The Infra Pod - The AI Analyst is coming to change Data Teams (Chat with Lucas from Gravity)

Episode Date: October 6, 2025

In this episode of the Infra Pod, hosts Tim (Essence VC) and Ian (CEO of Keycard) sat down with Lucas Thelosen, founder of Gravity and former head of product for data and AI at Google. Lucas shares his journey from leading teams at Google and Looker to launching Gravity, a company focused on bridging the gap between business users and data through generative AI.The conversation dives deep into the challenges of data analysis in modern organizations, the evolution of AI-powered tools like Orion, and how generative AI is transforming the way companies leverage their data. Lucas discusses the importance of semantic layers, onboarding AI agents, and the future of data teams in a world where AI can automate complex analysis and reporting.

Welcome to the InfraPod. This is Tim from Essence and Ian, let's go. Hey, this is Ian Livingston, CEO and co-founder of KeyCard, the greatest way to create trusted AI. I'm so excited today. We're joined by Lucas T. Lozen from Gravity. Lucas, what is gravity? Why did you start it?

And what puts you on this crazy adventure? Yeah, so I quit my job at Google. Like, I had my dream job, in essence, at Google, right? Like, a head of product for data and AI. Like, it's like the dream. That's what I always wanted to happen. And with everything happening right now in AI, I just couldn't sit still. I had to quit.

And, you know, I got four daughters here. That was a major, under, you know, major decision. But I got to do this because there's so much opportunity here of what we can do with AI now that we actually have the generative component, right? We had data. We had AI for so long. And now we actually have the generative component, which allows us to explain what's happening in your data to the business

user, right? And actually to take that last mile and actually get it to the person that makes decision and takes actions. And I ran the consulting team at Google before this. It's just, there's so much untapped potential out there. So what is gravity? Like, what's the problem you're solving? You're talking about like, hey, generative AI, now we can solve this problem. But like, what's the foundational problem, you know, that we've always had that you're solving? And why, how does generative AI, like, make that this is a moment to go do something like from the ground up net new again? Yeah. It's a nice. just generative AI, right? We're building on the shoulders of giants here, right? There's data,

of course, there's a foundation. There's, like, traditional AI with, like, forecasting and everything, normally detection. And then there's now this generative component to it, which allows us to talk with everything. Now, the problem that I had seen, you know, I worked with over, like, hundreds of Fortune 100 companies. I can't say Fortune 100 companies that wouldn't work mathematically. Everybody, even those well-funded companies, suffer. from bandwidth. There's just not enough time to dig through all your data and find these things. Every time we got in there, we got paid a lot of money to come in to these top companies and we found things that would save them millions of dollars. And then you take it to

a smaller scale, mid-sized companies, small companies, you can find a ton of value if you had the time to actually dig for it and then also connected to the business because that is, I think, so many times misunderstood problem or not seen problem where you have the business, you know, you have sales, marketing, product, all out there doing things. And they actually don't quite realize all the data that they have access to. And the people that know the data, right, the traditional DBA or something like that, they don't really are in the meetings with the business to understand what new problems they might be thinking about right now.

And they always ask, like, hey, can I join the meetings? But they just don't have the time. Like, nobody invites them to the meeting. And so it creates all kinds of gap in the business, of the technical and the business side. And now with the tooling we have, right, we can actually bring. that, like where Orion, our AI agent, can actually be aware of all the data there, right? It can even take in your meeting notes and understand what the meetings were about. And so fully context-aware, generate reports, generate insights for you that are tailored to you,

to your recent meetings, to what you just did and what you care about. Incredible. So you're seeing this is like a moment to redefine how we basically build and use data inside organizations. And I mean, you would have a lot of, like, inside of this knowing that you spent a lot of time building the Looker professional service organization. Was there a moment at Looker when you saw a generative AI, like, what was the moment where you were like, I just got to go make this big bat.

Like I just kind of go do this from the ground up. Was there some unlock, some moment as you look at the tech playing around with things that did it or are you just reading papers? What was it? Yeah. So long before generative AI, so let's go back to 2015 or so. I had this moment where I helped this major company with a marketing spend optimization, you know, and we saved them $600,000 a week.

