The State and Future of Data Tooling with Leigh Marie Braswell and Erik Bernhardsson

Hey everybody, it's Eric Thornberg, co-founder, partner of Village Global, a network-driven venture firm, and this is Venture Stories, a podcast covering topics relating to tech and business with world-leading experts. Hey everybody, welcome to another episode of Venture Stories by Village Global. I'm here today joined by 2 very special guests. Leigh Marie and Eric Bernhardson. Eric, Leigh, welcome to the podcast.

Thank you.

Thanks. Excited to be here.

So we're excited today to talk about all things data tooling. I have to say, it's our first episode on the category. And perhaps let's start with introductions. Leigh, why don't you introduce your background and how you got so excited about the space?

Yeah, absolutely. So right now, I'm an investor at Founders Fund. Prior to Founders Fund, I was an early engineer and then product manager at Scale AI. I guess sort of my data journey. I, you know, studied math and machine learning in college. Obviously, when you're in college, everything is very easy in terms of data is very clean. And so you don't really learn how challenging it is to set up a data stack. So when I got to Scale, and was working with our customers and just realizing how challenging it was to, to solve problems with data. There's just so many infrastructure challenges, analytics challenges, and so kind of learning more and more about that, realized how early we are in having tools that can solve a lot of problems in the space. And then at Founders Fund, kept hearing pitches about data platforms, and so wanting to dive in and really discover what that meant and what sort of would be the dominant data platforms going forward.

Awesome.

Well, and I, my background's a little bit different. I studied physics back in Sweden, but did a lot. I always grew up coding. Joined Spotify pretty early and built a music recommendation system there. And despite not really knowing much about machine learning, but that was like back in the Wild West when no one really knew anything about machine learning. So I was at Spotify for 6, 6 and a half years, built a lot of the music recommendation system. Also built a lot of the data platform there. I ended up open sourcing a thing called Luigi, which was like one of the first, I think, open source workflow orchestration systems that gained any sort of significant traction. Left in 2015 to join a company, a very tiny company in the mortgage space called Better, where I took a CTO role. And then I grew that team to about 300 engineers. And then I left earlier this year thinking, you know, if I never start a company, if I don't start a company now, I'm never going to do it. So I just have to do it at some point. And Having spent most of my career in one shape or form in the data space, that's basically right now where I'm like very focused on hacking on a bunch of prototypes and trying to make sense of like, where do you start? Pretty much.

Let's set the scene a little bit or set the stage. When we're talking about the data tooling landscape, what exactly are we talking about? Leigh, could you take a stab?

Yeah, absolutely. So I like to think of it, let's kind of follow the journey of data inside a company. So typically, You get, you get data from lots of different sources, right? You have product analytics data, you have maybe data from, you know, your sales team, marketing data, there's, there's a, there's a ton of data that is in a lot of different sources, and that you want to ultimately use to make business decisions. But there's a lot of steps it needs to go through before it can be particularly useful. So you start with the data, usually the first person that is responsible for wrangling it into somewhat of a clean format as a data engineer. So this person typically has a lot of infrastructure expertise. They're going to need to make sure that latency and storage and all of these sort of infrastructure problems are solved when you're dealing with massive amounts of data. And then once they clean it up a bit, it might go to a variety of different users. So maybe a data analyst who is now looking for patterns in the data or creating reports. A machine learning engineer, a data scientist, someone who's maybe modeling, you know, a particular, you know, let's say for example, using product analytics to model what sorts of behaviors lead to churn. And then the final sort of consumer is probably somebody who's, you know, using it to make a business decision if the model isn't just part of the product itself. So like a product manager. or an executive, or maybe somebody on the sales team. I guess, Eric, anything you kind of want to add to that?

No, I think that's right. Like, and I think you talked about it mostly from the perspective of like, you know, making business decisions based on data, which I think is the most important thing for most companies. And then you kind of mentioned it a little bit too, but like sometimes, you know, the core goal is to actually like let data drive your product, which is a little bit different. Like you're building a music recommendation system is like one example, right? And then, you know, you might have like a, you know, a machine learning team or whatever So that's maybe a couple more roles to add to that roster.

