Building a Virtuous Cycle of Analytics in Global Enterprises - with Barry McCardel of Hex

Welcome everyone to the Emerge AI and Business Podcast. Today's guest is Barry McCardel, co-founder and CEO at Hex. Hex is a collaborative data workspace designed to bridge the gap between technical data science and business decision-making. Both for organizations that want to move faster than traditional spreadsheets or rigid BI dashboards allow. Barry joins Emergis' Matthew de Mello to discuss optimizing workflows by shortening the lag loops between asking a question and getting a trusted answer. He breaks down how to move beyond static dashboards to help teams understand the actual reason behind business trends. Our conversation also covers why the true ROI of a data team is found in internal sentiment rather than quantitative logs. Today's episode is sponsored by Hex. Before we begin, a quick note for our executive listeners. Emerge invites enterprise leaders who are driving meaningful AI initiatives to share what they are learning with a peer audience. If you're moving real projects forward and want to be part of that conversation, you can learn more at go.emerge.com/expert. That's go.emerge.com/expert. Now the conversation with Barry.

Barry, welcome to the program. It's a great pleasure having you.

It's great to be here. Thanks for having me on.

Absolutely. Data-driven as a term in the industry for me is one of those where it— for those in the know, it has such a concrete definition and it actually is very important. But from the outside, really sounds like one of those like marketing buzz terms, especially in kind of the landscape we're living in now, where as we, we may have heard, I think it's come up about just about every conference in the majority of podcasts that I've hosted in the last year or so. But that MIT a case study that says, oh God, 95% of these, uh, you know, pilots are, are really not working out. But what's tending to happen, I think, underneath that discussion of data-driven and how to make your organization data-driven is that the tech stack is not getting streamlined, but rather you're just adding more tools on top. And the operative word for that we're hearing more and more from business leaders is this thing called tool creep.

Yeah.

Um, just from your, from where you are, what's your observation of, of what this challenge really is and what are the most pressing challenges enterprises face in becoming truly data-driven?

Yeah. So I think it is a marketing buzzword. There's, there's a handful of these that anyone trying to sell you something is going to say: data-driven, data democratization, seamless. You know, there's all these, these, all these words you're going to get. We all know the same things. It's, it's the taglines that adorn every booth at every conference you go to and marketing websites. And look, I get it. Like there is, it's all coming from, I think, a shared aspirational goal that, that I, I think is earnest. It's not like snake oil. People have this vision of, I want to be data-driven. What's that mean? Well, I want to make decisions knowing they're rooted in data. Well, why do you want that? Because I want to be more confident about those decisions. I want to be more confident they're going to be the right ones. I think most people kind of live their life in fear. I mean, I know it sounds a little extreme, but like, I think at work, really, like a lot of people, it's like, I want to make the right decision. You know, I want to like, companies will say another buzzword, like fail fast, you know, but like most people don't, because if you fail, you get you get fired or you get— people get mad at you. And like, we can get into a whole rabbit hole around like psychological safety or growth mindset or whatever. But fundamentally, I think what a lot of people are looking for for data is a safety net. It's either— it's, it's confirmatory, it's— and it's, you know, well, I want to make this decision and oh, I'm not sure if it's right, but hey, the data backs it up. And so if I go make a presentation and I socialize it to all the key stakeholders and it's got this data behind it, even if it doesn't turn out right, we can always go point back point that data and say. So I don't know, there's a lot of ways to look at this. People want to be data-driven. Okay, well, we're like 20, 30, maybe years into being sold tools to make us data-driven. And what have we gotten? Like, if you kind of look at the bounty of BI from the last 20 years, well, we've gotten a lot of dashboards. And the questions I ask a lot is, do you feel data-driven? Does this— is this making you—

do you have a bunch of screens just kind of yelling at you?

