S4 | E20 | Enterprise Data is Broken. Here's What's Next with Ethan Ding Co-Founder @ TextQL

Alex Hutchings: 00:00:03

Welcome to the Think Data podcast brought to you in partnership with Mydataworks. If you want to stay up to date with the latest breakthroughs and trends in the world of data and artificial intelligence, and if you're curious about some of the strategies that companies and founders use to launch data and AI products, then you're in the right place. Our aim is to bring together a diverse lineup of fantastic guests from the founders, through to accomplished leaders and product owners at some of the most fascinating data and AI companies worldwide. They will each offer you their own unique insight into what it takes to launch and scale a great data business. Thanks for tuning in, and I hope you enjoy the episode. This week, I'm joined by Ethan Ding. He is the co-founder and CEO of TextQL. Ethan is... building a data warehouse designed solely for the AI era where AI agents generate hundreds or even thousands of times more queries than humans. Before founding TextQL he worked in ventures at Contrary and Bessemer and he's become pretty well known for his thought-provoking views on AI infrastructure, enterprise software and where this industry is really heading. Well, welcome to the show, Ethan. It's really good to have you on. And for those that don't know, I know I kind of gave a bit of a brief intro into kind of who TextQL are and the high level narrative, but who is Ethan? And ultimately, I think it was what, four or five years ago when you looked to launch TextQL. Firstly, when was that Eureka moment and what's your backstory?

Ethan Ding: 00:01:46

Yeah, I'm not gonna lie. Like when we started this, I think I was like 23. Like I was starting with my co-founder, he was like 25. I was working as a data leader at a startup where I was basically the person who did all the SQL crunching and writing queries and executing and building dashboards for everybody else. And because of that, I was like, let's take a stab at trying to see if we can automate my job away. And so I partnered with my co-founder, Mark. He was doing AI research at Facebook. And the two of us decided, you know, we were on the younger side. And so we kind of knew going in. Like whatever we were going to be able to build, it was not like we aren't going to be able to build a company that's built off like having being like family friends with like the CIOs of all the big, big companies or anything. And we were going to have to build a product that's like a lot better. And so we went to work talking to a lot of customers over the course of like two and a half years trying to figure out like how do you automate like data analytics, right? How do you fully automate like the work of a data scientist or data analyst? was a very it was it was a very painful like two and a half years the models weren't quite good enough uh to do like texas sequel it was very very common in the crowded space i think basically every other company that started to do this at the same time as us uh went out of business but like for us the the they're like eureka moment that really takes us to today is uh there's a there's a there's a point in time when we we heard like back to back with every cio we spoke to like every every single like head of data every single vp of data like listen i would love to use your product but my data is really messy And that's kind of like what really, really like made us like internalize everyone's data is always messy. Nobody's ever said, oh, my God, I have such clean data. I can't believe I have clean data. And so we have to like build AI that can work on like messy data where it is. Right. If you have like 50 different databases, if you have 30 different BI tools, we should still be able to make use of all of them. And so that's kind of what takes us takes us to like where we are today, which is that, you know, we build at TextUL, we build like AI and data systems for like the most regulated, messiest enterprises like on the planet.

Alex Hutchings: 00:03:44

Interesting, because obviously taking a look at your websites and the case studies, you're obviously working and you've solved problems for some pretty well-known brands out there. So obviously going back to, say, one of those businesses, what typically is the problem they're trying to solve? Because obviously you're from a data background. There's no shortage of good analysts out there, some exceptional analysts, and there's no shortage of tooling to try and make their job easier. but you know what

Ethan Ding: 00:04:12

problem are they ultimately asking you to say to solve here yeah the the problem is something like i want to see a chart uh and like i frequently want to see charts my data is all over the place every single time i like make a request to my team it takes like weeks and weeks and weeks to get that information back to me sometimes it means uh the answer is uh hey by the way like we're not going to be able to get this for you until we hire deloitte for another you know 35 million dollars uh to clean up this data set and the story that we tell is like we don't need you to migrate we We don't need you to. like lock yourself into some data platform. We're going to clean up the data. Our AI is going to literally bring you the dashboard for you, clean the data along the way, do all the joins, do all the transformations, do all the searching and get you like the charts, right? At the end that you need to make your decisions.

