morphed into it right so that's think about it >> so my my thesis then became as I went across >> um I think San Francisco is running faster than the rest of the lot I think the a lot of the rest of the lot are running slower than they usually run because there there's also that um how would you put it? Um that inertia, right? So, um your initial response when you're so disconnected from it is that to dis to dismiss it and to say, you know, this is not this is not happening. Close your eyes, put your head on the ground. And I'm I was hearing a lot of that uh you know from people in corporate America where and and I could tell the reason why that was happening was underlying there was this friction there was this tension where they were reading a lot about joblessness and they actually felt like you know you you let's say you meet a a startup kid in uh San Francisco he's generally working 16 hours and he's genuinely working left right and center and then you meet someone in corporate America let's be real beyond the meetings that they're doing it's actual productive value of like 3 hours and then there's a ton of busy work, right? >> That's what AI hits immediately. And so obviously there is this defensive anti- AI sentiment um in corporate America. Now in that context, what I wanted to sort of ask you was I get the startups. I get that and they're also fairly un unstructured. They're tech first. They do a bunch of these things. But there's a trillions of dollars of value in corporate America in making corporate America itself efficient as an engine as this sort of extreme behemoth. A lot of this is also again coming from the background that I was from you know the the the snowflakes of the world the terodatas of the world it was largely structured cleaned processed data on top of which was these entire enterprises were built. Now a lot of these enterprises are fairly new to unstructured data where they're again building out new tentacles and just rapidly gathering a bunch of things and thinking okay we'll decide what we do with this later because they're not they're not built in that sort of startupy uh chaotic model. I'm curious when you differentiate between let's say you know the the the startups of Silicon Valley versus these high value comparatively slower um more defensive enterprise companies. How different is that in terms of the value that you're offering and in terms of the uptake that you might imagine them to have over the next 5 years? For us, we don't see a tremendous amount of difference to be honest with you because ultimately we're starting at the foundational level. So I think that's one piece as a general rule for AI startups overall. Um what I would say I I I think it's still too early to to have anything definitive mama if I'm honest. I I I there's no patterns in this business anymore. If I'm you know I I see this I used to think well okay you know this is going this direction so it will then go to this level you know all of a sudden some of the biggest tech companies are having layoffs why are they having layoffs it's become trillion dollar companies I don't understand you know uh like literally out of blue you see this you know companies who said we'll never have layoffs all of a sudden have layoffs uh the level of mergers and acquisitions happening so unfortunately the reason I can't answer the question directly is just the the patterns are just not there. I just haven't seen it. I I see all flavors. I've seen company today got acquired after only 18 months. Uh and I saw that company and founder wish him well and success. I've seen other companies who've raised several hundred million dollars revenue, you know, at risk. I I I don't know what to say. So the impact of AI in some of these industries, some of it's been inflated, some of it's very real, some of it has, and some of it comes down to the fact that you've got two very very large companies who are dominating a consequential technology like LM >> in anthropic and OpenAI and their ability and to distribute and innovate on a new product offering which could completely disseminate and just completely have existential risk on a on on a company that's been around for several years. I mean, that's real. That is real. Uh, and that's, uh, and so that's concern. So, that's what's typically happening. Um, so it's very difficult to say, well, is the ROI is it different? It's because I just think it's still too early to discover that and I think it's moving too fast. So, that's my that's the way I would look at it. >> One last question and then we're going to sort of wrap it up as well. Again, I mentioned earlier the the Palenteer CEO talking about sovereign AI for particularly for enterprise. Um, uh, Sat Satya from Microsoft came out with a very interesting essay recently where he said, you know, we can't have a world where frontier model companies basically absorb everybody. We need a world where we have shared prosperity of sorts. Very Chinese of him. Um but inherently what he was saying was, you know, you stay in your lane and enable everybody else to to to be able to generate that value and distribute it accordingly, right? Um what's your view on the way that you just re mentioned earlier about these two companies that are beginning to absorb everything seemingly? >> Yeah. Um and so there is there's always that question for every startup founder even is what I'm doing even worth it because if I can >> there