I think the problem is always like dashboards raise more questions than answers. Like I'll just speak for myself as an executive. Like we can get, we can get into my whole backstory, but I've been a builder and user of data tools my whole career. Hex was very selfish. Just the product I always wish I had. I identify as a practitioner on some level, but I'm the CEO of a reasonably large company. And so I'm like a stakeholder, right? And I get dashboards that I see that are of course built in our product. You know, I have this experience every day. I'm like, well, great. This metric is up until the right. Why? Metric is lagging. Why? Like, and, and I think the real data-drivenness doesn't come from the sort of like trivia dashboard stuff, which is like, oh, this, this number was going up. It's really like, can I better understand the predicate reasons for why things are happening and better predict if we do something different, what will happen? I mean, fundamentally, that's what people want. And whether that's this fear-based safety response or whatever, that, that is what people want. And so I think that now, in the year we're recording this on February 13th, 2026. AI is working, agents are working in a lot of places, and a lot of people look at data and they say, okay, well, this is going to be the next big area for that to work. And you said something in your, your prompt to me of like about pilots not working and all this. And it's— I do think like AI is— it clearly has this massive potential to increase quote unquote data-drivenness in that, okay, if I could ask questions, if everyone in my org could ask questions with natural language and get answers, well, that's going to increase the number of people who have access to that, which is great. That's exciting. How do you get that working in practice? Who is doing those things? Who's the arbiter of what is even true? How can you make sure that the decisions you're making are rooted in trusted, accurate analysis and that everyone is aligned around what that— those things are? That's where the rubber is going to meet the road. And if you're asking me the questions, what are the biggest challenges of making enterprises data-driven? There's a thing there that I think is still very, very unsolved.

Yeah, yeah, absolutely. All at the same time, I think we'll put the marketing language aside, we'll put the hype around this aside. If we did not live in this world where people were getting sold this stuff necessarily and they could just focus on what they should be focusing on, how do you think the data practitioner role would change, especially when, just as you were saying with the dashboard problem, your goals can change from, you know, month to month, quarter to quarter. You'll need new dashboards, board. So what should that person be focusing on?

Well, I mean, yeah, I mean, this is a great question, right? The traditionally, how do data teams work? Well, you hire a data team, you post job descriptions that say you're going to drive strategic analyses that will affect the biggest levers on the business. And you'll do advanced analyses. And we looking for someone with 10+ years of Python experience with rigorous routing and statistical method.

Buzzword, buzzword, buzzword.

What do those people wind up doing? Hey, Barry has a question and he needs an answer to it. He wants a dashboard he can look at every day. The CFO wants a dashboard they can look at every day. Okay. Let me go build this dashboard. That takes a long time. You have to go munch the data and pull it together and get it ready. Then you finally share the dashboard and you send it over to the exec and they're like, this is good. It's good. But like, you know, can I get some extra views for this? And this number doesn't quite make sense. So you go do like 2 or 3 more iterations of that. And then they get the dashboard and like, you kind of look, it's like, are they actually opening it? And like, most of them, they never do. But then the ones that do, you know, they do because you're getting a bunch of follow-up questions and you're like, okay, now here, here's my time to shine. Right. Time to do some strategic analysis on the why, why is this happening? Or can you give me a deeper cut of this, whatever, but those workflows suck. Most data practitioners, you know, they're either working in a dash BI dashboard that the company bought, or they're in a sprawl of other crazy stuff. I'm a SQL runner here and a Python thing here. And now I've got a cloud code thing running over here and I'm getting my insight. It's a matplotlib chart I'm sharing by taking a screenshot of a graph and putting it in a PDF of a deck. To attach to an email and that gets sent over to the exec and they look at it and they have more follow-up questions. But the lag loops are so long on this, it's weeks that really everyone just kind of gives up. The exec stops asking questions. They're like, it's going to take 2 or 3 weeks to get some shit back. Why am I going to wait 2 weeks? Might as well be never for trying to move quickly. And the data analyst, data scientist, this isn't really what they want to be doing either. So the whole thing kind of falls apart and sucks. That's the traditional way this works. So what should people be doing instead? Well, like, I think it's obvious, feels obvious, seems obvious that you should have most people in the org can ask natural language questions for the simple stuff. They can ask a question, get an answer that's trusted and rooted. You then have more specialized data scientists and data analysts who are also able to work with AI agents, but they're, they're more off the off-roading, going deeper dive on stuff for the new novel and gnarly questions that that are those deeper dive things, you should be able to have people be able to share, collaborate, and move it back and forth. And if you're in a data— if you're asking what the data practitioners should be doing, they should be in a deep dive and curatorial role where they're going and discovering new things and they're adding that to a corpus of context that can compound and then everyone else can benefit from. And then when those people hit the wall, they're able to get help from the data practitioners. We call this the virtuous cycle. This is how— what our vision is for how this should work. Really. And it just has never really been possible before. I mean, obviously I could show my product here and be like, X makes it possible. But you know, this is the— we can get into that. This is the thing that I think everyone— it's obvious to want this. It's how to actually make it work in the technology and in the way your team's organized. It's like much harder.

