Drill to Detail - Drill to Detail Ep.108 "Mode, ThoughtSpot and AI as the Next Discontinuity" with Special Guest Benn Stancil

Episode Date: July 20, 2023

Mark Rittman is joined in this special episode of Drill to Detail by Benn Stancil, CTO and Founder at Mode to talk about their recent acquisition by ThoughtSpot, the economics of the modern data stack and how AI and LLMs could likely become our industry’s next discontinuity.We don’t need another SQL chatbotThe new philosophersThoughtSpot acquires Mode Analytics, a BI platform, for $200M in cash and stockChatGPT-Powered Data Analysis using Cube, Delphi and the Code Interpreter Plugin

Part of me is like the pace of the development and just the things it can do that don't seem like they should be possible. It's hard to look at that and think like there isn't some sort of foundational change that happens here. So, hello and welcome to a very special episode of the Drill to Detail podcast, and I'm your host, Mark Rittman. I've called it a very special episode because I'm joined by none other than Ben Stancil, CTO and founder of Mode, and a very well-known blogger and commentator on the data analytics industry. So welcome to the show, Ben, and it's great to have you join us. Thanks for having me. It's great to be here.

Okay. So Ben, for the one or two people who don't know who you are, give us a quick introduction to yourself and your role as CTO and founder of Mode, first of all. Sure. So as you said, I'm Ben, and one of the founders of a company called Mode. Mode is a BI tool that's now been around for about 10 years. My role at Mode has currently, my title currently is CTO. My role at Mode is basically been bounce around and do a bunch of different things,

which is fairly typical for founders of startups. So my background is largely as like a data analyst. That is where I came into Mode. It's the things I was doing before Mode, basically working on data teams, helping answer questions for businesses, building reports and dashboards, all that kind of fun stuff. At Mode, I've done a number of different things, sometimes working as a data person, sometimes leading our product team, sometimes working on solutions engineering or support. Currently, most of my focus is on either kind of product strategy or doing things externally with the community, trying to understand kind of where the market's going, where Mode fits into that.

And that's where the blog sort of has its space as well. So technically, the title is CTO. I'm not leading Mo's engineering teams. More of thinking about kind of the bigger direction for what kind of technology you want to build, where you want to go, and those sorts of things. Okay, fantastic. So a few episodes ago, we had Pete Fishman on the show.

And he actually told some of the original story about, I suppose, about how Mo was formed and Yammer and so on. But maybe just tell us a bit about how your work at Yammer led into Mode and maybe how you work with Pace as well. Yeah, for sure. So Phish is kind of like the godfather of Mode in some ways. So I joined Yammer in 2012. It was shortly before the acquisition by Microsoft. So Yammer was acquired by Microsoft. I joined Yammer's data team, which was split into two halves. It was a data engineering team and an analytics team that was analysts or data scientists or kind of

whatever you want to call them, that were kind of responsible for helping people around the business answer questions. And so I was on that half. And internally, we had built a couple tools that were basically designed to help the analysts distribute their work around the organization. That that we weren't trying to just build dashboards and reports, but we also weren't like the data scientists who are sitting in the corner doing hard math, building algorithms for for predicting products that people are going to buy. It was much more we get a question from a product manager being like, hey, we have two different features. How are people using them? What do we learn from that? And we needed ways to work technically. We were fairly technical folks that could write SQL and Python and things like that. We needed ways to work using those tools, but also be able to distribute that out to folks that were not necessarily technical. So they were,

in some cases, product managers or marketing managers or executives. And so we actually had built an internal tool that allowed us to do this. It was essentially a SQL editor in a browser with some charts on top. The analysts would write SQL in it. They would send links to folks with the charts explaining their answers. And so that's really where Mode came from was we saw that product be really successful inside of Yammer. We saw it spread inside of Microsoft after this acquisition. And then we actually saw versions of this product pop up at other leading tech companies that were kind of building data teams in the same way, like Facebook, LinkedIn, Uber. And so that's me, my other two co-founders, Derek Steer and Josh Ferguson decided to say, if this is a product that is successful across these companies and is sort of the beginning of a market around how data teams are working, then is there a product to build more generally

here? And so in a lot of ways, the story behind Mode began with the team that Fish built, and the three of us all were on that team and kind of saw the success of the internal tool and were inspired by that, basically. so people speak quite quite well of uh of pete fishman or fish as you call him um what was good about him what was good about what was good about him as a leader and um the way he ran the data team have interest yeah so so fish fish is one of the one of the best people in this industry uh both i think in terms of talent and just in terms of people um and I think that there are two things about Fish that to me made him really, really good at what he did. One is he was

like a business-focused analyst, basically. His entire motivation was, how do we solve these problems to make the business better? He wasn't interested in technology. He wasn't interested in ego. He was someone who was kind of at his core, curious about how the world worked and wanted to figure things out and figure out ways to make businesses better because of that. And I think there was a lot of like motivation in that that drove you to think about, hey, how do we solve problems rather than sort of how do we politically position ourselves around the organization and things like that. And he was very good at it. Like he's, he's someone who is as, as sort of a adept problem solver as I've ever met. And so, you know, he, he built teams that, that tried to live up to that. We often didn't, but, but he was someone who could very much lead from the front about how he thought about business problems. The second thing is, is like,