It was insane. And then we went to the next organization, right? And it turns out the same pattern applied. Very different company, very different business, but the same pattern applied. And so I had this thesis at Loka that we should create templates of like, this is how you should do analysis for marketing spend optimization, cloud spend optimization, any kind of optimization scenario, really, root cost analysis for cohort creation, all that. And we built these templates and it turns out they only fit like 70%.

maybe 75%. And so you always needed to bring in the consulting team and they need to customize it for you. And it's really hard to understand someone else's template and applied to your data. And so people are like, well, you're just creating a scheme to charge more consulting hours here. It's like, no, it's just like we need to,

you know, there's this like fuzzy edge. And then when, you know, I was at Google at the time, ChatsyPT was the non-spot. I mean, Google had the large language models for a long time and it would have been so cool. But anyways, ChachyPT came out. This was the moment for everybody. And there was the tool I needed for the fuzzy edges, you know, for the remaining 30 percent.

So we can use the template and then customize it for that unique snowflake that is that given company for the uniqueness of that organization without having to bring in an expensive consulting team. And so I think we'll be so interested to hear maybe how you think about this problem because you've been at Looker and then you went to Google, you're definitely seeing a lot of like vertical solutions because I saw your talks around Google. next and a lot of the content is really about modernizing BI, like how do you deal with supply chain data, how do you deal with different things? So how I look at data teams or data solutions is usually for companies that has dedicated data teams that knows how to empower themselves, you have to go buy a BI solution like Looker or something similar that can help be able to like organize and come up with their own scripting. You have to look, you know, look at mail and kind of things like that to like build the things you want. And of course, you also rely

and other partners to help you, okay, I'm manufacturing, I'm doing supply chain, I'm doing something else, to help me figure out how to build maybe even vertical, faster reference solutions, right? And now you're coming with an AI angle. I'm very curious, like, for you, do you think that for AI, are you just trying to, like, I don't care what vertical you in, I don't care where you are. If you're just doing marketing, my AI will come to help you, and there's no customization or anything required. Like, how you think about the world now with AI-driven data analysts?

The world basically looks the same now. We're just put AI and then you're marketing with AI, everything looks good. Actually, there's still a lot of thought process to it, to even know, okay, you're a supply chain, your manufacturing. Can you tell us maybe like how you think about the problem, how does Orion where you're doing kind of fits that?

Yeah, so there is, is a lot of similarities across organization, right? It doesn't, but then there's also uniqueness to it. So I almost think of the world in a chess board there, right? There's like the different industries and then there's like sales marketing, product, so customer support, customer support or like account management is quite similar, right? If you are in pharmaceuticals or if you're in software sales, right, or if you're in construction, right, you meet with customers, you have a funnel and all these

things. But there's uniqueness to it, right? There's like, okay, different speed, different cadences of how you do things. When we started this whole thing, you know, we sat down and we mapped out like in that chessboard, in that chessboard of like industries and roles within those, what are the questions people should be asking? You know, and like, it's great to have access to industry experts that also can think about this. The AIs themselves at this point have also quite a good understanding of, you know, what a supply chain analyst with demand forecasting might want to do. So all that we created the questions you should be asking. And that's really foundational for Ryan now. Might be giving some IP away here. But, you know, it's really hard actually

to create that list. So I feel confident saying this. But then we also do quick onboarding. So when we onboard a customer, we pull in as much as we can about this organization. And if they have like any kind of onboarding materials they use for their own employees, we actually give that to Orion to read. So it understands how the organization talks about acronyms, how they think about, you know, what dates are important to them, what is their fiscal calendar, all these different things. Now the beauty, you know, when you're onboard an employee, it takes a couple weeks to get that. It takes Orion like five minutes to catalog all that and apply vectors to it and store it. Then it understands your business quite well.