So, well, I think if it were— all due respect, I think if it were obvious, uh, that MIT, uh, study would, would not be making all the, all the rounds right now. But I, you know, in your last answer, uh, dichotomy between dashboards and conversations, I think is a debate to be had. And I think for, for what you're, you're seeing at HEX right now, there's a clear winner in that dashboards tend to be this, this place where, oh, we're going to set in stone the things we care about forever, or at least for the next couple of months. And that already might— you might be giving most of those metrics and those views in those, in that, in that perspective that you had when you're, when your business goals were X, Y, Z on February 13th. 2026, a little, that might be actually holding you back from being anything else. Whereas making it more conversational, having a dialogue, a copilot rather than a dashboard. And you fill me in if—

Yeah, that's right. I don't think it's, I, well, I don't think it's either or. I think if you're trying to get clicks, you put up a thing called Dashboards Are Dead. Dashboards are dead. ThoughtSpot. I think they're so funny. In 2020, they put up a blog post, Dashboards Are Dead. Everyone wants to— you can kind of tell people are doing stuff just to get some buzz. Or there was a data debate that last week, you know, that the data stack is broken. Well, yeah, I mean, maybe I— nuance doesn't really market well, so that's why I'm bad at marketing. But, um, yeah, right. Uh, the reality is dashboards are fine. I actually think it's— I have the same opinion about dashboards I have today that I had for a while. They're fine. They're fine.

Yeah, they're like—

but they're just what's happening.

And now they're one tool in a pantheon.

Well, right now they are, but like, this is the whole thing I was saying is like, up until today, they've really been the end-all be-all, right? It's like the only thing you have. Like if you're— the question I ask sometimes is like, if you're a CFO signing the purchase order for a BI tool, what do you think you're buying? You think you're buying dashboards. You're like, great, this is the dashboarding tool, right? That's what people understand data tools to be. It's the end-all be-all dashboards. Said earlier, they raise more questions than answers. Well, now it's great. You have a dashboard that can be the beginning of a question. Those questions can be asked now. You, you don't have to go through that, you know, super long laggy cycle. You can just ask a conversational question. If you start a new conversation question, it can go and use an existing dashboard as context, or maybe it can remix it, or more people could build dashboards that customize to their views, or the agent can customize a dashboard for you depending on what it thinks you're interested in, or maybe it's what you're— what you might want to see that day. So I think the form factors are going to change a lot, like you get into some like weird sort of slightly overwrought, like ontological, like epistemological questions. Like, what are dashboards? What do they mean? Uh, but like really fundamentally it's like, yep, it's a collection of things that are curated to look at together. And I think that'll exist. It'll just become more of a minority of the workflows.

Right. And this kind of leads into a problem I was alluding to at the beginning of the show, which is tool creep. Like you're going to buy something maybe from a vendor, but they, they put all their chips in the dashboard basket and they're going to tell you dashboards will cure cancer, save your life when you might need a tool that's more multifaceted. But even more than that, you might need a partner that is more focused on where data is driving through the organization and how do we get that data to answer the questions you're going to have, not on February 13th, 2026, but on June 30th, 2026, and then February 13th, 2027. Is that a little bit more what the, what the solution is here after we get past the—

yeah.

I think that's right. It's like the dashboard, the, the traditional way is, okay, we're gonna build the CFO dashboard, big sprint to do it. We iterated with the CFO, we built their dashboard. Now it's like under glass indefinitely. The, the question, I think that relationship with these artifacts will just change. It's like, there's not one CFO dashboard. Maybe there is a base one, but then as things are changing, as new metrics emerge, as other things, that, that dashboard can adapt, it can change. Maybe the CFO, what they really want. I mean, there's a, there's a, there's a jobs to be done thing here for people who haven't heard that phrase. It's, it's kind of just like a way, mental model for thinking about products is like, what are the jobs to be done that someone's hiring a product to do, right? Like, you, you, you can say like Zoom is for video chat, but like really the job, the fundamental job to be done is like collaboration. Like you want to be able to collaborate and connect.

Exactly.