is someone in anthropic who can who has infinite tokens has loops engineering loops engineering agents engineering products >> welcome to the welcome to the startup world so I'll answer I'll answer very quick I will be respectful of of your time as well so I think two things one is um I think there's going to be more data regulation and there's going to be more data enforcement at the sovereign level than ever before. I I think larger companies can pay a billion dollar fine without thinking twice. It's a rounding error. I think that being affected to companies, the large majority of companies uh is going to be a real real problem. And I think because data is now not just an asset for companies, it is a national asset and it will be protected uh by by sovereign nations. And I totally understand. Sovereign clouds, neoclouds, I see that happening and morphing more and more. On to the other more fundamental question. Welcome to the day in the life of startup founder. Some days the highs are high. You think you can change the world. The lows are like what was I thinking? What the hell was I why why did I do do this? Basically, um my advice to founders, go understand that it is a big market and go after a big market. Number two, frustration. Motivation is a very good catalyst if you want to build a company. For me, this was the frustration. I've s frustrated this problem for 15 freaking years. I need to solve this. Okay, it's a nice way to basically go after it. Um, can Anthropic build a good product? I wish them nothing but success, but that is not their focus. So, this has been around for a long time. I've seen companies that tried to build a competing solution uh with companies that were hosting on their actual platforms. AWS have been notorious for this. Microsoft like Microsoft less so, but you know AWS had this where you know you have companies that are running on AWS and AWS have a competing product. Some focus on it, some don't. That is just the nature of of the industry. So I don't I don't I think there's still a a huge room for opportunity, but it is a foot race. You've got to execute, you've got to focus, you've got to drive further forward. And I think that's what differentiates a business irrespective of the idea. No question. Second, build a great customer experience and they'll stay with you. No question in my mind. So ultimately, serving customers, demonstrating value, which is what we like to try and do and be very clear about that is is something that I think that has huge value and I think that's what will drive companies to be successful in the longer term. M can you give me a a sense of where you think all of this is going to go in the next 12 months? And then uh connected to that, I'd love to get a sense of who in your opinion are the three most exciting uh startups pro probably not the not the more popular ones uh but but really doing interesting work um that one should be looking out at. >> Reason I can't answer that question and I say this with respect is I'm just 110% focused in what I'm basically building. Uh, and so I wish a lot of my startup founder companies lots of success. If I select three, I'm going to get offended by another 30. So I just want to be respectful of my community that I'm in. And I I do have have a large founder community. So I'll be respectful of that. But I I do think that we're going to see interesting. We're going to see three potentially trillion dollar IPOs this year. I mean, I think that's a very exciting time for the industry. I would like I was interesting to see how the market absorbs that. Um I think the level of innovation is going to be very I think merges and acquisitions are going to definitely increasing more than ever before. Um so and I think the SAS apocalypse thing I think was completely overhyped and I don't think that was really a thing that distracted me. I think it was a distraction. So we'll probably see more distractions similar to that. But the harsh reality is and I'll just tell you right now there's lots of companies that still run on IBM mainframes. There's lots of insurance companies that still have lots of fax machines. There is lots of you know that it takes a long time to dramatically change. Now those markets are not growing. Those products are not growing. I totally understand that but change takes time and just it's important to remember that >> makes a lot of sense. You thank you so much for taking the time out and sharing all that insight. I'm I'm excited for the product that you're building and I'm uh curious how that sort of is able to reduce the amount of slop that everybody seems to hate. uh courtesy of you know making sure that the foundation >> you've just done the you've just done the BDR pitch reduce the number volume of slop so we'll we'll we'll give you some commission on that one alone thanks very much really great to have to be here >> likewise and thank you so much for all of you guys let me know in the comment section below when you're building AI systems how do you deal with unstructured data do you think that is a problem if you're part of an enterprise how much of useless um redundant data do you guys have to deal with and uh you know do you think that can make a huge impact in the way that we deal with AI in enterprise but nonetheless this was zedi you are watching beyond the bubble podcast powered by noodle seed studios thank you so much for watching and I'll see you in the next