There's still some nuances to it because not everything is documented properly. It's actually really interesting. One of our customers now, they hired the next analyst. And they're like, hey, could Orion onboard our next analyst because now it actually understands our business and our data super well? It would be great if it could explain it to the new hire. And so I'm looking at the very first launch video you guys have, it basically has a looker in the center first, right? Like I can connect your looker and I can basically hunt,

uses the word hunt for insights there. I don't know how evolve you are based on the very first launch video, but is that the assumption? Maybe talk about how Orion works in general. Like how does the customer typically start to leverage you? Do you have to have liquor first and then you connect you? and I had to say what role I am, like what are the things you require to get the context and the data

to even do your job, the first place? So like a little background on Looker, I was running data and infrastructure for a company. AI was part of my purview as well. This was 2013. So like one of the earlier AI hype cycles. And I needed a central place to run everything from. So there's one single source of truth,

like a ground layer of what that we can use and we can all agree on. And that's why I came across Lookup. because Looker has this base foundation in there, LookaML. And so it's a very good tool to use for all kinds of AI use cases. And so as we started Orion, starting on Looker makes a lot of sense because there's a structure to it that we can use for our quality assurance component to always check back with like, this is correct. Now there are other tools out there.

So there's like a variety of different data catalogs and semantic modeling layers that Orion can actually attach to that the list is quite long. So we started with Looker, of course. That's where a lot of my friends are. and a lot of former customers. And now we have expanded to the wider universe of semantic layers of things that explain what your data is. Because if you let AI lose on 6,000 tables in your database, it will find all kinds of things that are not necessarily appropriate or potentially incorrect. So we need something that grounds our quality assurance agent into some source of truth.

Incredible. So I'm kind of curious, like as a result, you're building the software. You're obviously thinking deeply about what the software does in terms of how we, we gather, consume, ideate about the data that we have about organization and think about how that data informs what we're doing. How do you think something like what you're working on changes the way that we actually build our data teams and actually build our companies, right? Like, what do you think the delta is in the future, like between, you know, let's say

pre-generative AI, very static, lots of building out very static views or, you know, that 30% that you couldn't implement to a world where. we have highly rich, highly dynamic actionability and corroborability because of the inference thing that we can do on top of the data set that we have. Like, what's the delta and the team look like? And then I'd also love to understand, like, how do you think that changes the way we run our businesses and the types of business we can build? Really good question.

So I think the work of running hundreds of queries, right, running multiple permutations of something, those are gone, right? You don't have to do that anymore. That's a positive, I would say. And, you know, AI doesn't get annoyed. it runs the 36 permutation if it needs to, right, to really get to something where you want to go home, right, you want to be done, where that then gives you time for, and I think this is where it really gets crucial, is to think about data and analytics and insights from data

as a product. And you become not, you're not head of data, you're head of product management for data, right? Because in most companies, data is really the second product. Let's take the example of Uber, right? First product, ride sharing. Second product, okay, maybe not Uber eats, but you know, second product data, right? And so you need someone who thinks about how data is evolving and how it gets into all these teams, how it's being used, which additional data may need to be captured, which, like, data sets might, you might want to buy it to enrich your data further, but really owning data as a product and be strategic about it. Because all of the grant work, right, I mean, there's still the work to be done to organize the data and make it very

clean and structured, which there's other teams working on that and AI tools working on that problem, so I'm excited for that. But then the work that we are doing of going through it, right, permutations, cohorts, all these things, you don't have to do that anymore. So you can really focus on the strategy side of things. Until your second question, how does this change the way we run businesses? So this is why I'm getting really passionate because I was at Google, right? I was at Amazon. I was at Walmart. I was at Uber. Like, you have these highly resourced organizations that can afford, you know, all these MIT, Stanford, McKinsey guys. And yeah, they grow much faster than the market.

But now you can actually bring it like a small company over here in the Midwest or something can afford similar technologies at similar level of sophistication and you're really leveling the playing field. Like I don't think the vast majority of organizations realize this yet. But, you know, this is not just true for data analysis. It's true for lawyers, right? Google had thousands of lawyers. So they had this amazing power in the market

where like a small company had no chance. But now they can afford an AI lawyer that's actually really, really good. I'm super excited about that. Yeah, it's really fascinating because there's so many things I think AI can do that we are all like believe it can do, but actually in reality it's really hard

to actually make it do everything at the moment. I want to maybe dive in a little deeper into how you just mentioned like semantics, layer is really important, right? So we want to talk maybe some of the things Orion is relying on. So when you come to a customer that has all different kinds of tables and data, right, because nobody has a standardized schema across every industry and every company and even the same industry. It's all internal. So when you come to a customer like this, it sounds like there has to be a definition of what this data is, right? And how do you standardize into something