And could you maybe give a bit of an overview of how this has changed over time? Like if they were sort of like different waves of these kinds of companies?

Yeah. I mean, I think when I look at it, I started working at Spotify in 2008 and that was like the golden days of like Hadoop, which was this like terrible distributed system to do, to process data. And in a way it was like, you know, it still was like extremely valuable because it was like the really like the only thing that worked back then. And then I think in order to write like a simple query, like, you know, how many users churned yesterday, you basically had to write initially a whole like Java code, you know, 200 lines of Java code. And then eventually like people started writing, building wrappers in Python, whatever. But so that's like kind of back in the day how it started. And then of course, like if you had an insane budget, you can always go to Oracle and buy like a cluster. And then over time, I feel like, you know, like there's been a shift in the last like 5, 10 years, like towards more like data warehouses so that this shifted to more like SQL, which has opened like a whole new consumer base. A lot of product analysts and other types of people are now running queries much easier. I think, you know, putting things in the cloud has helped a lot with that. And I think, you know, we're still sort of, in my opinion, not even halfway, like probably just like a quarter through. that sort of transition towards, like, you know, commoditization of tools and building things that are easy to use. But it's like a whole new world already. Like, it's just so much easier to work with data. Still pretty hard, but like so much easier in the last 10 years, which is exciting.

You know, it's kind of challenging for me to even imagine a world where, you know, you didn't have something like a Snowflake or Fivetran or something like that.

Yeah, it was terrible for so many years. It was like the dark ages, right? And like, You know, companies like Spotify, we like valued data so much. It was like so important to us, uh, that we did it anyway, but it was extremely painful.

Yeah. Lee, we, we just gave a, um, sort of general market map for, for people who are really outside the space and, and, and, and maybe still slightly confused. Can we go deeper on some of the spaces? And, and by deeper, I mean maybe give like an explain it like I'm 5 version of, of some of these subcategories.

Can I just add that, like, it's not just people that are outside the space that are confused. Like, I'm inside the space kind of, and I'm like insanely confused. There's like that, you know, when I look at the market landscape, there's about 20,000 products. Whoa.

Yeah, no, it's, it's incredible. And I mean, part, part of my motivation for, um, I, you know, I wrote, wrote a blog post where I kind of divided all of these tools in, into major categories based on end user, because otherwise it's just very, very challenging to orient yourself. So I guess, you know, the way that I like to think about it is for each particular persona on a data team, whether it's a data engineer, a data analyst, a machine learning engineer, or, you know, somebody as far downstream as a product manager, like what tools make their jobs much easier? And so, you know, if you're dealing— if you're a data engineer, you just have so many different infrastructure challenges. And there's, there's so many different categories, like, for example, maybe You want to be able to serve real-time predictions, or like have data operate on data streams. There have been multiple products that have popped up to help with that. Maybe you want to make sure that there's no inconsistencies in your data. So there's been this sort of new emergence of the data quality category, these tools that can automatically sense when your data is bad in some way, whether it's inconsistent or The distribution is changing. So I don't know if we'll have time to go through all the kind of different, I would say, subcategories, but I like to think of it in terms of people and the problems that they have of getting the data from the format that they receive it to something a bit more readable to give to the next sort of person.

Yeah. And so if you imagine in the next few years how the space might evolve, Do you imagine evolving just as new positions evolve or are created, or how do you see that? Yeah.

I mean, and I've talked about this with Eric quite a bit, but ultimately, hopefully the stack will consolidate quite a bit. So right now, I think people are taking a ton of different tools off the shelf to these very specialized point solutions, but over time, imagine the, the sort of most dominant tools will begin to become more useful for their persona or potentially expand outwards. And so if you think about some attributes of tools that might lead to this, which of these tools are very collaborative, right? Like, which of these tools involve multiple personas, maybe even a technical and a non-technical persona? I think a good example here is a data catalog. which is basically an interface for you to see, you know, where is your data coming from, the lineage of it, who is, you know, using, using the data the most. I think that's a really exciting category. And then, you know, I think you'll also see tools that, as Eric was saying, that are abstracting away a lot of sort of the, the developer ops or other types of specialized engineering that make their end users much more productive. Those will continue to sort of dominate as one of the more successful data platforms. So that's some sort of predictions that I think will end up happening.

