Value Driven Data Science

[00:00:00] Dr Genevieve Hayes: Hello, and welcome to Value Driven Data Science, where data professionals become strategic experts. I'm Dr. Genevieve Hayes, and today I'm joined by Nicholas Kelly. Nick is the co-founder and chief AI architect of Delivering Data Analytics, a consultancy that helps organizations turn data, BI, analytics, and AI into confident decisions people actually act on.
[00:00:29] He is also the author of Delivering Data Analytics, How to Interpret Data, and the recently released The AI-Driven Data Team, and previously appeared as a guest in episodes 70 and 71. In this episode, we'll discuss how data professionals can evolve their careers for the AI era so they can take charge of what their roles become instead of fearing what might happen to them next.
[00:00:58] Nick, welcome to the show.
[00:01:00] Nicholas Kelly: Thank you, Dr. Hayes, and thank you for the intro. It is a real pleasure to be back
[00:01:04] Dr Genevieve Hayes: When talking to data professionals about AI, one thing I've noticed is that attitudes tend to fall into one of two extremes: AI boom or AI doom. At one extreme are those who are excited about the new opportunities AI will create, the chance to do more interesting work or finally launch the business they've always dreamed about.
[00:01:27] But at the other end are those who lie awake at night wondering if their job will still exist in 12 months' time and are saving every dollar they can for when the job apocalypse finally arrives. But regardless of where they fall on that spectrum, few of the data professionals I've spoken to have a clear sense of what to do next.
[00:01:49] What skills should they be building today to be ready for tomorrow? Nick, this is a question you've recently been helping data teams to answer through training workshops and your new book, The AI-Driven Data Team. However, before we go down that path, 12 months ago, when you last appeared on this podcast, your focus was on delivering data analytics and insights to stakeholders primarily through BI dashboards.
[00:02:20] A lot has clearly changed for you since then. So to begin with, can you give our listeners an update on the pivot you've taken in your career since you were last on this show, and what motivated it?
[00:02:35] Nicholas Kelly: Yeah, it's an interesting journey, Dr. Hayes. I think what precipitated it was we started to notice our clients not request many dashboards. That was probably the first thing, and we had one particular client who had blankly said, "Hey, Nick, ChatGPT can do 50% of what you do right now."
[00:02:55] And he was like, "Okay, we're not gonna need your services anymore." And that was probably two years ago. So the writing was on the wall, and so really I had a choice at that point. Probably being a consultant, we see this stuff a little bit sooner than an employee for a company would see 'cause we're sort of on the front end of it.
[00:03:14] The things that could be cut kind of more discretionary spending, so we see that first. So I would say it was market-driven for us. We saw that, and so we had a decision. Do we take an honest look at ourselves and say, "Okay, things are gonna change, and we can either react or be proactive"? And at the time I decided I wanna be proactive.
[00:03:36] I don't wanna just figure this out after the fact and be looking for scraps basically try and do this on my terms. And so that was the idea. And to do that, I was like I need to research this properly." And Dr. Hayes, you would know this well, you've been doing AI for years, and when I tell people, "Look I built my first neural net in 2001, I've been doing AI since then," people are like, "Really?
[00:03:59] What? AI existed back then?" It's yeah it did. Most people's perception of it right now is, ChatGPT and Claude and all those things, and that's all good. It's certainly that, but it's been around a lot longer. And In some ways, it was a large pivot from people's perceptions.
[00:04:15] Oh, Nick does dashboards, and I certainly do that. But dashboards are, and always have been, an interface to deliver and enhance and enable action. That's always the way I've viewed it, and for many years, it was one of the optimal ways to do that. Now that's changed. So I would say the front end has changed.
[00:04:35] The back end's the same, in more or less terms you could say it's largely the same. But the engine for delivering that insight is now, instead of dashboards, it's pivoted into Teams, into text messages into WhatsApp, Viber. The last mile has completely changed with BI.
[00:04:54] And we still do occasional dashboards, but we mostly do the getting to the last mile with insight. That's been, like, the major pivot, I think, is if you were to tangibly look at it, it's that. That's what's changed, and we had to get ahead of that. So there was, like, a bunch of things we did. I talked to my publisher.
[00:05:11] He's like, "Let's do this." We have to have the public perception that we're shifting into AI, even though we're not really. We've been doing it all the time. But you have to have that. You have to manage that perception. And then we also bought a GPU rack, which was crazy very expensive,
[00:05:27] But to have a seat at the table, that was one of the things we said we needed to do. So that was like our journey it was market-driven, but also wanted to have that public perception of change as well and apply that change to ourselves and also make sure we're just getting ahead of it.
[00:05:43] Dr Genevieve Hayes: For me, I started my career before Power BI and Tableau became, standard issue for data analysts. So back then, the data analysts in my team, they were just effectively hand-coding the equivalent of dashboards. So hand-coding these reports in SAS that'd ultimately be output into Excel or a PDF report.