he's a genuinely really good guy, like honestly, and that sounds sort of, but that makes a big difference that the data industry can attract a lot of people who do this. There's a little bit of a, like a bedside manner problem among analysts where they want to come in and tell you how smart they are. And, you know, they've got the numbers and that sort of stuff. And, and Fish would never do that. Fish was someone who, who was there for his people. He was there again to help the business. He was one of the most egoless guys I've ever worked with. And so yeah, he is a true superstar to me in the space, both in terms of the talent that he has,

but also just the quality of human being that he is. Fantastic. Fantastic. So what was so good about Yammer as well? Because Yammer what was so good about Yammer as well? Because Yammer, people speak well about Yammer as well. Was there anything about Yammer in particular that was a good, I suppose, founding ground or kind of place to start what you're doing now? What was good about Yammer? So I can mostly speak about it as a data org. Yammer was one of the first consumerization of IT SaaS companies. So the approach that Yammer took, for better or for worse, was basically to build

Facebook for companies. And this was before Facebook for work or whatever it's called, workplace, whatever it's now called, existed. But the idea was like, hey, this kind of Facebook newsfeed thing works really well for Facebook and Twitter. What if we built it for collaborating at work? And what that did was the way they decided to build the product was, well, let's build it the same way Facebook does, which is oriented very much around user adoption, user oriented very much around trying to figure out how to make it viral. Let's like test features. Let's do all these sorts of sort of consumer product style development around the product rather than thinking about like, you know, how do we sell contracts and we measure ourselves entirely on ARR. Obviously ARR mattered, but a big focus of it was just like,

are people using the product? What is our daily engagement? Like daily engaged users was Yammer's North Star, not ARR growth or something like that. And so I think as a data organization, that opened up a lot of doors for us because it meant that everybody was thinking about a metric that was more easily movable. Like everybody was thinking about how do we drive engagement in this product in the same way that Facebook is always thinking about daily active users or whatever. And so I think that that gave the data team a lot of influence because it wasn't just, how do we close deals?

It was, how do we increase the number of people using this product on any given day by 10%? And you're sort of much closer to that problem as a data organization than you are, how do we close a contract that's going to take nine months to negotiate where most of that feels like it's handled through giving the right sales pitches, having the right conversations with the right people across that account. You know, when you're driving engagement, it's a much more sort of tactile problem for data folks. And so I think that gave us a lot of influence at an enterprise company that was not typical, certainly certainly at the time and still kind of isn't typical today but but it was a it was a breeding ground i think for a lot of ways of thinking um for enterprise products that is a little bit uncommon okay okay and then obviously the other thing people know you for now is is your is your sort of your newsletter on fridays um and everything

you said so far wouldn't necessarily lead into it someone a blogger or sort of like an opinion person. So what was the genesis of that and what drives you on with the kind of the articles you write? And how does that fit with your role as a kind of a CTO at Mode, really? So the impetus behind the blog was like it's all been sort of an accidental thing with, with, there is no grand strategy here. Uh, so, so basically when we, when we first started mode, um, there were three of us, there was, there was me, Derek and Josh, as I mentioned, Josh was, Josh was our founding engineer. Basically he, he was building the products like day one, Josh, we got to build a product, go build a product. Uh, Derek was our CEO. Uh, he was personable and well-lik which is good uh you know he was

someone who could be the face of the company and josh and i couldn't um and so he went out and talked to investors and customers and things like that you know he was he was suitable for that me not being technical and not being you know uh someone you want to put in front of investors i like basically had nothing to do um and so the very first thing I did was I just started writing a blog. And at this point, it wasn't kind of traditional thought leadership. It wasn't like five things to build a successful data team. It was things that I was interested in, which was just like analysis of public data. So the very first blog post on Node's blog was actually about Miley Cyrus.

It was kind of like 538 style stuff. I started doing that as a way to, technically, it was like a way to try to drive interest in the product. Partly it was just like, well, this is a thing we'll try and see what happens. It ended up being something I enjoyed. It ended up being something that worked reasonably well. For a long time after doing that, I ended up bouncing around these other different roles I mentioned. And then at some point, we ended up hiring sort of the appropriate experts in those roles. And once again, out of a job. And so I was like, all right, well, I'll try the blog thing again. It has taken on sort of a different character from what it was now 10 years ago. But that's really what kind of led to it was like, I'll try to try to see what happens here. And, and again, the, the, the overall like strategy behind it at this point is, is not existent. It is, I write one this week and then the strategy is, well, I guess I'll do it again next week. But, but for me, it's like kind of thinking out loud. It's, I've been thinking about this space for a long time. What's fun to me is

hopefully fun to other people. And, and so if I try to talk about the things that I'm interested in, then the hope is that other people will find it interesting. But there's not really any giant narrative this thing is all rolling up to. I have no strategy for what happens on day two here. Yeah, excellent, excellent. So in terms of topics for this episode,

there's three things I want to talk to you about really. One is obviously about mode and it's particularly topical now because of the recent announcement that you made during Snowflake Summit. Then also there's two topics that I've really picked up on from your newsletters, but also things that I'm interested in. So it's the economics of the modern data stack industry. And finally, talk about LLMs and AI and where you see that going, because that is the kind of topic of the day at the moment, really. But let's kind of start off with mode, really.