What's a dash— what's the job to be done for a dashboard? Is it like what people really wanted is I want to understand my world. Dashboards were the only thing we really had to do that, and they were actually a pretty poor way to accomplish that task. Now You can just think about how to solve that job to be done in a bigger way. And maybe the best way to solve it is I get an email every morning with a natural language summary about what's interesting, what's going on. I can interact with that. And if I want, I can click through to a dashboard that's been built for me that I can visualize. But maybe I don't. Maybe I just want to look at the natural language summary of my email. Maybe I want to reply to the email and have a whole conversation with Hex's agent or, you know, some agent via email about what's going on in my world. And that's actually the way to solve that. And so I, again, like we, we, there's a lot of like category questions and all this. Like I think again, going back to my point is like most people think about data, they think about BI, they think about dashboards. I just think, wow, there's all these new ways to solve that job to be done that aren't grids of tiles with charts, even if that's still useful sometimes. Oh, and it's, I, you can kind of go and do this and I think it's actually like a trap for a lot of data people. It's almost like a uniquely like little trap because it's, um, We're all quantitative and we think we can measure everything and justify everything with numbers. Well, maybe, uh, as it, as it turns out, like data is primarily a service function. You're there to support and enhance other teams. And if you ask what's the ROI of HR, you can sit down and like try to sharpen the pencil. I'm like, well, how many lawsuits would we have every year if we didn't have, or, you know, what's how much more are we able to retain employees? But like really like. At the end of the day, the way you think about the ROI for a function like HR, finance, or other support functions is fundamentally, as a CEO, I will tell you the way I think about it, which is, are the other functions in the company feel like they're operating better because of that? In a way that is very hard to measure, it's impossible to measure. But like, if I go and ask my head of sales or head of engineering, you know, hey, are you getting what you need from HR and finance? Are you, are you able to, are they solving problems for you? Are you hiring and paying, you know, all these things? And if they're like, oh yeah, they're fantastic, I couldn't imagine life without them versus Well, actually they're kind of a pain in the ass and I've been trying to get this role approved for 6 months and I can't, and blah, blah, blah, blah, blah. It's very apparent as a leader, like where, what functions are going to feel, are feeling high ROI versus not just through that sentiment. And in consumer products, you call this NPS or not just consumer, but in some products you call this NPS, Net Promoter Score, which is basically, would you recommend this to a friend? And it's kind of a funny way to look at it, but if you really went and asked like the head of marketing at a company, would you recommend the HR team to the head of sales? It's a funny way to say it, but like, you know, it kind of captures a lot right there. Like, are they actually good partners? And I think about this for data teams. It's like if I went to the head of marketing at your company and I said, would you recommend the data team to the head of sales? If you were catching up with the head of sales and the head of sales was, oh man, we're trying to solve this thing, I'm not sure what to do. Would you, the marketing leader, go, oh, well, you got to go get someone from the data team engaged? Or would you be like, you know, between us, the data team is kind of a waste of time. I don't get a lot from them. That I think is the purest actual test. And you can sit down and try to sharpen your pencil and do an ROI thing and all of that. And again, going back to what I said at the beginning about being data-driven, you kind of want to do that because, you know, we're going to hire these data heads, but we want to make sure we can represent to the board that they're going to be— that it's a worthwhile investment. And here's the ROI of our data investment. And we're going to spend this many millions of dollars setting up a data warehouse, a data ETL, and buying tools. We need to make sure it's all paying off. Well, I know it sounds like a dodge, but at the end of the day, I think you could actually just really run a survey internally. And be like, how's the data team doing? Would you recommend them? You know, would you give up headcount to get another data person? How many headcount would you give up to get another data person? To get how— what if you were to give up one headcount, you know, what would you want from another function? What would you want more of? I think that is the brutally honest way to look at it. And it's not quantitative. And so it won't satisfy someone who needs like an ROI number. But that person, I think, is looking at it all wrong anyway, which is that's not really the way to measure a data team. In the data team output. I'll tell you firsthand, just yesterday we have a customer, Descript, amazing company, great PR company. Like they're using AI to enhance the way people can edit video and audio projects. It's really, really cool. Like if you— I don't know if you use them for your stuff, but a lot of podcasts and a lot of video productions do. It's really cool. The dimes we're paying for it are not like— that's how I would encourage people to look at it. Yeah.

And it sounds like your solution isn't just a product, and I appreciate your candor earlier saying, hey, I'm not just here to peddle, peddle stuff. And hey, if you have success, that's why you're on the show, you know, to talk about where the examples of these capabilities are really making a difference for your clients. But all at the same time, it sounds like your solution is not just a product, but it's also, you know, there's an educative step that you're doing with a lot of the folks that you're working with. Tell us about just what those first steps are like to become truly data-driven.