[00:06:06] So to me, dashboards were the next evolutionary step beyond those hand-coded reports, and it doesn't actually bother me, the idea that the dashboard tools are going to ultimately cease to exist. It's just, oh, okay, so we've gone from one set of tools to another and to another again. But there's still that need for the information that was conveyed by those tools.
[00:06:37] That hasn't ceased to exist. It's just how that information's conveyed
[00:06:43] Nicholas Kelly: 100%. That's really, that's the shift is that the tools are changing. We still need the data. We still need people to act on that information. So none of that's changed. But yeah, totally. In some ways, it's like the latest tool, and it's more than that as well, but it is like the latest tool.
[00:07:03] It still requires people to adopt, to use. You could argue we haven't solved BI adoption, you can't say it's fully adopted in most organizations. When we get involved it's usually like you would say, I'm just picking numbers here, Dr. Hayes, but let's just say 20 to 30% adoption.
[00:07:22] People who should be using dashboards aren't, actually using dashboards. It'd be great to get that to 70 or 80%. So we haven't solved that. But that's in part because of the tool. The cause of that problem is the tool. With this advent of AI now, like we can get much higher adoption, not of the tool, but of the actual insight, people acting on it
[00:07:41] Dr Genevieve Hayes: Clearly the BI dashboard is dying. Do you think There will ever be a point where people completely stop using Power BI and Tableau, and they completely get superseded by AI? Or will they always linger on?
[00:07:54] Nicholas Kelly: That's an interesting one. I can't answer that in what it looks like in the future. But there's definitely a trend for organizations that currently do have software teams to upskill those software teams to build their own tools.
[00:08:05] We're seeing that. 100% seeing that. And in some cases, while they're not getting rid of their BI tools, you can see that trend where, "Hey, we built an app, and now it's, surfacing those insights into teams." If you look down maybe 12 months, do you need the dashboard? If you wanna put it into PowerPoint and stuff like that, maybe.
[00:08:24] But ultimately, you could just build a tool to be able to do all that. In part that's driven by the capabilities of the organizations. Do they have AI engineers? Are people capable of building the tools that will replace that? But I do see a trend. Probably not gone altogether, but certainly Power BI will have the capability to follow that evolution.
[00:08:44] I would hope Tableau does too, but I would love to see them evolve it more. But you can see the momentum with Copilot and their integration with Power BI that you could definitely see that evolving
[00:08:55] Dr Genevieve Hayes: Yeah. Because what I'm thinking of here is I remember when I started work in the insurance industry this was before R and Python became mainstream, and everything was done using SAS. And then R and Python became increasingly more and more socially acceptable, and the finance teams across multiple organizations started to say we've got this massive bill each year for SAS.
[00:09:21] Do we really need this anymore, or can we shift over to this free tool?" So I suspect that will ultimately be the end of many of these tools, where people say we've got this massive bill each year for Power BI or Tableau. Can we cut this from our budget and then have you switch over to the thing that you really wanna be using?"
[00:09:42] Nicholas Kelly: I think you're spot on. I think that's what's driving it, is that licensing costs. Certainly in the case of Tableau, that's their struggle. Since Power BI is already part of I think it's the E5 license, maybe E3 license from Microsoft. It's there.
[00:09:53] It's already there. Hey, guys, might as well use it. We're paying for it. Yeah. Certainly it's budget driven, and they're gonna start to see more of these capabilities coming out from their team being able to do this. Why are we paying this amount? And then also the pressure for what they're paying for cloud compute inference.
[00:10:12] That's also gonna push, "Okay, we're paying all this for inference. Why aren't we doing more with it?"
[00:10:17] Dr Genevieve Hayes: Yeah, exactly. Everything comes down to dollars and cents at the end of the
[00:10:21] Nicholas Kelly: Yeah. That's right. Yeah, fine
[00:10:24] Dr Genevieve Hayes: So what I'm hearing here is that essentially the role of the data team is going to remain more or less unchanged, but AI is fundamentally going to change the way in which data professionals will need to work going forward.
[00:10:40] Does that sound about right?
[00:10:42] Nicholas Kelly: Yeah, sounds right. Maybe three months ago, I did a training at a company. It was a BI team. And I was testing out a new workshop/course on how to take data professionals and train them to also be AI-enhanced, like the thesis of my third book.
[00:10:59] And I had a great deck, of, all the content, all the theory and we got about halfway through, and I was like, "You know what? I need to practice what I'm preaching here." I'm- I've been talking about the theory. So I just said to the team "We've got a choice.
[00:11:11] I can carry on for two more hours and show you slides," and, I think they're valuable slides, but would it be better if I just built something in front of you right now? And unanimously everyone's like, "Yeah, let's do that." And I said, "Okay we need to pick an idea."