Alex Hutchings: 00:04:58

Interesting. So obviously going back five, 10 years, there's so many different legacy data platforms and data on your point, very disparate data setups. And obviously in the new world, you've got the big players out there. How does TextQL work across the older legacy systems? And obviously the more recent, I'd say recent in general terms, cloud-based SQL environments, does it literally work agnostically across all of those? And how does it clean those in practice?

Ethan Ding: 00:05:24

Yeah. So like the really hard, so for the past like 35 years. If you wanted to build a database company, you had to go to VCs and basically say, I'm going to have this like algorithm or something that's going to be slightly better or more efficient than the other systems. Right. You have the transitions from like renting like DB2 mainframes or like like DB2 databases from IBM in their basement to Teradata databases in your basement to cloud like like like databases for AWS, but storage and computer coupled. And then you've like seen the transition to write open storage, open compute, open everything else. And um And the story that you kind of have to tell investors, because building a databases company is, it's a monumental endeavor, right? It's next to like building an AI lab. It's one of the most capex or like labor expensive, like R&D projects you can embark on. And you have to kind of tell VCs like, hey, we're going to lock them in. We're going to use my proprietary table format and my proprietary storage format and my proprietary query engine and my proprietary, like DSL and like query language. And it's like, oh, we're going to do this thing. They're going to know everything. They're going to have to learn our system and then they'll be stuck with us and they'll pay us like, you know, infinite dollars like forever. Right. If you look at the revenues on some of these databases like that have been around for 40 years, they have like 98 percent gross retention, which which means like every single year, like they may maybe experience like 2 percent churn. Like a very high gross margin product. Yeah. And so like we were we we we we thought to ourselves like we have to build all of the abstractions that make it possible for you to do joins. between these things just in place. So which we had to build like Rosetta Stone infra that lets you query from one language into every single language, right? We had to build Rosetta Stone infra to make it so that every single table format can be unified into one table format in the cloud, right? Or on-prem depending on where it lives. We had to build loading protocols to make it so that you can bring those things together into the same plane where they can like actually see each other. They can fit on the same computer. We basically had to reinvent like the entire like data ecosystem around like not needing to migrate and not needing to like, you know, have a human do ETL jobs for lack of a better way to describe it.

Alex Hutchings: 00:07:35

Thinking back to the early days of TextQL, obviously you're still early on in terms of your journey, but I'm assuming there was some questions or some problem statements that you got asked if you could answer or solve. Then I'm assuming you would have gone, actually, we'd love to be able to solve this. So how much of your...

Ethan Ding: 00:07:53

product development was based on the feedback and demands of those early clients and how did you then evolve this in time i think from the perspective of like from the perspective of like a like a early career like younger founder um i think a lot of i think like this the the steve the lean startup like advice is pretty bad like you don't generally like want to feel like like you do whatever like the customer tells you to you want you want to like figure out like what what causes them pain. And then you want to like figure out like, can that thing even be solved? Right. Like, I think we've got a lot of feedback around like things like, oh, one of the big things is we got a lot of feedback around transparency and like making it so that the user can interpret like what's going on beneath the scenes with the AI and be able to be fully deterministic and like the outputs, right. Make sure that all the answers like perfectly accurate. There's a place for that. And there's a place in our product that you can define determinism. But what we really found is that it's more because customers didn't. like the one out of 10 experiences that like might've had like not a good time. And it's like those, as those like a one out of 10 experiences shrunk to like one in a hundred or like one in a thousand as our products evolves, they stopped complaining about like determinism, even though the product wasn't any more deterministic. It was just like better at producing like a non-deterministic result that the customer liked. And, and so like the, the, there's a lot of like framing things where like, it's like, why do you feel bad? Oh, because somebody complained to me. Why does that make you feel bad? Because I can't like explain to that person what's going on because I don't have full visibility and control over like how to fix what was going on. And so like for a lot of these things, like reading between the lines is what's really, really important for us in the AI space, right? Especially like because we're working in an industry where like determinism is extra important, right? On the regulatory side. Yeah.