I literally told someone a couple of weeks ago, a prospect— my sales team hates me— I told someone a couple weeks ago, I said, you should not buy Hex. Interesting.

Yeah, but that's how they know you, they can trust you. That's why you just, you just got a whole bunch of credibility with this podcast.

It'd be a waste of your time and mine.

Yeah.

Yeah. I could sell you a bunch of licenses, but it's not gonna work. Right. Cause they were, they, they're, they're, um, they were looking for a magic pony. They weren't, they weren't willing to think about, okay, well, how are we actually organizing our data team and what we're doing?

Yeah.

They were actually just looking, they were like, we just wanna replace Tableau. Salesforce is ramping up prices. That's the main innovation of Tableau last few years. And, you know, we want a cheaper thing, but we want to be able to one-for-one recreate every dashboard we have in Tableau and in whatever we get. And I was like, you know, there's really a different way to think about this. Look at this. How's your team organized? I was asking a lot of questions. I was encouraging them to look at this a little bit differently, and they were kind of just like, shut up and give us the price quote. You know, I was like, look, I don't think you should buy— I'm not gonna— I'm actually— we're not gonna bid on this. Like, you shouldn't buy Hex. Like, yeah, like, you should probably just renew with Tableau, honestly, and just pay costs.

Like if you're not going to take it that seriously, right?

If your vision of data-drivenness is like, we just want these dashboards, I'm like, yeah, you know, go do that for another year. Keep an eye on what we're doing and maybe it'll, maybe it'll come around. And so I just don't like, I think that there's like a, how you actually view these things. One thing I look at a lot is like, do you actually have a data team? People are like, well, the AI can be the data team now. We have AI data analysts. Like, yeah, no, I think you still want humans. I actually think the humans are more important than ever because How do you trust the analysis? How do you know what's true? Who's encoding the semantic understanding of the data? Who's partnering with the functions to actually think about what they're trying to accomplish and then how to really encode that in a way? Like maybe GPT-7 will be so good that it'll be able to do this all for us. We'll see. But I, I really, we always want to go find who are the people within the org that are going to be thinking critically about how data is partnering. And I say data like the function, but also like data as a sort of like collection of You know, literally how the, how the, all the, all the stuff you're investing in and stack you're building is partnering with the rest of the org. That's really interesting to me. And I, I want to hear that people are thinking about that critically and people care about that. And if not, I'm just skeptical they're going to get value out of Hex or any other data product really. Yeah.

Well, on that note, do you think this, this not just the product, but the philosophy and the, the strategies around being truly data-driven that are really going to work? Is that only a set of industries or a set of kinds of organizations, or are you just saying, ah, well, some of these guys aren't, aren't going to last anyway because they're kind of stuck in the past? Trying to get a sense of where, where is that data stuck in the past?

Yeah, stuck in the past, or look, looking from— I don't know, sometimes people, their vision of being data-driven— here, here's, here's what I would tell you. The— if your vision of data-driven is You're just going to have a lot of dashboards around. You know, there's a better way to do it. Either you're going to figure that out now or figure it out later. Maybe, maybe that's not your biggest problem as a business. Like, I don't know, but I would encourage people to really fundamentally think— maybe this goes back to the first thing we talked about. I would encourage people to really fundamentally think about why you want to be data-driven. Do you want data as a safety blanket? Do you want it as like a confirmatory step? Or are you really trying to better understand your world? Are you really willing to let everyone have access to information. And there's a tension, I think, sometimes with people's relationship with data where they're often— there's a confirmation bias thing a lot of people are looking for. They have a hypothesis and they're just looking to make the data, the chart look a certain way to confirm what they want. Or are you really willing to work from first principles on understanding what is true about your world and what you should do differently and challenge some of your assumptions? I think those are the types of people who get the most value out of data.

Absolutely. But just for maybe where the conversation's going around, like a frequent chorus that we hear on the show is that we want the vast majority of AI strategy, something that almost anybody listening to the show, no matter where they are in the global economy, can take home as an element of advice is definitely don't wait to jump in the pool, but you want to get in the shallow end just because all, you know, the most fanciest solutions not every, you know, factory has to operate the way, the same way Apple does.