Yeah, I mean, I think kind of like my sort of point of view on that is like there's so many like specialized roles in the data landscape today. And maybe this is like a terrible analogy, but I almost feel like it's like, you know, you're like in the late 1800s running like a textile mill or whatever. And then like electricity comes and like suddenly you're like, oh my God, I'm going to have to, you know, build my own generator. I'm going to have to like make power lines. I'm going to have to like do all the switches and that kind of stuff. So you have to hire all these like experts for that. And then like over time, I feel like, you know, with the, you know, with new technologies, like those things tend to commoditize. Like, you know, suddenly there's a power company, they could just run lines to your textile mill. And I feel like it's a little bit similar thing with, with sort of, you know, developers. Like, you know, the textile mill like person, you know, who runs it is like, actually, I shouldn't be in the business of doing this. Like, I'm an expert at like making fabric. And I think there's a little bit of a similar thing where like tech companies, you know, over time that, you know, what is the value of developers? The value of developers is to take business logic or, you know, business logic and express it in code. But now, because there's so much like complexity around working with the data, it's gotten a lot less, but there's still a lot of complexity. You know, they get sucked into so much like infrastructure and configuration and resource management and provisioning. And that to me strikes me as a very huge opportunity where I think there's going to be both sort of despecialization of roles, but also sort of a consolidation of the stack into fewer players and better tools that take care of larger parts of the stack, if that makes sense. I don't know what you think about that, Leigh.

No, I mean, that totally makes sense to me. It's interesting. I think back to my time at SCaLE. And there's probably 3 engineers, like the infrastructure lead, a particular engineer who understood how our ML stack worked, that if they weren't there for whatever reason on a day when something went down, my job that day was to figure out where they were and to get them to help us figure out how to get things up again. So it's just like there's this still, even with all these new tools and how the ecosystem has evolved, Still very— you have to be an expert in certain types of tools to make sort of sense of large amounts of data and have systems up and running.

Let's zoom back in. Why are there so many different data roles? What are the resulting problems and benefits?

My theory is that it's very hard still to work with data. And so you kind of need an expert on the infrastructure side. You need an expert doing machine learning. You need an expert— you know, understanding business need and translating that to SQL and packaging it so business people can understand it. You need people building the pipelines. And so, you know, and all of those things are kind of hard to do. And so, you know, it's rare to come across people that know like, you know, 2 or 3 of those things. It's sort of exponentially harder and harder. And then so you end up, you know, kind of instead opting for sort of specialization where you hire one person for each of those things. I think, you know, on the landscape side, this is sort of maybe a separate thing which is happening, which is that I think tools right now in many cases are a result of sort of commercialization of open source projects mainly. And they tend to be a little bit more like sort of, you know, widget-type tools. They sort of, you know, come out of like a need to solve like one particular kind of subproblem. And I think that also drives a little bit of fragmentation on the tool side, uh, which doesn't help with, you know, when you have already a problem with specialization on the, on the, uh, talent side. So there's sort of 2 things that I think are both sort of reinforcing this, I think.

Yeah, I mean, as an investor, really excited to, to see tools that, that are less specialized, that, that maybe can involve multiple users in the sort of, you know, data lifecycle. And, you know, thinking back to my time as a product manager, an unintended side effect of a lot of this specialization is just how challenging handoffs become. So, you know, scoping something that involved like a new data source, a new model, an A/B test, maybe you're also using machine learning, or you're having to like write some report. I mean, it could take weeks months, even at a startup where, you know, I would say velocity is pretty high. So, you know, a lot of specialization means a lot of coordination, and I'm really excited to hopefully, with better tooling, kind of avoid some of it.

Why is the startup ecosystem and data stack so fragmented and expansive?