And so you gave a kind of a positive history earlier on of mode. But the question I have, really, is maybe at the start, what led you to build something rather than pick one of the tools that was out there at the time? So why build rather than buy for you at the start? So there wasn't really a product that solved the problem we were trying to solve. We were in this in-between spot where it wasn't traditional BI. There were BI tools out there. Tableau was established at that point. It wasn't as big as it is today, but Tableau was there.

There were the micro strategies and the business objects and things like that that were kind of traditional BI tools. And then there were kind of data science tools. There were things like RStudio or there were Jupyter Notebooks or tools that were very much kind of technical tools meant for technical folks. The problem was they were both kind of too far on either end of that spectrum where we weren't, we as a data team didn't want to use a BI tool

that was just designed for basic reporting because we needed something that was kind of closer to the metal. We needed something where we could write custom queries. We could answer these questions that weren't what's happening, but like help us figure out why it's happening. We were doing all of this kind of investigative work that required digging a lot further than you could get by kind of building a simple dashboard. But the other side of that, like, like we couldn't send the CEO of Yammer,

uh, uh, an R markdown file. Like it couldn't be like, here's our analysis. It's just open up a Jupyter notebook and run it like that. That didn't work either. And so we needed something that allowed that kind of had two sides. It allowed us as technical folks to do the work we needed to do, but allowed the kind of non-technical consumers to have an easy way to look at it, to collaborate around it and things like that. And there just wasn't really much product there that would help us do this. So we ended up building a thing. There's also, to be entirely honest, like a bit of a bias in Silicon Valley. You got a bunch of engineers there,

you know, going to be inclined to say like, I can build a better version of this. And we had a team that in their case wanted to do that and actually could. And so I think that's really where we went that direction instead of buying something. We just couldn't find anything that actually solved that need. Okay, okay.

The product was spun out then to become Mode. And I suppose I was going to ask you what kind of user persona is it aimed at but i guess it was aimed at people like you so maybe product-led growth or or people who were more technical but wanted a bit more than just kind of i suppose just writing sequel themselves i mean did you have a persona in mind and did you ever kind of cite an idea about the size of the market you were going for at the time so we did um our persona at the time was there were sort of two ways there was a sort of the persona that we imagined we would sell to and the persona we probably did sell to the persona we sold to was was basically data teams that looked like ours at the time when we

first started it there weren't a ton of those. There were some, and certainly in Silicon Valley, they were becoming increasingly popular. These were, again, not BI teams and not data science teams in the sense of they're trying to win Kaggle competitions, but data teams that are trying to solve business problems. Tended to be somewhat technical. They were all over the place in Silicon Valley. They were starting to pop up in media companies and healthcare companies and fintech companies. And so we were selling mostly to those folks. The audience that we pitched, we have old pitch decks from 10 years ago now, was like anybody who writes SQL. And that to some extent was true. And that was a very big market then. It's still a very big market.

But it's slightly more nuanced than that when you're really solving the problem for people who aren't just writing SQL, but people who are these kind of problem solvers who try to solve problems, you know, again, using data, answering questions more flexibly. Speed matters a lot. It's not just report builders, but, again, people who are kind of partners to the business to help them. Okay. So how did you see Mode in, say, relation to look at the obvious other player in this market or certainly the very well-known player in this market? Yeah, it was. So we actually, in some ways, the two tools were very complementary. We competed and for most of

Mode's sort of history, we always would have like two or three competitors that were the folks we'd run into. Now that set evolved over time, but Looker has been one of the most consistent ones there. But that said, in a lot of ways, the competition with Looker wasn't so much a direct one. It was more of what style do you want to favor? So Looker is shaped like a traditional BI tool in that you build a data model, you get reports on top. The way that that all works is slightly more modern. You're not building a data model with a drag and drop thing. You're building it with code and kind of a YAML-like interface like the LookML.

It runs in the cloud. It runs on top of cloud data warehouses. It pushes compute down. It doesn't have like a data ingestion. So it's architected in a way that is kind of for what I would say is the modern data stack, which is a phase we may be kind of getting out of now. But it was architected very much for that world. But it was traditional BI. It was the role of the product is to build reporting for the rest of the business. And it's like very governance focused. It's a sandbox for non-technical people to explore data, not an environment for analysts to answer questions outside of that sandbox. To them, writing SQL was kind of a backdoor. They had

this SQL runner thing, but they always saw it as like a tape hat for when the look email didn't work, but it wasn't really where you were supposed to be. Our approach was, there's a lot of questions that you don't actually anticipate, that you can't sort of pre-model, that you need something that's flexible, that's faster, that's those sorts of things. And so really the competition with Looker wasn't so much, these are two of the exact same products, which set do you prefer it was more do you favor the heavily governed sort of tightly controlled traditional bi data model or do you favor the like fast flexible write a sql query build a dashboard really quickly kind of approach to it and so when we won against looker it was when teams that favored our model when we lost looker it was when teams their model it wasn't so we lost a looker, it was when teams favored their model. It wasn't so much of a like, hey, these are a feature for feature thing.