Alex Hutchings: 00:09:37

Definitely. And in terms of your early kind of go to market motion and kind of getting that product market fit, what were some of the early challenges you came up against in terms of getting this product?

Ethan Ding: 00:09:47

to market and obviously getting people ultimately paying for your solution yeah i think we we talked to people like from like day one I think my co-founder, my co-founder is like one of the best engineers in the world. And he basically said, like, I'm not going to write a line of code until we have a customer that says, yes, I will pay for this. And that was like a grounding, like foundation of like our entire company is like, talk to the customer, talk to the customer, talk to the customer, spend time with the customer and like figure out like what they need. And so like, like the first thing we did, like I think the first like three months of starting, we got a sales coach. We learned how to like reach out to customers and we signed like, I think like a, like a $10,000 like, like contract or something in the first like, like, like five months or so. Basically every other customer that we signed like churned for the first like year and a half because the product didn't work. But it was a lot of like, like with enterprise data products, right? You, you have to get it into production to find out that it doesn't work. So like you can like, like, like build the thing in locally and then stare at it and it just like. looks like terrible because because you can't even test like you know it on a real production workload like you can run it on eight tables and it looks good and then it's like okay cool like can we run it on a customer's like environment and and the answer is like no not until like they give you the chance to try and if they give you the chance to try now you have to um like go through the security and everything else and if it fails you only have so many chances at that i think we took we were really fast with like getting into customers rebuilding the product and then taking another swing at it so i think we went through like eight iterations of like the product like building it taking it live losing the customer rebuilding it taking it live losing the customer etc i think a lot of our i think one of the things that a lot of our competitors didn't do right uh is is like take that many shots i think a lot of them like took like one or two is what i've like kind of heard like from like various people who worked at those places um and so like the like the the velocity at which you take those reps like did a lot because it let us iterate through like way more architectures than and find the one that finally did work yeah in the end.

Alex Hutchings: 00:11:49

That resilience as well, isn't it? That kind of resilience and that kind of North Star where you know where you want to get to, but you're okay to roll with those initial punches and use that to your kind of benefit to kind of reiterate, redevelop and then go back in again. Do you think your kind of customer profile has changed at all over the kind of last few years in terms of the type of customer that benefits from TechSQL solution? Or is it now almost seen as you could solve problems across all the ecosystem and landscape?

Ethan Ding: 00:12:15

Yeah, like... we specialize in messy data we specialize in distributed data we specialize in rows and columns and we end up being able to make the biggest difference in regulated industries because they have such messy data um but like like a like a like the way we think about like the market is if you plotted like an x y axis of like how messed up is their data which is often like a correlated with how old are the companies in the industry how long they've been around um how much regulation oversees them and as a result like how many how many like different distributed databases do they have right so like like like a, like a, like a large bank has basically every single database on the planet over the course of like the past, like, you know, 150 years. Whereas like the, the, the, the Y axes is like, how many analysts do they employ? Right. There are industries where like, you just don't need to be that data-driven to like win. Um, right. Like the, the, or alternatively the type of like data-driven is like very, very solved, right. Things like advertising and, and trading. Those are, those are solved like markets, right. Like those are, those are markets that are like extremely, extremely efficient. Things like CPG and like retail. um like pricing like in in in terms of like like you know walmart pricing these are like relatively solved markets like these are fairly like fairly efficient systems because like there's like like you know the the data is like not that regulated and you can be like reasonably like well balanced but things like insurance policies and underwriting and like like a lot of those those those problems inside the large insurance companies those are i guess like just less like well-solved problems there's a lot more surface area where you can like like you know we ask ourselves the question Like, can you... can you hire 100 analysts and produce another you know times their salary or something or something like like 10 to 20 million dollars stuff like roi and if that's the case at these like companies um then that's a good company for us to get into and and and that the the the willingness or propensity for that to be the case kind of varies a lot across across the companies yeah

Alex Hutchings: 00:14:04

that's fair and it goes back to your point right at the beginning about obviously you as an analyst and you wanted to find a product that effectively could replace you or do a better job than you And I know there's a... I'm fortunate enough, I've recruited analysts for years and obviously more recent AI engineers. But what's your honest thoughts on a product like TestQL and kind of the lifespan of a data analyst now? Because obviously there's a lot of people listening. The data analyst used to be seen as that kind of entry level way into data. And then obviously they climb the ranks and specialize. But obviously with this tooling, so efficient, people can interrogate it. get their queries and results pretty quickly. What's your thoughts on assessment of the data analyst role as we currently see it?