Yeah, I mean, I think I mentioned some of it. Like, one thesis I have is that a lot of it is sort of commercialization of open source project. I think maybe another thing, like, I don't know, this is like a sort of broad accusation that maybe it's harder to prove and maybe people will disagree with, but like, I feel like a lot of companies I look at like are not maybe trying to answer, like, you know, they don't have an ethos. Like, what are we actually solving for? It feels like more like they're like, we have this product and we want to make it better by adding more features. But what's the actual goal? Is it to enable more productive teams? Is it to reduce downtime? I don't know. I think there's a little bit too many widgets and very specialized features in the space.

Yeah. It's interesting because teams, I think, are getting more and more used to just buying a lot of different, even closed-source solutions. You look at not data, but let's say observability or monitoring. The average company probably has multiple solutions that they're using for that. I was kind of doing a deep dive into, you know, Datadog and New Relic and a lot of these new startup competitors, and people are totally fine paying for 10 different solutions. So definitely companies are used to kind of just adding more and more and more, which, which doesn't really help here.

I think an interesting thing is like when I look at fragmentation of a space, Part of me actually gets excited. I spent 6 years in the mortgage space, and like mortgages are like notoriously fragmented. There's like 10,000 mortgage lenders in America. And part of it is like, so far there hasn't existed any scalability advantages. And so maybe it's a little bit different, but like, but what I, what I think it's like a sign of often when I see fragmentation is that demand is there, but supply is not very good. And, and so I think that applies to both mortgage companies, but also data landscape very clearly is that there is very clearly demand for better tools. And the reason why we haven't seen, you know, sort of winner-takes-all effects quite yet is that none of the tools are like, you know, often in many of the subsegments. It's not true in maybe all segments. Like, there are certain tools that are like getting, you know, increasing traction. But, but in many subsegments, you don't really see that. And that's why I think you see a lot of fragmentation too, is that, you know, none of these tools are like that great. Like, they're like good, but not like amazing.

Yeah. Why don't we zoom back out again and talk about what are some of the most exciting categories of startups here, or maybe some of the biggest pain points felt across the stack?

Sorry. I mean, I think it's like all over. Like, I think at a meta level, it's like, you know, sort of consolidation of the stack into like fewer layers. I mean, getting more specific, I think things that I'm excited about is You know, how do we get rid of this like infrastructure garbage? Like all the sort of, you know, pain that data teams are dealing with, you know, around Kubernetes and Terraform and AWS and Docker and that kind of stuff is something I'm very interested in. I'm very interested in workflow scheduling, like automation of like complicated data tools or data pipelines. I think there's a lot of stuff around like machine learning where like right now it's like you kind of have to build everything yourself, but I feel like there might be more like off-the-shelf type tools that could like be like good enough for a lot of use cases. Like one of the things that I, I've been thinking a lot about is like LTV prediction. Like a lot of companies have problems around like, you know, just putting in a value of like an LTV, a lifetime value on a customer based on, you know, maybe data they've observed so far and sort of estimating the future LTV. That to me strikes me as a problem. Like you could probably like kind of solve that as a black box. reasonably well in a way that like most companies could just like buy a service for it. But I think there's other things like that too. I've been sort of interested in this like idea of like a probabilistic knowledge graph. In a lot of cases, you can actually, you know, do machine learning on sort of, you know, the structured data pretty well and, you know, get a lot of value out of it, you know, multimodal data could be images or text or whatever. And I think, you know, without really knowing what, you know, the machine learning under the hood. And I think There's some exciting things there, I think, longer term, where like I see sort of machine learning getting a little bit more accessible. Of course, if you really, really want to do like machine learning well, you're always going to have to build your own algorithms. But I think for a lot of people, you might not need to in the long run.