Which one do you like to look and feel of better? They tended to just solve kind of different philosophical approaches to the way to deliver BI to customers. Okay. Okay. So I'm just considering here that there might be some people on this listening to this who haven't really used mode or or are struggling to kind of in a way or trying to picture in their mind and i'm in particular the fact that you've got the python side to mode as well um maybe just kind of just paint a picture what using mode is like and and and also how does python come into it sure so so basically the the very simple version

of mode is it's a sql, again, with some charts on top. Basically what you can – so it's a product with two sides. So there is like a technical editing experience, and then there's a presentation experience. The way it essentially works is you connect it to databases. It sits on top of warehouses like Snowflake or BigQuery or Redshift or SQL Server or whatever. Analysts or technical folks can come in. You have this kind of IDE for running queries against that data, for taking the results of those queries,

loading them into Python environments. There's a very flexible and powerful visualization engine that sits on top of it that allows you to do a lot of the kinds of things you could do in something like Tableau. And then you can take the results of all of those things and package them up into a nice dashboard that you can arrange kind of in the way that you would expect of dragging in little cells and things like that. So the idea is if you want to, say, quickly create a dashboard, as an analyst, you could open it up, you can run a couple SQL queries, you can put a handful of charts on top of those SQL queries, and then you have a URL you can share with someone that they can always get back to that result.

They can refresh it on a click, they can schedule it, have it delivered to them, that kind of stuff. And then if the non-technical person wants to come in and explore it, they get kind of a, again, a Tableau-like interface

to come in and drag and drop their way to explore and get it further doing drill downs and additional filters and all that sort of stuff. So the idea is that there are a lot of dashboards that people ask for. If you are a business user, hey, like, can I get, can I get a new report on the number of signups that we've gotten from this marketing campaign? The way that traditionally would work is analysts would come in, they'd write some SQL query,

they'd take the results, they'd stick them in Excel, they'd make a chart, they'd copy the chart into an email, they'd send an email, and that's it. We basically try to streamline that in one place where you can do all of that, where it's like write the query in mode, put the chart in mode, send the URL to the person, and then they have it all packaged together. When they want to run it or drill into it, they have it kind of live against the database. Again, it is similar in structure then to a BI tool where you might build that in BI and a BI tool, that kind of dashboard and BI tool. But in most BI tools, you have to go in and update an underlying data model to get there. There is, in Looker's case, editing the LookML, which again is more of a, it's like,

it is sturdy and there's a lot of governance in that, but it's also slower where if you don't have that data model, then there's a lot of work to do to go in and kind of update that underlying structure. And so for us, it was more of, can we really quickly enable analysts once they get those questions to answer them kind of by querying the data directly, by building data products on top of that directly. And it really favored this kind of speed and flexibility approach, again, over one that was tied down to a data model. Now, again, obviously data models also bring some structure and some governance that is valuable. So that's where the philosophical approach between the two products is different.

Okay. So originally when I planned out this episode with you, my next question was going to be a polite dance around the topic, really, of, I suppose, why was Mode, in my mind, like, say, one of the indie bands from my youth, like, say, Talking Heads or the Pixies or whatever, where it kind of had a lot of influence and was very well respected, but wasn't particularly kind of, I suppose, the biggest in the market, really. Or certainly wasn't sort of maybe as talked about now as it was maybe a couple of years ago um but then um but then there was the announcement uh during snowflake summit of um of mode being acquired by thought spot which was kind of very interesting um because it certainly put the back in the headlines and it certainly started to kind of like i don't know put you back into the center of things so maybe just tell us a bit about um what that acquisition um but also what was the what was the lead up to it what's the rationale for it really um and um

and i suppose also lead will lead on to where does mode fit into the thought spot sort of story but i suppose why did you start talking to them and what was the lead up to this yes so they kind of answered all of that in in. The arc of that is all kind of the same. So to your point of like mode sort of had presence and then faded, I think a lot of that is kind of just like Silicon Valley hype cycle, to be honest, that people care about the next new thing, not necessarily the biggest thing. And so if you look at Mo's sort of revenue growth, it's a chart that doesn't have like big, it hasn't like flatlined, it doesn't have big dips. It's basically a chart that's kind of been steadying on up. And so I think that the mind share of Mo is one that back when we were early, it was kind of, oh, this novel new thing of SQL and Python all together. And, you know, over the course of that, people get, the company continues to grow, but people find the next, the next new thing. And so the, the companies in the headlines aren't always the biggest. I think actually Looker in some cases is a good example of this.

Looker as a business has gotten very big. But in terms of Mindshare, Looker is a thing that, that is not nearly what it was in 2018, 2019, but as a business, it's probably five or six times bigger than it was then. And so I think part of that is just, you know, that people care about the new, new thing and in Silicon Valley, not necessarily the big one for us. And so, so what does that mean for us? And like, why do we end up with hotspot? Basically for mode, we focused on these data teams. We saw ourselves as either complementary to existing BI tools or in cases when companies were very focused on doing BI the way that we wanted to do it, which was this kind of fast and flexible thing driven by data teams. We could be the entire BI solution. But oftentimes, especially at large companies that have multiple tools, they would have kind of a reporting tool for reporting. And then mode was the thing where they would deliver the kind of ad hoc analysis, the strategic work, the kinds of

things that were the questions that needed to get answered outside of the BI tool. That market of like selling exclusively to data teams isn't huge. Like it is not, it's a good market. They're valuable customers. But in order to build a really, really big business in this space, you get pulled into being BI. And I've like said this a number of times in the blog, basically, that there's a whole bunch of companies that start off saying they're not BI tools