Ethan Ding: 00:14:48

Yeah, I'd say like, like, if you think about what, what is the job of a data analyst, it's not writing SQL, right? You can write like a trillion lines of SQL and like no dollars will appear. Like, like people are employed to like bring dollars into bank accounts. And then because they're they're they're paid and their compensation requires them to take dollars out of bank accounts um so like how does it how does an analyst produce money um and an analyst produces money for a company when they write some sequel they do some stuff they create some charts but the charts don't create the money the charts inform a decision the decision is slightly better than the other decision that would have been made without those charts that that's what we believe as an industry and and the extra you know marginal money from the if if expected value gain from you making that chart or dashboard or forecast or whatever helps you make a better decision than the alternative world and that decision that delta in like ev on that decision is how you make money and and and so like let's say like and today like the the typical analyst like lifetime is is spent 75 or something percent of that is basically spent on cleaning the data preparing the data right i think like everyone knows the jokes about like oh like like a data scientist like spends 90 of their time cleaning the data and then hits like one run button and like like trains the thing for the remaining 10 of their time that that that is the the case for like almost all data-driven tasks and so like the role of an analyst we see as going from basically like uh like it's a little bit like like uh like like somebody who might have previously that manually dug for coal um and now you have a machine that like digs out a lot of coal right but every single coal like nugget of coal might be like a like insight that leads to a decision where now you can hit you can cover a lot more surface area you but so can your competitors and so now you know if you're like the the pricing analyst at like walmart competing with the pricing analyst at like target now you're like like now you're you're building these machines that produce these decisions um right you're building these like loops of like agents that are looking at i don't know like like price of like diapers or price of like uh like like like shelf life or like out of stock like like trends i'm trying to like inform decisions that like help your your company um while like you know competing with like the analysts at the other company that's doing the same thing. on like some cadence and and so like i think the number of like insights and decisions that analysts produce will like go up like by uh by like like like orders of magnitude we personally believe as a company that like the demand for analytics is infinite and therefore if you let a human produce a hundred times more insights the demand for insights will maybe a thousand x right and and that means that you can actually hire now like 10 times more analysts but it doesn't look like analysts like filtering through filtering through like like making charts anymore. It's going to look like analysts saying like, hey, you should make this decision. I know you were going to do this. You should do this. Trust me. And then like build like more and more like systems and infrastructure around that like decision making loop.

Alex Hutchings: 00:17:39

Yeah, it's a really interesting point because I had someone on this a while back who said something similar. I think it was around they freeze up the analysts to do the work they probably want to do more of, you know, getting close to the business. you know, actually advising businesses on certain, using that data to form insights and actually help that organization to, as you say, grow revenue numbers and obviously attack new markets. I certainly see that. If you're in the kind of company now where you've just brought TextQL in, what does that kind of adoption process look like for teams? Are you, are TextQL going in and training the analysts? Are they actually just working with that CDO or CIO on kind of how to derive more value from this tool?

Ethan Ding: 00:18:22

Yeah, we typically come in through, like, the CIO or, like, CDO office, sometimes the CFO office, and then, like, plug in, like, all the data sets. And then afterwards, like, the UI is, like, pretty, like, intuitive for people to, like, start using. And people, like, blow out, like, the, you know, have some, like, giant bank of, like, Jira tickets they haven't, like, addressed. And a lot of people, like, get a lot of dopamine from, like, going through, burning through that, like, Jira ticket queue to, like, answer everything. But then, like, once you actually have, like, everything that people have asked answered, now you kind of are faced with a problem of, like. what do I do with all this like excess capacity? And, and, and we, we work with our clients to figure out what are the opportunities that you can do 10 times, a hundred times, a thousand times more analysis. Right. And, and like produce like, like decisions, right. Um, there are things like quarterly forecasts that you probably can't do like a thousand times because like, you know, like, like a minute by minute forecast of like, well, your revenue is going to be a minute from that. It's probably not very useful, but there are other things like, um, like like like like fraud detection where you know like going from turnaround times of like five minutes to five seconds actually does produce a lot of like like money on the table right there are things like they're like like pricing optimization there are things like a broad like like like underwriting where you know if you go from like underwriting like the age bracket of like 18 to 25 to underwriting the age bracket of like 18 to 18 and a half 18 and a half to like 19 right like being able to do like finer grain like analysis on like smaller signals um lets you like absolutely dominate like uh like the like competitive like dynamics and like a lot of these markets yeah interesting what's next to you guys i know obviously you're growing exponentially there's obviously