to help your AI agents be able to understand the data much faster and quicker and doesn't have the go everywhere. And given your looker experience, I'm very curious, like, how you see that exercise to be done well, because I feel like there's either, I'm going to do everything for you, I assume you don't even know what a semantic layer is, and then you probably even know my semantic layer, because I'm a product coming in to trying to understand your data, or, you know, like, here's my semantic layer, you go and figure out how to do it completely on your own, which also I feel like is very difficult, right, for a variety of reasons. And so how do you really get your customers to be able to map what they have, which sometimes

they don't even know what they have, into something like your product, because I feel like that's the usual number one problem is like, I have data. I don't even know what they are, and I don't sometimes even trust it. And you have this magical AI product. Like, we can do everything for you. But like that dance, is that a challenge and how do you sort of overcome it? Yeah, no, it absolutely is a challenge. Like any company right now that has a structured data layer has a huge advantage because it's not

just important for Orion. It's great if it has a semantic layer and it works really well, but any AI tool you want to use, the more structure you have, the better prepared you are for this moment right now in time. And if anyone wants like, what should I do? I don't know what AI tool I want to use. Well, maybe start there. That's a good, good first step. Although, little word of caution there, there are a tremendous amount of companies from Snowflake, data breaks, Google, right, the big guys here, all focused on solving that for you. They're all to go through their own metadata, right? They have the query log.

They have the information in what all the data was suggested and which API it came from sometimes even. And so they can actually enrich that data for you pretty quickly using their own AI technologies. So if you are on the fence, we were just talking with a company who hasn't done any of this yet, it's all raw data, none of it has been structured. It's like, well, try out, you know,

I know some of them are not in the like life, you know, yet to the market accessible versions, but there are already pretty good tools out there. We even connected Orion to a couple databases without any semantic layer and just to read what it can find. And it's doing pretty well, but it does need someone who says, you know, this is the date we use when we define a customer is now a customer, for example. You just have to have someone who is the judge on that, right?

We don't want Orion to make that call for you. But yes, the tooling is getting vastly better and fast really quickly. So I'm excited for that future. we're building for that future, for sure. Orion is not focused on solving that for you. But yeah, any big database provider is trying to solve that right now. And so, yeah, you alluded to the idea that, hey, we can also even listen to the transcripts, right? The meetings and all these things. So I'd probably just categorize all of this as unstructured data.

At this point, there's PDFs, there's notions, there's all these, like, crazy data sources. So I'm just curious, are you leveraging these as well? and how do you overcome the sort of like noisiness? Like we can be talking about anything in a meeting. And so there's a lot of like deciphering, right? When it comes to like how much data that we want to be able to leverage from those kind of sources because it feels important, but also like has all kinds of stuff in there. Yeah.

And I mean, we are running in the tens of millions of tokens all the time. So there is a diminishing return. Actually, it gets like you can probably chart a curve on when too much context. actually yields worse results. And that's why the multi-agent approach is becoming really popular. We started from the beginning with a multi-agent approach where, like, one AI does a certain task. Okay, then it completes it, sends it over to the next AI, right?

And so with these note takers, for example, we don't really go to the transcripts. We have the note-takers already summarize it up, and then we use the summary of the transcript. So that way, the window that the AI has to go over, our AI, is much, much. much more limited, onboarding documents, right? As we onboard a new customer, and they have these onboarding documents, those are actually really good, because most of the time, they're not written by other AIs. They were written in the olden days before AIs, and they're full of really helpful information to understand for Orion and have context. And then every once in a while, like, we have a customer who just told me about this example where they had a competitor posted a

blog post, and they wanted to know, hey, Orion, take a look at this blog post and tell me how our own data verifies or actually has counterpoints to that blog post. Like they did it. It was amazing. Orion had like, okay, we're seeing the same trends over here in our data, but over there, this is very different. And you might want to point that out. So it's great to see these ways that people now start to use a tool like Orion, where

this was really not possible before. You couldn't take a dashboard and say, hey, dashboard, I just got a blog post, compare our dashboards to this one, you know, that just wasn't really possible. There's a couple of other use cases. It's like we have a customer where they have like contractual obligations and then their own data of how things are going in that large system. And so whenever there's a potential breach of contract or like obligations to the customer, they need someone to investigate. Did we really or did we not? Because the customer is saying they want their money back.