Well, yeah, right. Well, this was funny. When I was at Palantir, we would go in and talk to clients. I was there, you know, I started there 2013. I was recruiting 2013. It was like the rise of big data. We'd come into all these customers and we had this brand and we were starting to emerge as a known quantity, uh, from the work we had done in the government. And that was part of trying to bring us into the commercial space. And I'd go and meet with customers and they'd be like, great, we want to do some machine learning.

What?

Okay. What do you want to do machine learning for? Well, We just want, we want to get a machine learning one. It's like, well, why? What do you want to do with it? And they're like, well, we'd like to predict what our inventory in our factory should be, our inventory should be, and how many units we should be making 2 years from now. All right. Do you know what your inventory is today?

Right.

Well, no, of course not. You know, there's no way. That's, that's basically unknowable because, you know, we've got, it's like 3 different ERP systems and we've got our data lake project and that's going to complete in 3 years. And then theoretically be able to do something, whatever. And I was like, well, okay, I have an idea. What if we, what if we start by just helping you see what your inventory is today? Like first let's get that picture and like you can just make better decisions like day to day, just based on like being able to count things. And then, and then, yeah, we may, we'll do a regression on top of it and we go look at some seasonality trends or whatever. And like people didn't, it was funny because people, it was like a hard thing to sell because people wanted the magic pony. They wanted the like—

more than they wanted to know what their inventory was. Right.

That's so boring. And we've been living this way for a long time, but like I could go to the CEO or I could go to the board and tell them like, we've got this magic ML thing we're doing to do predictive analytics. And it's really buzzy. Fundamentally, a lot of the value came from, I think, these kind of like fairly simple things. And I, you know, it sounds funny to say, cause here I am, I've got a product that I build and sell that's, I can tell you how advanced and how magical it is and all these cool things you can do. When I actually look at what a lot of people are doing, I was mentioning this, this, um, CEO, fellow CEO posted this thing about how Hex is changing the way people look at the questions she had in the screenshot or look at the questions most people are asking. They're not like, go build me a predictive model understanding exactly what the world is going to look like in 3 years using a random forest, blah, blah, blah, blah, blah. It's like, what's going on with user trends last week? It's like fairly straightforward questions.

Yeah.

Um, and then those can often ladder up into more sophisticated things like, well, okay, what happened around then? Why did that happen? And what are some predictive drivers of this? But like the, the base rock stuff is something most companies I think actually still have a really hard time with. And as much as the gee whiz stuff where I can show you our agent going and doing something amazing with writing all these different queries and bringing data together and Python and all of that, I think if you kind of look at like what percentage of people in organizations have access to even do the basic stuff, like if you went to a company that spends $2 million a year on a BI tool, right? And you went in and you went desk to desk And you said, can you pull up that tool right now and show me and go build a metric? What's a data question you have? And they told you a question. You said, can you go answer that using this BI tool? What percentage of people actually could? Vanishingly small. It's actually very small. And so there's kind of just like a base 101 set of things that I think if we can help people get better at and do, there's just going to be a lot of value from. And, you know, we come into organizations all the time where you now— and now the dynamic is, well, we want to give natural language self-serve to everyone in New York. Yes, you should want that. That's good. How do you get there? Our, our views is, is like, go give the agent to your advanced data users first who are able to understand it. It'll help, help, help with good context. The agent can be a little wrong and still useful in that those are expert users who are gonna be able to like understand whether they should be trusting those numbers or choosing the right table or not. And that context exhaust that's thrown off by that is what's gonna go enable everyone else. Trying to roll out a natural language self-serve product, like from zero to one is the first thing you're doing with AI as an AI product. Project is like trying to enter a building by jumping into the 18th floor. It's like, theoretically you can do it. It's not the best way to do it.

Helicopter or something.

We set up a trampoline, but there's a high likelihood you might not make it into the building that way, or you may not like the way that you— like, a better way to do it is like go through the front door and walk up the stairs. And like, it takes a little longer, but like, you know, you're going to get to the 18th floor, right? That's the thing again. And it's essentially an analogy, to be honest. I stole from that prior experience I was referencing where we, uh, a guy who works with Andrew would explain this to customers that way at Palantir. He would say, well, trying to go do like build a neural network as your first ML step is like trying to jump through a building by jumping off the 18th floor.

Right. Right.

Let's start by counting things.