Yeah, with ML in particular, it seems like right now you have 2 choices. It's either you do absolutely everything yourself, or, you know, integrating a lot of different point solutions and putting them together yourself, or you have these very, very general AutoML services. And so I think, yeah, Eric's totally right talking about sort of what's the middle ground. Can you figure out use cases that are broadly useful for people and maybe not even have to have an ML engineer? You can have someone semi-technical that can set up something very general for predicting LTV, right? So that's, that's definitely super exciting. I think I talked about these a bit earlier, but, you know, I'm really excited about tools that are collaborative, that foster interaction between technical and non-technical users. Really interested also in kind of giving, you know, specialized engineering superpowers to people that might not be specialized in that form. So super excited about sort of ML prototyping tools where you take an ML engineer, maybe they don't know anything about front end, but you give them some sort of way to create an early version of an ML app that they can share with others or even potentially deploy themselves. And so you're taking out the need for them to have a specialized infrastructure engineer or front-end engineer working alongside them. So yeah, super excited about stuff like that.

Yeah, I think the point about collaboration tools, I think to me is quite interesting. I think right now that this is sort of so many handoff points. And what I often see at teams that I'm definitely guilty of this is an engineer or a data scientist will generate a chart and then screenshot that and put it in a Google Slide document and then annotate that and send it to some third party who will add some comments in it. and then email it around. And I don't know, there's this sort of almost like an assembly line of information and insights and owning the feedback loop, like how sort of information disseminates throughout the organization that I think is quite interesting. It feels like, how do you package insights and how do you share it and how do you do that end-to-end from the data scientist without having to change the format like 3 times? I think there's some interesting opportunities there.

Yeah, and also, I think of a blog post that you wrote earlier about sort of how do you build and grow a data organization. Not only those sorts of problems, but also what happens when, you know, you're someone who is non-technical, maybe you're on the operations team, and you need data to make some sort of decision, but there's no sort of resourcing for you from the data team. You're just going to try to do it yourself, and that leads to also a lot of potential you know, bugs or things that are built that are very brittle and ultimately wasted time.

For sure.

What about more broadly, what tips and insights for people building data tooling startups?

Yeah, I think one thing that I feel like a little bit sometimes happening is like people sort of index a lot to the existing tools and existing roles and existing organization structures. And and try to optimize for that. Like you know, for instance, right now I feel like there's a lot of startups, you know, trying to make you know data run on Kubernetes, and I'm like, I don't know, like what's the underlying problem? Is Kubernetes the right tool? Like I don't know, maybe it's not, maybe it is. Like I like and and I think and and maybe that's like a bad example, and but but I think to me it's like worth sort of going like a couple of like wise. below that? Like, what is the underlying problem to actually solve? And maybe the right— maybe you need to rethink how we do these technologies. And I think that's going to be interesting to see how all these companies play out. Because if you're sort of building a startup on top of an existing platform and then that platform goes away— I'm not specifically talking about Kubernetes, there are many other examples too. How are you going to react to that? I think the other thing is sort of thinking about this defragmentation of the landscape and consolidation of layers. How do you position yourselves in a world where there's going to be probably increasingly winner-takes-all effects and more full-stack services? How do you do? Do you partner with other companies? Do you try to buy them? I don't know. I think that's going to be an interesting sort of story to play out in the next few years? Like, what's going to happen with all— with that landscape?

Yeah, I think thinking about, you know, if it— no matter if you're starting a startup or you're growing your startup, what is your strategy, right? Like, I, I do agree that there will be a smaller number of startups than, than there are now in the, in the long run in these sort of areas. Um, so you— what are you as one of these startups? What is your moat? What is your plan to— or what is your plan to develop a moat? Like, is it that your tech is 10x better than everybody else's? Is it that you're going to create some brand for yourself that nobody else can match? Are you going about counterpositioning in the right way against incumbents that just fundamentally can't switch to the pricing model that you've developed, for example? So just really kind of answering that question early on and making it very clear to people that you're hiring or VCs or whoever. what your moat really is, I think is helpful.

Yeah, totally. Absolutely. 100% agree.

Maybe to head towards closing here, what are other predictions we have for the future? Or perhaps just because you spoke about it a little bit, what are other misconceptions people have about the space when they're investing in it or thinking about where it's going?

Ali, I'd love to hear your thoughts on the investing angle of it.