Now to do these investigations, you know, thousands of times took a huge team to do so. It's quite annoying work. That's a perfect use case or something like Orion. I'm so intrigued because I have so many questions there's so many of ways we can take this and I just keep coming back

to the same question I think at a high level we're like generative AI we have a sort of semantic layer I'm thinking about this like programmability you have which is sort of

like broadly I look at what you're doing is okay you're building like this sort of like agentic data layer for the business like that's kind of what it feels like I don't know if you define it that way but I'm curious like

other than just you know large language models what other like technical breakthroughs had to come through to enable you to build this sort of next generation data visualization and exploration tooling. Was there other pieces that we're missing or is it just like, oh, now we can really do something we could never have done before. And it purely is just a large language model. Yeah, I think on the last mile problem, right, the one where we just like explain this data to me, right, that it truly just was a large language model that was missing. I think

the step that is crucial, though, to us is that we don't just want to explain a dashboard to you. We actually want to think about Orion thinks about who you are in your organization and the questions you should be asking because I wanted to be proactive, right? I wanted to say, hey, Lucas, you know, on Monday you have this meeting with this customer. Let me create a little business review deck for you. So you have it going into your meeting. I want that like chief of staff that you have with Orion. And so the last mile, yes, we can totally, like tell me what this means to me and there's a very valuable thing but then the ability to think of questions and investigate required us to create these big workflows right and there's a bunch of

people now working on how to orchestrate them in some ways we're going back to human psychology of teams and how teams should operate with each other and because we have that has been studied for a long time and the LMs are built on materials that is based on our lives right so they're behaving very similar to how teams behave so we have all these many. Azure agents now that are keeping the team focused on the goal and like you have the creative one and you have the one that's skeptical and you want to create a good team dynamic. In terms of technologies that enabled this all, I mean, you know, I was there when the first, you know, Redshift, Big Query came online and, you know, before this, it was quite hard to have

these massive data operations. And then people starting to, there was this brief moment, well, not that brief actually where everybody just going wild west. You know, you had hundreds of tablo dashboards and then all that. And I'm glad that we came back to, hey, we should have a central governance layer because that was, you know, 2005 or so business object. That was the whole point. So I'm really glad we have that again because that really does enable, you know, use cases like we have here with Orion. If it was all Wild West, it's very challenging. I mean, I would say any business run on that kind of infrastructure is destined to have disagreed. agreements in a bunch of meetings because people have different numbers that they're looking at.

And so I'm very curious, like, when I think of AI business or data analyst, it's almost like a catch-all title that can do everything in any data type or even any company organization, because typically somebody wants some data to help inform a decision, and it could be all types of ways you get to that conclusion, right? And it's really interesting. Going to your website, you have four different type of, like, solutions, target different people. There's executives, right? There's data teams. And your case study is the blueberry farm, which is super intriguing. You know, I'm just trying to piece together, like, do you start with one of those personas in your referenceable solutions? Typically speaking, like executive team wants to do this first or this is marketing or just this data team or do we try all parallel threads?

Because I'm thinking of a blueberry farm. Maybe they do have all kinds, but it's not so clear to me. Everybody starts the same place, you know? Where do you find your success usually comes into? And how do you even get them to trust your AI analysts, right? That trusting process can go further in that company. Yeah, and there's the original vision, right, which was like, okay, this is your analyst for everybody. It's very much to your point.

It's very broad. like I want to build an enterprise brain. I want to be the trusted, the trusted tool out there in the background that enables all these support, you know, all these support use cases for your organization. And I want to even take it further. I want to take it to action. You know, if I can help you find that your cloud spend is inefficiently allocated or your servers are idle somewhere, why don't I just shut it down for you?

Like once you trust the engine that comes up with the analysis and the recommendation, why don't we just take the action for you? And that's what I have done many times before in my career, right? Like, now I can do it with AI. So I'm really excited for where we're going. But where we are right now, the most common use case that we have, and I think that's where the most highest urgency to buy is, is with product managers.

Most of our customers are actually other tech companies. They need to provide something interesting to their customers. So Orion creates the analysis for their customers. And it creates a QBR decks or the business review decks for the account managers. It highlights upsell opportunities and potential sharing risks. This is a true reporting at scale. There's 10,000 customers that lock into your portal. They all need their own custom report.