that like we sell embedded analytics. We sell to data team. We sell now AI chatbots. These things aren't BI. There's something different and special and more narrow and more focused. It all sort of blends. And ultimately, at the end of the day, everybody gets pulled into being like, well, can you build some dashboards? Can you build some self-serve reporting? Can you sell to business users? That's what customers are going to want. And it's where

the money is. And to your question about economics that we'll get to, if you're chasing huge Silicon Valley VC backed outcomes, you're going to find yourself slowly drifting towards building a traditional BI tool. And so our arc was, okay, well, we built this thing for analysts. Analysts really like it. We found ourselves slowly drifting towards building a BI tool. And I think to your point of where was mode in the market, I think a lot of it was we were building a traditional BI tool and ain't nobody excited about that. We can sell it and make money, but it's not the thing that gets tech crunch headlines is, you know, another company builds BI. We were building from that direction. So we were basically building a thing that we were, it was popular with analysts.

We were starting to build more and more kind of business user self-serve type of functionality. We really expanded our visualization capabilities, things like that. ThoughtSpot was a company that very much started with that world. They were very focused on the business users. They've done a great job of building a very flexible self-serve tool. It's something that's loved by those folks. It is not a technical tool. ThoughtSpot is a tool that does not have a terribly technical side to it. Data teams often would find ways to say like, great, we like ThoughtSpot for our business users, but we're not going to work in it to answer our questions because it's just not built for that. And so they found themselves kind of drifting in our direction of, hey, we got to solve this problem for like,

how do we build a BI product that is also loved by data teams? We were a data team product that we're starting to build a BI tool that needed to be loved by business users. And so they actually approached us because they were starting to think about what is it they can do to build a more kind of complete end-to-end solution that also is appealing to data teams. They knew that we were popular with those. They came to us and said, hey, this is kind of the direction we're going. We realized we were basically going in the same direction as them. And so it's like, if we're both kind of trying to build the same product from opposite sides, and we have these, these assets that are sort of completing the vision for each of us, if they were combined together, we realized it made a lot of sense to bring those two together. And so that's,

that's ultimately where it came from was, we were the data tool that was slowly building towards BI, they were BI that was slowly building towards being a data tool. And it was like, we could really shortcut a lot of this by by putting these things under one okay interesting so so and that leads on i think to the next topic really which is about the the modern data stack industry in general okay so so you you've certainly been there right from the start really with this um and um i mean just just tell us tell us i suppose give me a potty to kind of couple of minute i suppose history of the modern data stack industry and your position in it. And I suppose to where we are now, just if you could summarize it in a couple of minutes, really.

Where's it come from and where's it got to, you think, at the moment? So my view of the modern data stack is, people have this question, like, what is the modern data stack? My view of it is basically, it's an era, it is like we had the era of big data prior to it, which was basically Hadoop and sort of that ecosystem and the idea of sort of NoSQL databases, we're moving past Oracle, and we're going to dump everything into, you know, Impala or whatever. And that was kind of, quote, unquote, big data. Now, it's like, what products are a part of that? I don't know. It's like, doesn't really matter. The point is, there's a philosophy to it, and an idea of how these things will change the way people think about data. To me, the modern data stack is essentially like a philosophical, the sort of philosophical child to the era of big data.

It was, oh, actually, instead of doing everything in Hadoop and things like that, we could just put stuff in cloud data warehouses. We're going to focus back mostly on things in SQL. It's going to be cloud based data tools. It's a lot of sort of end user facing products. It's not, it's not Oracle and Microsoft and sort of big, you know, enterprise sales in that way. It's much more kind of product led growth style companies. So the kind of joke that I have is I think like the modern data stack, like what are the tools in the modern data stack? They're basically data tools that launched on product hunt. Because I think that's like one one the era that it basically aligns with two

they are companies that tended to be sort of end user facing silicon valley startups um like oracle has a cloud data warehouse but i would not consider it a part of modern data stack kind of because doesn't really fit that philosophy um and so so i think like that's basically the arc of it and and now i think we're sort of starting to get past that a little bit where where we're going to enter the next phase of all this stuff which is ai AI, which we can talk about in a minute. To me, Mo's role in this is it was very early in this phase, kind of pre even the term. It fits into the category though, because it was a cloud-based tool. It was a thing that was focused on working with cloud data warehouses. It was a SQL-based tool. It was selling to data teams that

were kind of like business-focused data teams, not, again, traditional data science teams. In a lot of respects, Mode was early in that. I think that the timing of when we started Mode was two or three years prior to when the market really got going for this stuff that the first customers we had, and especially the first investors we talked to, they were like, who are these analysts you sell to? Like, dude, is anybody writes SQL anymore? It was all kind of some skepticism because we're still in this like big data world. I think by 2016, 2017, we had moved pretty well into the modern data

stack world where things like Fivetrain were very popular. DBT was starting to emerge. Everybody was moving to Redshift and then Snowflake. People were much more comfortable with data in the cloud. All of those things, that transition happened probably the 2015 to 2019. And so Mode preceded that a little bit, but I think was very much in that set of, again, companies that were starting to think about data as a very SQL-based thing, as a cloud-based thing, as a thing that was designed to sort of drive business decisions and those sorts of things, as opposed to kind of the big data we're going to predict the future with massive amounts of data hype that was in The Economist in 2012. Okay. How much do you think the modern data stack was driven by venture capital and venture capital funding? As much as I suppose the companies that were consuming it, the companies