Yeah, I don't know if it's necessarily consensus among investors that there will be a lot of consolidation. I guess kind of, at least if you look at the number of startups and the amount of funding in some of these sort of point solution areas, you know, it doesn't always make sense. I think when it— when you're thinking about, you know, all the sort of successful elements to growing a startup, I mean, one of them is timing. And so I'm always thinking about, like, you know, is it the right time for a really sophisticated point solution in a certain area? And yeah, I guess that's one sort of thing that I'm always noodling on.

Yeah, yeah. I might start with my most controversial predictions that, you know, if you listen to this podcast 10 years into the future, Feel free to reach out to me if I'm wrong about this. Like, one of the things is like, I think there's been a lot of success with like open source tools, but like my feeling is actually we're kind of past the peak of open source as a business model. Like, I think it made a lot of sense like 5, 10 years ago when there wasn't really a good distribution method for software. And the main one that existed was like, here's an open source thing, install it yourself. And, you know, so we have a generation of like Confluent and Mongo and a few other ones like doing really, really well. But I'm not necessarily convinced that, you know, if you're a VC and, you know, you find out about the commercialization of some random open source package and, you know, it has like 10,000 stars on GitHub, that that's necessarily like, you know, a strong investment thesis at this point. Like, that is not necessarily going to be successful. I think the world is a different place right now. And for a lot of things, I actually think it makes more sense to build it in a proprietary way. So that's like one thing I think. Another like semi-controversial thing is like, I, I, I've been a big fan of SQL, like, and I've always like kind of, I wrote a long rant blog post once about SQL, like how much I love it and how much every other garbage query language should go away. On the other hand though, like, I wonder if it's like, you know, the pendulum has swung a little bit too far. Like, I feel like 10 years ago, like people was, you know, maybe 15 years ago, people were trying to do everything on the database. Like they're trying to, you know, use triggers and stored procedures and all kinds of stuff. And it turned out that putting business logic in SQL maybe isn't a great idea always. And so I think in the long run, code is a better way to express business logic. So I predict a little bit of a trend back towards code. And then another semi-controversial, or maybe not, I don't know, maybe this is like mainstream now. But, but I, I think, I think there's been a lot of hype around like AI and machine learning, and, and I think it's fine, right? Like there are some like really cool things you can do, but, but I think relatively speaking, I think there's more room for like sort of boring data science tasks. And that's not to sort of diminish their value. I actually find them like, you know, personally like very interesting, but like things like, you know, building pipelines or you know, building metrics or dashboards or, or, or just doing, you know, product analytics. I think in a way, when I look at like the sort of legacy companies, you know, if you go to like a big bank or, you know, they often have a bunch of like sort of pet AI projects going on, but like 9 times out of 10, I don't think they know where in that conversion flow users are experiencing the most friction and dropping out. Like, I think there's so much like basic product analytics And you can almost like tell when you're using some of these products, right? Like trying to register a bank account online, it's just like a terrible experience. And so to me, I, I think there's just so much basic product analytics that the world is still not doing, uh, that I, I think, you know, that's where I see a little bit more of growth and data demand in the next 10 years. And then maybe finally, like, you know, more of like a positive prediction is, is I think we're like so early in cloud adoption and in data adoption. Like, I think, I think, you know, I look at the cloud and, you know, we've been, you know, using it for 10 years or something like that. And I think it's amazing how many tools we have now and how much more productive I am as an engineer. But I really think like that the tools we have are still pretty early stage. Like they're still kind of hard to use and they— I don't know if they are, they make me as productive as I should be. They definitely make me, make it possible for me to build very scalable, very high-performing things, but it takes a lot of time and a lot of frustration. So I think, and then I look at like just how many companies are still running things on-prem and how many tools people are still building in-house that really should just be factored out into SaaS products. I think this is probably 10x upside for the cloud in terms of market size from here.