So that's really the number one use case right now. That doesn't mean I want to give up on the executive weekly report, right? Where Orion on Sunday evening, it writes up what's all coming this week and what you should pay attention to any optimization scenario. There's so many opportunities. But the product manager is right now on number one buying persona. What do you think is the wedge opportunity, do you enter an organization? Do you think the thing that gets people to say, you know what, let's go and reinvent or, you know, let's evaluate new data tools after they spent like 20, the last four years investing a lot into their data stack?

Is it, hey, that stuff didn't work. Let's try something new. Is it, hey, this is going to make us like a lot more efficient? So it's going to make our team a lot more efficient. What's the core unlock here that you represent? Because I think a lot of what you're saying makes a ton of sense. I'm just also very curious when you're thinking about your good market motion, when you're thinking about what you're doing today.

Like, how do you think your customers are thinking about you? Man, a lot of good questions here. So that is actually one of our maybe challenges right now. There is no line item we're replacing per se. Like we're building on the shoulders of giants. The database needs to be there, right? Like some sort of data catalog or semantic layer needs to be there. Usually the analytics team is understaffed.

Or in the example of, hey, I want to create 10,000 custom reports for each one of my customers. Yeah, we have never done this before. This is a brand new thing we want to do now. And we see that as a competitive differentiation in the market. And that's why we're buying all right. So there's that. I would say, you know, I don't think there's a future where you need a huge data analyst team. That's maybe my hot take here.

You need someone to run strategy of data as your product and your organization. You need someone to own the architecture of your data layer. there is another potential secondary hot take we could talk about. And that is we are coming across organizations that have not invested in a data stack. And they're like, well, can you just do the API calls? It's really easy now to do API calls. And now everybody's doing MCP servers, right? We will see as I'm not sure if that could be interesting.

There are use cases where you need the data centralized to do massive joints. But there's a lot of use cases where we don't need to do that. So we have some, like, let's say, a medium-sized Shopify account owner, right? We pull in their Google Analytics, we pull in their Shopify, we pull in their QuickBooks, if needed or so. And actually, you know, Orion's in-memory capacity right now is about 12 million rows. Most organizations are well-served with that. I don't want to say we don't need databases anymore, right? There's still a reason for them.

But even if you're lacked behind on your investment in the data infrastructure, there might actually be a very opportune moment in time right now. I think that makes a lot of sense, right? And I think that also mirrors, yeah, because there's no line item. But the broad sense of, from my understanding, of, like, data teams, the data stack is we spent, you know, a long time investing the last four, five, six years since 2020, specifically investing in building, like, modern data infrastructure. So we understand our organizations. We have seen some upside. We've seen some downside. It's still broadly arcane.

We've also seen a deep level of consolidation. And I love your hot take around, you know, the future of the data. analyst. I mean, what about the rest of the data world? Like, what do you think about things like data engineering? What do you think about things? What do you think about the OLAP, sort of, you know, business analytics use case versus more of the online operational use cases for driving real-time data decision-making, let's say, in a product? Like, what do you think about the future of those aspects as well in the context of what you're building with Orion? I think on the data engineering side, it will be quite similar

to data analytics where a much smaller team can handle a lot because I know all the managers of AIs, right? I think that's the, this is important skill set that, that anyone who wants to be active in this field going forward will need to have, which is managing and delegating. If you love just doing everything yourself and put your headphones on and, you know, like, that's me. I think you have a tough future head because delegation and I mean AI, not delegating to other people here, right? Like, and then piecing it together and making sense of it and say, yes, this is good. This is a good ETL script. This is a good pipeline, right? This is good transformation. And that is a pretty advanced person, right?

And you've got to pay attention. You've got to be there when the results come in and you've got to be reading it mindfully. It's a bit of a different job than a lot of people had right now. The train left the station, right? So if you're not on it, I would run fast to jump. That is the skill set of the future. We talk a little bit about your internals. Of course, we're not going to go through everything.