Absolutely. But, but you like, we, we say all the time, you don't, you don't want to, you know, just start with a toy application or, you know, so, so then folks will listen to the show. and they'll hear, oh, start with a use case that you read about somewhere. And we write about use cases all the time. Maybe they heard about it at Emerge, but it's still toy thinking in all of the ways you described in that last answer of like, because they're still thinking, oh, I need my ML solution first. We need an ML win. You didn't, you weren't thinking, or, and this is the future that to answer our question that will come about when we see these industries truly data-driven, the hype is past us. We're so far away from the AI hype cycle that, you know, it's 2007 compared to the internet adoption. But everybody's asking, how do we do our inventory better? And then they used an AI solution along the way because everybody knows what— that's, that's what they do. They—

why?

Go ahead.

Why do you, why do you think— if I can, I can turn around. Please. Why do you think AI projects are failing at such a high rate in enterprises?

Ooh, man. I'm actually, this is the second time the, the first week I've been working here for 4 years and, and this is the Second time, first time since I've been working here and it happened twice this week. I get questions on the show and I got recruited for another podcast and they want to ask me about this. So tune in, everybody. I'll give you a taste of this answer. I think a lot of it is, is, is toy mentality. I think a lot of it is misunderstanding. I think it's a lot. A lot of it is the C-suite getting an education on AI by watching their kids make bad book reports and fundamentally miss it. Maybe they were the kind of kids in school who'd like, you know, pay a nerd $20 to write a essay paper and they think that they can kind of do the same thing maybe with their marketing department. And they, they say yes in the meetings, they're kind of half listening and getting back to what success measures they dreamt of in college and were relevant maybe 10, 20 years ago. And they don't— their heads are in the sand because that's not the world or the economy we're going to see in the next 5 to 10 years. And after that, to sum up a much longer explanation of what's going on in that MIT report, I'll put it this way, if 95% failure on an initiative is bad and we should never do it, ooh, I got bad news about business and capitalism. Got real bad news. 99% of businesses fail, so then what is it about? It's about what does the 5% get everybody else? So let's look at the 5, let's look at the 95, and ask ourselves, or even, you're a CEO, I'm part of a, —capitalist organization, I'm part of a business, your profits are, or what you take home, or what you're bringing back to the shareholders, is a tiny fraction of your overall revenue. Are we looking at that ratio the same way we're looking at this MIT study? Yeah, it's an interesting question.

I mean, it's an interesting way to put it. I think what you said on most things fail, most businesses fail, is true. I think you're looking— this survey though is selecting for, or these types of surveys generally are selecting for businesses that are successful. Like you don't wind up getting rolled up in an MIT study if you're like a failing startup. No, that's true. Yeah. Right. Right. Enterprise, that is a going concern. And I don't, I think this, this stat may shock, may scandalize people more who didn't, who haven't been part of a lot of enterprise IT stuff. That's me. Who didn't live through—

you said it fast.