Oh yeah, no, I completely agree with that. It's really interesting seeing sort of— I guess I'll go in order, but sort of your point about how ML is definitely only one tool in the toolbox. And sometimes it's made out to be like it's going to be the tool. And you see tons and tons of startups doing sort of small things in the ML stack. I do think that there are probably only— I don't know, I think there are only a few sort of core ML categories that really make sense, given how early people are in their ML adoption, the really basic, like, infrastructure stuff versus some of the fancier stuff. And then, you know, you're seeing how early people are in their cloud journeys. So a lot of the current tools that they're using today, it's probably a bit out of necessity. Like, for example, you know, you hear about so many different problems with the cloud providers' offerings. And I imagine a lot of successful startups over the next few years will be looking at this particular solution and this cloud provider. It's really incredible the run rate of some of these solutions. And can you spin that out into a separate startup, address some of the core developer pain points, you'll probably have a very successful company. So I think it'll be really cool seeing over the next years more developer-friendly, cheaper, sophisticated ways of solving core infrastructure problems.

Yeah, and just to expand on the point about machine learning and AI versus more boring stuff, because I think you said some really insightful things, Leigh, is I almost feel like, in a way, a lot of it comes back to like, you know, org changes needed and, and like ways we, we operate. And, and like, you know, think about it, like let's say you're like a chief marketing officer at like a large financial institution, like data, like product analytics is kind of a threat to you actually. Like, because, because your job for the last 20 years has been to like have all the answers and like work with agencies and all kinds of stuff. And, and whereas like, I think I think re-changing, you know, changing the, you know, to different model of how we build consumer products to be data-driven suddenly means you don't know anymore, have the answers. And I think from that point of view, like, I think data, like using data to make business decisions or building consumer products or whatever, it needs, it necessitates a bigger change in org structures and workflows. than maybe saying, oh, actually we should use machine learning to power our chatbots. And so I think for that reason, there's been a lot of buzz, I think, in legacy companies to adopt AI or machine learning, whatever, because in a way it's easier for them to sort of conceive of that. But I think there's a much, much bigger upside in the more painful transformation of how do you go from sort of a pre-data decision-making process to like, you know, post-data decision-making process. And that I think is going to be ultimately extremely good for consumers and, you know, anyone using these products.

Yeah, I guess so far we have talked a lot about tooling and infrastructure, but it is also a people problem. Like, it is, it is a behavior shift to go from, you know, I'm an executive and I'm just making these decisions because I know, versus I'm an executive and I'm making these decisions because I have a report that shows that this is the right decision. And also, I'm going to measure like it after the fact to make sure it was the right decision.

Yeah.

And, and yet, I was talking the other day about, you know, why, why aren't there very many product manager tools, like tools that tell you what to prioritize next? And I think a lot of it is a behavior thing. Like, do you as an executive or product manager, whoever's making a prioritization decision, You know, if you had a perfect tool that could surface exactly the decision you should be making, would you use it? And yeah, I think that educating people is a lot of the problem there.

Yeah, and I think it's an interesting way to phrase that question because I think people, if you ask them, will say yes, but like in secret they'll probably say no, right? Like they may not want that tool.

Right, or like you'll have it and yeah, you won't understand how to use it properly, or Yeah, I think it's very interesting that there hasn't been, outside of like the project management tooling— I mean, you see some startups in the space, but there's currently no magical product oracle tool where you like put in all of your, your data or like your KPIs and then it outputs, you know, what you should do next.

Yeah, Leigh, Eric, this was a fascinating deep dive into data tooling. For people who want to learn more, about the space or follow your work individually, where might you point them? Leigh, let's start with you.

Definitely follow me on Twitter. And yeah, I kind of post— I've been starting to blog a bit. So post everything to there. And then yeah, my email is also on my Twitter.

Yeah, I have a blog, which I occasionally write things to. It's erikbern.com, E-R-I-K-B-E-R-N.com. And And on Twitter, I'm Bernhardsson, just my last name, which, you know, occasionally I write a lot of random thoughts about data tooling, not all of which are correct, but, but, you know, sometimes it can be a source of heated debate on Twitter.

Awesome. Leigh, Eric, thanks so much for coming on the podcast. It's been a great episode.

Thank you so much for hosting us.

If you're an early-stage entrepreneur, we'd love to hear from you. Check us out at villageglobal.vc.