But is there one challenge you have to go through that's probably not obvious for the most folks when you're building almost like an AI and analyst? We talk of semantic layer. We talk about context rot almost, right? We heard about that term. Is there something else that, hey, this is actually quite unique and this is not that easy to figure out? I mean, I think the challenge that every AI company has,

and we have it too, right? Anyone who wants to take it beyond a, you know, we vibe coded this and we have now proof of concept out and a couple of people are using it. The problem is if you want to be in the enterprise, you have to be correct 100%. You can't have it be incorrect, right? And so anything that was a prototype,

is going to fail flat, right? Because as soon you lose the trust in a second. And so for us to, you know, we build more complicated and complicated systems to ensure accuracy. That is the big investment that we have made, just a very, like a incredibly diligent quality assurance run, which I would argue that is where a lot of our IP is. The second one, and this is interesting as we think about the future, right, like I'm building an AI knowledge worker, right?

we're building a knowledge worker here in the past, right? Ian, you were saying, right, we have invested billions in this data stack. We have invested billions in a data stack that serves the knowledge worker. Now we're building the knowledge worker that uses these technologies. And that's a really interesting shift in thinking. But in order for that knowledge worker to be well-rounded, it needs to have more than just access to your data. So we put a proprietary data set together of like outside information that is actually quite, you know, sometimes we just knew that there was, you know, a special event happening in

the news and that's why things dropped. Or there was a hurricane in Florida and that's why, you know, this happened. And so we had to put that in because not everything can be explained in this lockdown silo that is your own organization. And, you know, a good analyst, a human analyst, they would be like, yeah, duh, of course. That's why this number went up or down. And so we had to add that to or Ryan, just to be a more well-rounded knowledge worker. Awesome. Well, we're going to jump into our favorite section of our pod. It's called the spicy future.

So give it to us. What's your spicy hot take that most people don't believe in yet? Oh, man. Okay, so this is where I get in hot waters. But I have worked with hundreds of organizations on data, you know, through my time at Looker, through my time at Google. And the number of times I came across like a brilliant data analyst, it's rare. It's like a diamond in the rough.

Like you come in and people are there to make it to the end of the day and bench watch another Netflix show. Like they're not passionate about what they're doing. They're not going to dig a little deeper, right? As soon as they find the something, they're going to take that and present it up. Like I've seen horrible analysis where like, I don't know, people forgot to copy down the formula and now the profit was actually incorrect, right? like big companies where maybe even sometimes publicly traded companies. And so I was so frustrated, right?

Like, I'm German and I'm very stereotypical German, right? Like, it's a way of doing things. And what I'm building here, what we're building is a really good analyst. Like, for the first couple months, we had Orion compete against me. And I was beating Orion. I found better things. I found it faster. And the last nine months or so, there's no chance.

Like, even if I had 40 more hours, like, I would get frustrated with this organization. I would like the date, why did they structure it this way, right? I don't want to do this anymore. Orion doesn't get frustrated. It keeps going and it finds the things. And then it writes them up very nicely. And when you give it feedback, it takes your feedback into a cone. It doesn't get upset that you just didn't take it analysis as it wants it presented. So there are some human flaws we have. Right. We have the flaw that if we see a green number, we just move on. We only pay attention to red numbers. Orion doesn't have this bias. Right. It doesn't get tired. It doesn't get annoyed. I mean, I think there's still amazing analysts. out there that would find very unique things that an AI just wouldn't think about right now. But there's, I think, over 900,000 analysts in the U.S. You know, as you said, it's a very broad title that a lot of people have. Every organization has analysts somewhere. I would say there is a – it's also a very short tenure, right?

A lot – if you're a really good analyst, you become management. You move on. So the average age, I think, across analysts is like 25 or so because it's a good entry-level job to get into. So that's a very hot take. I think if you are a young analyst, right, who wants to get ahead, think of data as your product or become the architect of the data underneath because those are really important jobs that are always going to be there.

I'm a CEO of a company, right? I want a human there that is in charge, and that is always going to be a job. Like, how much do you think our existing software is a reason that these data analysts are so checked out, right? Like, if you think about their job, it is unfortunately oftentimes, and I don't mean this in a negative way, I just mean, like, the day-to-day. work of being a data analyst, unfortunately, has become the equivalent of, like, the modern line cook at a McDonald's restaurant, right? You're basically taking, like, orders from somebody

you don't know to solve a problem you're detached from to deliver them up, like, some fries so they can, like, you know, eat their low value, high caloric food, right? And so do you think that's a manufacturer of the software and the limitation software has had and what enables organizations to do? Or do you think that's just, like, the way organizations work? Oh man, that's a hot take right there. The analyst is the... Yeah, that's a hot analysis, right? A hot analogy.