that were funding it, and how it was paid for? I mean, how much was VCs and VC funding a critical part of the MDS, do you think? I think in the second half, a lot, in the first half, not nearly as much. So there was this phase to me of the, you could kind of, I guess, break the modern data stack era if you wanted to. This is like getting really sort of in the weeds, but like in the three rough phases to me, there's this kind of like proto phase of it, which was when the early tools in the space started to pop up. So this was Mode and Looker and Periscope and Chartio and Fivetran and Stitch and those sorts of folks. There, I don't think actually VC money drove a lot of that. There, I think a lot of it was like piggybacking up on Redshift, really. And just like, oh, this cloud data thing is a thing. We need some basic tools

for it. Like Mode was not a, you know, mode is in its course of its history raised like $80 million. That is by today's standards, a relatively paltry sum compared to what a lot of data tools have raised. You know, Looker did not raise some astronomical amount of money. These were things that were like, yes, they were VC backed, but it wasn't, it wasn't like crazy hype. I think a lot of this was the beginnings of a new market and, again, piggybacking on top of some transition away from legacy databases into the redshifts of the world. The second phase, I would say, was pre-pandemic. And this is where things started to really get hot with DBT and all those things. I think that was, to some extent, VC-driven.

Though, again, part of that was just people were starting to really move workloads into the cloud. Snowflake was getting really popular. This was pre-Snowflake IPO. And I think that was like VCs were starting to froth it up some, but it was still a lot of like exploratory. There is real value here. We're not exactly sure how to find it yet. We're starting to build real products the the post pandemic phase the the 20 late 20 basically post snowflake ipo so 2020 to 2022 i think is entirely vc driven that was that was like extremely fluffy that was where they were

like new data startups all the time. They were raising. This was like when the default term sheet was 20 million bucks on 100 million dollars for an A with like a pitch deck and no product. Like this was this was a phase, I think, where the modern data stack was the hype. And so a ton of companies got started building very narrow prompt like solutions to problems that are real, but are pretty small. And so I think that was all VC dollar driven. And there will be real things that come out of it, but that was where we started to see just the astronomical rounds and the billion dollar valuations on companies with barely any revenue. And that I think is a world of a bygone era at this point yeah interesting i mean you think about um i suppose i think there was there were some blogs that you wrote a while ago this is how five trend fell this is how dbt labs fell and and and they were kind of interesting

you know because they were probably the first time that people started to sort of question i suppose the economics behind it or at least maybe the valuations behind it and what that meant for for people I mean I've had um you know Tristan on the show a few times really and talking about I suppose the funding they've raised at dbt labs and you know his rationale was well let's raise it while we can because that then secures the future of the sort of the platform and product going forward but it does also introduce its own complications you know if you've got staff who are now um but maybe their options are not worth as much as they were and and it means you can't raise money you can't get good staff i mean what do you think the effect of

maybe the drive for funding is going to be um and these companies that do have these big valuations um where do you think they go from there really is? Is it a problem, do you think? Or is it not a problem? I mean, what do you think on that? It's some of both. Tristan's perspective, I think, is largely right, which is if the money's free, take it. If you can put several hundred million dollars into your bank account and you have that cushion to ride out a potential downturn, which obviously we ran into, or, or, you know, you can, you can fund yourself for years on top of free money, like by all means go get the free money. And so like, I, I certainly don't think that that is, that is a bad idea. I think the thing that is dangerous is if you go out and get the free money and then you raise $300 million and you spend a hundred a year, and you're not seeing the growth to get there,

that's where I think you find yourself in a real problem. And so I think there will be, to me, different results here that these companies have. There are some that raised at huge valuations on not a whole lot of revenue. And I think those are ones that are going to find themselves in a real bind where they've got to either really, really scale back. Like they simply can't grow into that in two or three years. You're not going to get – if you're at $10 million in revenue and your valuation is $1.5 billion in this market, you're not going to go 10 to 150 in two years. Like that's just not going to happen. And so you're going to have to, something's going to have to give there, whether or not that's really slowing down growth and scaling back what you're spending or if it's selling or whatever.

But a lot of these companies also, they raise money and they haven't spent it that much. And I think in that case, you're mostly fine. You have a little bit of an issue with options kind of being underwater or whatever, you can adjust that. I think that the black mark of a down round may start to go away some where everybody kind of takes it. It's seen as the prudent thing to do is to kind of mark it down because that's what's realistic. And again, if you can go out and raise $300 million, not spend that much of it, and then mark yourself down, your valuation down by 60%, is that a bad thing? No, that's probably smart. Good for you. You probably did it right. And so I think in some ways, the optics of that stop looking as bad. And companies just sort of change their burn profile to fit the times, but they also don't have a lot

of money in the bank. The ones I do think that are problematic are if you're... There's this metric that David Sachs, actually the CEO of Yammer, has talked about some in his post-Yammer day called a burn ratio, which is essentially how much money you spend to how much new revenue you add. And I think that's a pretty good proxy for the companies that will find themselves in trouble versus not. Where again, if you're spending $100 million to add $10 million in ARR, that's a math problem that does not add up. If you're spending $20 million to add $10 million in ARR and you've got $300 million in the bank, okay, fine. You can do that for a long time and build a a really big business it may not

be the business you thought you were going to build two years ago but like that's not the point point is to build something lasting and sustainable and i think if you've got a bunch of money in the bank easier to do that than if you don't okay so do you i mean you talked about um five crown in one of your blogs and dbc labs has said i mean do you do you foresee the way this is going to go, that will be those businesses who have the smaller kind of data products? Or do you think, where do you think they'll go? Do you think it'll be consolidation? Do you think there'll be zombie companies?