Alex Hutchings: 00:20:02

a lot of excitement around you guys as you said earlier kind of there's competitors have come and fallen by the wayside you're still going and still growing but what can guys who are listening here kind of see and expect to see from you over the next kind of 6-12 months?

Ethan Ding: 00:20:15

I think for us we're going to be building deeper and deeper integrations into like our existing clients I think like like for us like we work with some of the largest healthcare and like financial services companies in the world and so like going deeper on like various use cases and be able to find you know if i can find something at one bank that can like reliably like turn into like you know 500 million dollars of pnl impact right every single year um then i probably copy paste that at the next bank and maybe they're 80 smaller but like that that's like 550 million dollars of like pnl impact every single year um and so like like yeah like for a lot of these nodes like it's it's a lot of like like stuff like that um like going deep embedding deep and then figuring out like what kind of decision making loops do you have and what kind of data could i be like bringing to the table to like help.

Alex Hutchings: 00:21:01

Yeah. And what's your final message to say CDOs, CFOs who are listening to this about, you know, when they should speak to TechSQL? When is that moment? You know, is there a moment in time where they kind of have that eureka moment and think, oh, we definitely need some more eyes on this? What is that point?

Ethan Ding: 00:21:19

I think a lot of teams like try to build this or trying to build this out internally today. And like, first you try to hook up an LLM with a database. Then you. uh figure out that you want to hook it up to like multiple databases then you want to figure out like you can load them into an environment to like compare the data um and then you try to figure out like hey how do i do permissioning across these databases to all the users then you want to look at the user logs and figure out like how do you reuse some of the work so that you're not like burning as many tokens every time and then you look at it and you try to figure out like how to define metrics across databases if you're going to do those new workloads and like how to store that context and how to save that context how to manage that context and how to like like maintain to sow the truth as time goes on. this is a lot of the work that like usually people come to us like by like like stage two or three which is like they they they add like the third data source and then they're like oh i can't i don't want to like you have one table format you deal with another table format you figure out the like the thing in between and you get to the third one and you're like i don't want to and then you look at like the long stack of like like 12 to like 15 bi tools that you have and then people like kind of like lose it that's usually like when we when we try to come in

Alex Hutchings: 00:22:26

So yeah, they've already basically in short, they've tried it themselves, failed and they're gone. Let's wave the white flag and get you guys in. Cause it's just,

Ethan Ding: 00:22:33

I don't know if it's like failed. A lot of it is like a really happy, active users. And then they're like, I want to scale this and I want to scale this. I want to centralize everything. And the, the abstraction to centralize it is it, it ends up looking like, like you're like basically for all like software products, there's a, there's, there's, there are things that it's worth building in house because internal, like understanding like Trump's External like tech complexity. And then there are things that are like technically sufficiently complex that you don't want to build in-house. And from what we've seen, it seems like this is pretty technically complex. Yeah. But, but, you know, maybe like Fable 7, like one shot to this. And that's the end of the conversation.

Alex Hutchings: 00:23:13

Let's hope not. I think it's a fascinating conversation. I think as someone who's recruited in analytics for years, actually seeing this tool and your messaging and the complete transparency around that kind of role of the analyst and the benefit that this is bringing to organizations. I've got every confidence it's going to be a huge success. So thanks for coming on this morning, Ethan. It's been really interesting. And we will, your website, and as you can see some of the UI people listening here on their website. So when we go live, be sure to tag you and some of the case studies in as well.

Ethan Ding: 00:23:44

Yeah. Thanks for having me. I appreciate it.

Alex Hutchings: 00:23:46

Thanks, Ethan.