Like, like I did, I was referencing like 10 years ago, like these projects failing. What I see and what I, what I feel is that it's because this is a new technology that has such enormous potential. There's a capability overhang that I think everyone feels because they can see their kid do a book report or they can use it themselves. People are using AI personally now, whether for their personal life or sort of as a prosumer. And I think everyone's seeing that there's this capability overhang. You can see these demos of things getting vibe coded and all of this stuff. The hard part about actually applying something to an enterprise is that the demo is easy, that getting these demos are easy. Getting it actually working is very hard because fundamentally organizations are very averse to change. And especially for really exciting and shiny things like this, people wind up putting these plans together that have unrealistic expectations. They're very waterfall mentality. We saw this a lot at Palantir where we would often be coming in after someone failed on a project where it would be this big bang waterfall where we're going to do a 9-month project to get the data lake first set up, and then it's gonna be another 6 months of this, and then we're gonna implement this, and then we're gonna build all the semantic models up front, and then we're gonna do UAT and then roll out of stage 1 into phase 2 and phase 3. And you know, it's gonna, the, the Gantt chart looks really long and it, these things actually just sort of fundamentally don't work because they're not, they're not an iterative mentality and mindset. And I think the companies that are seeing the most success with AI right now are ones who have a high experimentation mentality. And to go back to the point we were making which is like, well, a lot of things fail. I think it can be okay for things to fail. Yeah. If you are, have a mentality that's like, we're gonna run rapid experiments, not as toys. We're gonna go try to do real things and very quickly validate what's, what can work, what can't. We're gonna encourage people and promote and, you know, recognize people who are highly adaptive and are excited about change and are driving these, this pace of experimentation and actually have a little bit of humility that I always get a little skeevy when I hear about like, um, companies with like Chief AI Officers. I think that's like the, the new Chief Innovation Officer where I'm always like, man, I don't know what authority do you really have? And are you the person that should be, I think that role, the only way that role works is if you are effectively helping set the predicate conditions for a lot of flowers to bloom in the org. And then you are going and looking for the ones that are blooming and making sure they're getting more sun and water. If that, or if that role is, I'm gonna go determine all of the flowers that need to be planted in our perfect little garden boxes and have a whole strategy and plan around it. That is much— I generally think that doesn't work. That winds up being a lot of projects that are going to fail because the central planning version of this, I just think doesn't actually work. So I really encourage people to have this high experimentation and creativity mindset. And I'd rather 1,000 flowers bloom and maybe a couple of interesting ones, even within the AI labs, by the way. Let me just bring this up as one. Yeah, go for it. Claude Code, which is taking the world by storm, was like one dude just experimenting with it. There was a couple of people just experimenting with it. And hacking around. There was no senior executive at Anthropic who's like, okay, and then in fiscal Q2 of 2025, we're going to begin the Claude code project. And then after 3 months, after 3 months of 5 engineers development, we're going to then be on Claude co-work. And then this is the companies that are executing the best, right? This right now are effectively on random walks. They are trying things, they're empowering people, they have a, uh, they're creating the predicate conditions for a lot of innovation. And when they see something that works, they're like, we're going to go put more more light and water on that. You can, I think a lot of organizations can learn from that.

Absolutely. I know we're up on time. I haven't, I'm having so much fun. You're a great guy. I don't, you know, ask the audience. I don't usually say this on mic, but you're a great conversation. And I think you got a really, I mean, I'm sure like you were saying, like, you know, all the sales team hates me, but that those folks always make the best guests.

And I think you're making a very strong, but I hate, yeah. Turns you more linear style than I tend to think.

But I think with this audience, you're saying a lot of things that are getting said in hushed voices and in gossip. And that's my ultimate, as I was saying before, that's my ultimate WPI, worst performance indicator. How much is your organization gossiping, especially to avoid, to tell truths they're afraid to say in a more communal context? That's a big WPI for me. For me overall, and definitely the sign of a not very data-driven organization. I think you might agree. That's a good, good way to tie it back. So, so if your organization has a lot of gossip and you're listening to this show, you're listening to this episode, you got to, you got to turn that ship around. Definitely need a lot more data-driven conversation for a lot of the advice that you've been giving on the show so far. Just, just as a last note, any advice just for leaders out there trying to make heads or tails of this, the MIT study, the the, the, the marketing language and beyond and the tool creep?

Rapid experimentation and learning. The more I think having some humility and subordinating your own ego that you're going to figure out exactly what is going to work and what the future is going to hold. Things are changing so fast. I think fundamentally, like, are you an adapt— are you an organization that excels at adapting, trying new things, having bottoms-up creativity and being able to cultivate and amplify that? Yeah, I think those are the organizations that are going to work the best. The more that you think you're going to have some big top-down strategy around AI or anything else these days, the less credible I think those are. So that's how I would encourage people to maybe think about it.

Absolutely. Barry, thanks for the advice. Thanks for the great conversation. Love to have you back. So much pleasure. So yeah, yeah, have your people tell our people. Absolutely. Thanks so much for being on the show.

My agents will talk to your agents.

And they might not all be human, right?

All right, take it easy.

Wrapping up today's episode, I think there are 3 key takeaways from our conversation with Barry. First, enterprise leaders should measure the ROI of data teams through internal sentiment and partnership quality rather than relying on quantitative logs. Second, organizations can optimize workflows by shortening the lag loops between a business question and a trusted answer, moving away from static dashboards that often raise more questions than they solve. Finally, successful AI adoption requires starting with expert users and fundamental accuracy before attempting to implement complex predictive models. Position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap. For the opportunity to showcase your solution to the executives currently funding and scaling global initiatives, partner with Emerge to reach the decision-makers holding the strategic mandate. Secure your partnership at go.emerge.com/partner. That's go.emerge.com/partner. For further executive-level analysis and to join our network of leaders delivering workflow impact with AI, visit emerge.com. On behalf of the team at Emerge, we'll see you on the next episode.