Do you think that, where do you think, in a couple of years' time, these products and companies will be? So it kind of depends on the product. I think there will obviously be some different results here my my kind of rough heuristic for that is it's less about the finances of the company probably and more about the problem that you're trying to solve so so take take something like i mean like data catalogs i'm i'm like generally a skeptic of data catalogs. Roughly, I think there'll be some stuff

that comes out of that that's pretty good. But like, that's a fairly new category. Like they've been around in some form for another for a while. It's a fairly new category that the entire idea of building a data catalog that could be a three or $4 billion business is unproven. Like I don't know that you can, or data observability, data observability may be a

better thing where like there are a handful of data observability companies, things like Monte Carlo and Metaplane and Big Eye. The total number of data observability customers in the world is probably south of a thousand. Like those folks probably do not have a thousand customers across the entire set. Can you build a multi-billion dollar business on top of data observability? Maybe, I don't know, but like it is certainly an unproven product space that the entire category could itself be hype from the modern data stack. We're like, that may just not be a sustainable category period to build a super huge business. We don't know.

It may be huge, but it's like you're trying to build a whole new category there that got very frothed up in a couple of years. And so part of the challenge with building one of those companies is you've got to prove that the market exists at all. I think in BI's case, now, obviously, I have some bias here. But in ThoughtSpot's case, ThoughtSpot's also obviously a company that raised a lot of money in a really strong market. And we'll have to continue to... The market is now different for ThoughtSpot as it was for everybody else. However, ThoughtSpot is selling BI. And so that is a market that exists. Nobody's going to question whether or not BI tools can grow into multi-billion dollar

valuation companies because they've been around for a long time as multi-billion dollar companies. Like you can be a multi-billion dollar BI tool. It's been proven multiple times over. And so I think in that case, yes, there's sort of work to be done in all these companies to kind of get back to the valuations and things like that, that, you know, they would have had in frothy times, but it's a proven market and, and it's more about execution and going out and building a great product and selling it. And I think that's a very doable thing. And again, in, in mode and thought spots case, something that I'm, I'm very optimistic about our ability to do in cases like data observability or these categories that are sort of new, you've got to both build the product and prove the market.

And I think that's a hard one. Okay. Okay. I suppose the other risk these businesses have is disruption from something completely new. And I suppose AI and large language models have the potential possibly to be that disruption um that will kind of change the way we do bi and so on i mean maybe just to start off with this this final section maybe i get a little potted summary of of of why you think llms and ai are interesting and and i suppose also why these could potentially be a disruptive element really in the market compared to say just another bi tool coming along so my i mean i have a few different thoughts on this i generally i think they are they can do crazy stuff i by short answer is how they're disruptive is like

it it it can do wild things that don't seem like it should be possible. I went through this phase in early COVID where I was basically two weeks behind everything that was happening. When it first started, I'm like, this will be a few weeks. And then a few weeks later, it's like, no, this is going to be a few months. And a few months later, this is going to be forever. And I kind of was never caught up to the pace that the thing was moving. I think AI, I've kind of had the same pattern with AI where I would be like, ah, this thing is like when ChatGPT came out, it's like, this thing is cool, but it can't do this. And then three weeks later, it's like, oh, wait, actually, somebody figured out it can't do that.

And so part of me is like the pace of the development and just the things it can do that don't seem like they should be possible. It's hard to look at that and think like there isn't some sort of foundational change that happens here. I don't have a there's no sort of grand theory behind that. It's just like, look at the thing. And you're like, wow, you know, this is this is a very novel thing that that seems like you could figure out something to do with this that would be very disruptive. It also is, to me, kind of a fundamentally different type of technology. It's weirdly creative. It can do things that like – it's not just a computer that can run math really fast.

It's a thing that can actually, it is almost human. Like I get people will be like, it's not a human, it's just a text prediction or whatever, but like it has a lot of human characteristics. And so I think like there's a lot of things there where it could very much change the way that, that, you know, we think about interacting with computers or what computers are sort of designed to do.

In terms of BI, you know, I think there are, to me, lots of ways it could change things. I don't think that way will be, and this was sort of the last thing I wrote about on the blog, I don't think that way will be ask it a question and it writes a SQL query for you. I think there's a lot of reasons why that exact thing is hard. I think you may be able to ask it a question and you'll get an answer, but I think that it won't be just like text to SQL bot, that being kind of a popular thing to build. I think instead it'll be the kind of thing where like people figure out entire new paradigms

for how do we interact with data? How do we ask questions? How do we solve problems? Of which LLMs are a part of that problem and figure out some interesting new stuff there but what that exactly looks like i think we're all kind of still figuring out yeah i mean you think about the effort we go to now to create these very elaborate um i suppose elt routines and semantic models and writing sql in a certain way with a certain