But it feels that way. Sometimes when you're like, you know, even in organizations I've worked in and it's the data analyst is just a really hard role. I can understand why if you are a data analyst, why you would find it really difficult to do that job for a very long period of time, you know? Yeah. No, absolutely. And I think the, I mean, I was like my team, we once called ourselves the query monkeys. We felt like, you know, we were just being told to do. like a perform a little act, right?

And then like, hey, can you tweak the analysis so it makes my boss look more favorable? But like, you get those requests back and you're like, man, like that's really not what I'm here for. Yeah. And I think the really good professionals I've seen in the data space are actually, you know, practitioners that love digging into data. I think of this one guy. He was the VP of marketing. And man, he was sharp.

He was good, right? He would pull up his SQL workbench and just write some SQL himself. and he felt like it. That's where probably an analyst should be, like be a practitioner in the field, like a data-minded person in the field, like in the actual business. And I think that is where you feel empowered

and you actually are connected to the decisions that you're making and influencing. And with tools that we have now, you don't need to open up SQL Workbench anymore. You can just talk to your data and have Orion do the groundwork for you and get the insights back. Then data analysts right now, right,

you're stuck between a rock and a hard place because you have so many requests coming in, the data is not there or it's not structured well, right? You weren't in the meeting, so you don't have the full context. And then you sent the report up and nobody ever told you anything. If it was used, if it was helpful, right, like you just send it off, that's a tough world. So I think if you step it up and say,

hey, I'm just going to own the strategy now for my organization on how we mature with data and which data sets we should bring in, right? Like, that was my path out. Like, I just changed my title to be data product manager. So nobody complained, turned out of work. Yeah, reminds me the word data scientist in the first place. So I think maybe just related to this question of, I think we look at AI, has the power actually able to do way better than humans,

as you alluded to. But fundamentally, executives still trust people, right? Because I'm seeing somebody that I hired that coming from this background, showing me, telling me this story. And so we still believe in human way more than machines. in a lot of cases. And do you see there's a conflection point that happens when, like, someone like your chat TV is so good, I trust it over time.

But it's just the product yet you're delivering and the quality of outputs getting better to be the trust factor to kind of tip over those folks? Or do you think there's something else that needs to happen? We're like, you know what, Orion, it's going to be so much better than even my analysts. And what do you see is the thing that needs to happen to get through that? Yeah. Well, a good friend of mine, I told him, look, I always want to have a human there that tells me, you know, even the lawyers. Like, I'm not a big fan of lawyers, but you've got to have them, right?

So I have my lawyer. I know he's using all kinds of AIs behind him and he's charging a lot of money for it. But I still hire him because, you know, that's a trust factor. And so I told my friend about this. And I said, I think that's the same for data analysis, right? You always want to have that person there that even if Orion created the slide deck for you, right? There's a person telling you about it and it's their career at stake here. And so that's why you trust them.

And he told me, well, you're having a real luxury problem here. You have too much money to spend. And that's why you can afford having a human that you hire just so you have someone to look at. And it's like, if you get into the really competitive fields where margins matter, why can you still justify that you want a human in between you and the AI? I don't know. I mean, I think we'll have to see how the next couple of years here develop. I think there's a value there to have someone thinking about, you know, where things are going in your organization with data.

I still employ humans, right? I still employ a bunch of them. I value that. I value the exchange of ideas. But let's see where things go. It's a very interesting world we live in right now. And I think it's really fortunate that we can be part of the conversation, you know, and be in this conversation. Yeah, we have so many questions we can continue, but just based on time, for folks that want to able to learn more, more about Orion and what you guys doing. Where can I find you? Yeah, bygravity.com. So we have Orion here, which is our main product. There will be other ones. So bygravity.com is where you can find more about us. Awesome. Well, it's super great to have you on our pod. Thank you so much. This was super fun. Yeah. No, thank you. This was great. Yeah. I hope I, you know, I get sometimes really passionate. I talk really quickly and my German accent,

so I hope it all went well. Thank you. Thank you. Thank you.