dialect and so on you can imagine yeah i think maybe there's an idea that you just have a lump of data and you just point the LLM at it and ask it questions. I mean, there's a lot of reasons why that's a very compelling idea, but it's also practically a lot of reasons why that couldn't happen tomorrow. I mean, what do you think on that as a sort of an idea? Yeah, I mean, I think that's exactly right. I think that it sounds very feasible. And when you do, you build little demo bots of like, look, here's some sample data and I can

ask it a question and oh my God, it can answer them. And this is the future of analytics is me just asking questions. And I don't think that's quite right. I think it's a much harder problem to do that in the real world when you have a schema that is 5,000 Salesforce tables and they're all named crazy,LMs are a piece of the puzzle that can make BI better, but you still need a lot of the things that BI does really well. LLMs don't really work on just writing raw SQL on top of tables. You need a model for them. You need ways to explore it visually. You need the kind of governance that you get out of traditional BI tools that LLMs can't really shortcut that. They can shortcut the drag and drop user interface, but they can't shortcut modeling the data, trying to explain to the machine what ARR means. You have to encode that somewhere. And so while those things may also be like evolve with LLMs as part of the process, I think fundamentally, like the way that BI works is still going to need a lot of the elements of BI. And so it's not so simple as just like, oh, we have a bot that we prompt engineer into writing

SQL queries. You still got to build all these kind of core elements to BI. And so to get back to the point I was making earlier about like every tool eventually realizes they're a BI tool. I think a lot of these LLM bots that write SQL will suddenly realize, oh, we got to build some charting. Oh, we got to build dashboards. Oh, we got to build like scheduling. We got to build governance. We got to build access controls. Those are all the things that make a BI tool a BI tool.

And again, I think, you know, one of the reasons I'm optimistic about the ThoughtSpot acquisition is that's what they have. And they've thought about integrating LLMs into that instead of kind of trying to tack BI onto a chatbot. Okay. So I suppose the final question for me really is, do you think then that whilst LLMs and AI might not be the thing that replaces BI tools, that could in fact make the role of an analyst engineer maybe obsolete? Or certainly the very technical and the specific role of an analyst engineer, is there a chance that could be replaced

by LMs, really? Almost to the point where it becomes forgotten about in the same way that maybe a map-produced engineer would be forgotten about now. Maybe. Alternatively, do you think analytics engineers will be around to stay or it's just a phase we're going through i do i do think they'll be around i don't know that they will quite have the same hype obviously that we've they've sort of been built up over the last few years. But the role to me is still, there's, and LLMs could

actually even make this more important, is you need some way to express, basically like someone who is expressing business logic into data, that is encoding all of these things of what like English terms mean, how do I encode that into data structures that aren't at all sort of fit for answering those questions? I think that is very much a real problem that will persist. Do LLMs make that obsolete?

Probably not. Do they make it where like analytics engineers might express that logic and an LLM writes the kind of semantic model that then translates that into something that is working against the data. Maybe I could see that, but I still think there is like a skill in the kind of data modeling element that, that doesn't make the world go away. But certainly it could be in the same way, all these things could be transformed by LLMs. It certainly could be transformed in some way that is sort of hard to predict.

But it's hard for me to imagine a business user just being like... That's what's hard is if you're a business user that doesn't understand or isn't familiar with the weird schemas in Salesforce and Marketo, it's very hard for you to tell it exactly what to do and to make sure that it's right. You just don't have that access to that domain and i don't know that llms are particularly close to to making the actual encoding of that okay okay we're almost out of time now so just to round things off really ben i mean it's been fantastic speaking to you and we can probably speak for hours but i'm conscious you've probably got things to do and so on um but but how do people find out more about about mode and the acquisition that you're going through now with

with uh so with uh with thought spots um so uh i mean the for for mode and thought spot there i you know we're in the process of doing kind of the final close of the deal once that happens i'm sure there will be uh various things that go out about it. Certainly, if you want to check out Mode or check out ThoughtSpot on both sides of that, the teams are now working together to one, think about what the combined vision of this looks like. We'll have a lot more to say about that in the coming weeks and months. And as well as if you're a ThoughtSpot customer and want to see Mode or Mode customer and want to see ThoughtSpot, we certainly see this as being something where our view is that the combined experience of these two products can be something that's really great. But we want to make sure we're also

protecting the experiences that people bought on either side of it. The way this doesn't work is if we say, ah, at Mode, you have to now do things totally differently. We recognize that people bought Mode for Mode and ThoughtSpot for Thought we want to we want to make this very much an additive thing where where those experiences don't go away so um you know feel free to reach out to anybody at either of those places um you know you can go to the website you can email me and happy to put you in touch um i also for my stuff uh it's it's mostly on substack which is just ben.substack.com fantastic and when will you be at another conference? Will you be at the ADCO and S conference, for example? I will.

I will be at the DBTs conference in San Diego, and I may be at a conference in London in September. There's, I think, big data London or London big data, I think, in middle of September that I may be at. The DBTs coalesce is the next one that I know for sure that I will be at, The Coalesce, DBT's Coalesce is the next one that I know for sure that I will be at, though, in San Diego.

Fantastic. Fantastic. Well, Ben, thank you very much for coming on the show. And it's been great speaking to you. And I look forward to hearing more about Mode and Thoughts Upon the Future.

Awesome. Well, thank you so much for having me. And yeah, there'll be lots more news on that front shortly.