[00:11:26] So There was like, I don't know, 12 people maybe there. And so we said "Everyone pick an idea. You're all gonna go into," not to promote any tools, but Figma for wireframing. And Figma recently added in this AI capability called Figma Make. And so all of them go in, tell it a problem you're currently dealing with in the company, like you have to do something, right?
[00:11:50] For example, build a dashboard for showing credit delinquencies. Cool. So they went and all came up with wireframes, and then we picked one. It's okay, what am I gonna build? Let's pick one. And so I built it there on the spot from scratch, Dr. Hayes. That laptop I was using had nothing on it.
[00:12:11] It didn't have Visual Studio, it didn't have Docker, it didn't have SQL or Postgres. There was nothing on it. And so it took an hour and a half, and we had a functional app, so much more valuable for the team. They got to see this is what's possible. They wouldn't even have contemplated building that.
[00:12:29] That would be the software team that would've done that. And so then they saw, and I think this is a really important point, Dr. Hayes, they don't need years of software development experience. It helps. It's not a bad thing to have that. It's definitely not a bad thing.
[00:12:43] But if you have some competence in, which is the basis of my thesis here, if you have some competence in data, you work with data, you're already dealing with technical problems. Now, you're gonna have to expand that domain of the things that you're dealing with. But using AI, in this case, AI-assisted coding you can be way more productive and, for me at least, fulfilled in your role.
[00:13:10] You're doing more. You're doing more of the things that are interesting, you get to build something, who doesn't like building stuff, it is so cool to build stuff. I get real- joy out of just making something with my hands. Like just going into the garage and building something, or let's say cooking something, or whatever it is.
[00:13:27] It's there. You can see it. People can react to it. A lot of time when we're working with data, especially on the back end, we don't get to see the bot we've built, and you usually get to hear complaints from the end users, and that's not really fulfilling. Even if what you did was really good, if someone put a bad front end on it, you're not getting any of that good feeling from what you've done,
[00:13:46] so I think it makes things a lot more rewarding for people if they're willing to stretch a little bit out of their comfort zone. But also, I think people need to see that it's possible, that it's not that far off. You could do this. With a background in, especially like, data engineering those folks should 100% be AI engineers, you know, MLOps, DevOps.
[00:14:13] It's really not that far off. And, even if you'd, s- asked me this six months ago, I'd be like, "Eh, that's borderline." It's borderline. Now, it's "Why are you not doing it?" It's more like that. There's a developer, Michael Abrash He wrote this book.
[00:14:25] It's basically the bible for computer game programming back in the Doom, Quake era.
[00:14:30] Yeah, back at the end, I think it was the end of 2025, he said some of the coding models now are on a par, if not exceeding his ability. And this is someone who's can code in Assembly, right? Profound good coder, so if that's the case, then most people with a technical background, that is now within your realm. If you can just use some good best practices in terms of engineering there are absolutely risks, but it is within your realm of possibility that you could start using this stuff to build more and have a more fulfilling career.
[00:15:06] Dr Genevieve Hayes: So that our listeners can get a clear picture of what it means to be a AI-driven data analyst or data scientist, can you walk us through the process of what it would look like for a
[00:15:15] data analyst to build the equivalent of what would've been a Power BI dashboard two, three years ago?
[00:15:29] Nicholas Kelly: Sure. So what that journey looks like, let's say you're a BI engineer, and We wanna take a typical dashboard, let's say Power BI, and instead now we're gonna surface all of that information directly into Teams. There's two ways you could look at this.
[00:15:45] One of them is a major barrier. You have to go research everything. I just gotta figure out all the architecture, all the technologies, and feel like you have to learn all of the, let's say TypeScript and Node.js and maybe React, and then, you need REST API experience, and you have to build a database da, all the things.
[00:16:02] So you'll be inundated with new things. Possibly a vector database, there's a lot. Now, you can do that, and that's great, and I would still advocate that people need to learn all of that stuff. You shouldn't be doing this blind, but we actually tested this with a Tableau developer roughly this time last year, To see if we could get her building and delivering insights directly into Power BI.
[00:16:27] And we didn't get rid of Tableau. Fortunately, Tableau's got a pretty nice semantic layer and XML format to how their workbooks work. So we were able to just take the definitions using code, and we talked her through it. And helped her basically use Cloud Code and ChatGPT at the time to help her with the architecture and go, "Okay, this is the structure, what needs to happen."
[00:16:50] ChatGPT allowed us to come up with a really good plan that we put in Excel. So here's the overall architecture, and here's the plan, here's all the features that need to go into it. So basically, we gave her a few different templates that she would be using, but she largely did the work.
[00:17:06] And just having good rigor around a software development process, which was obviously new to her as well. But you can always look at this and relate it back to things you're normally doing in BI. Like, when you build a dashboard, you don't just throw it out there. Of course, you do testing, and then you want to validate all your calculations. You have to make sure your data's good. You've got quality. You have to do ETL. There's all of this stuff you're doing in the BI space that has not necessarily direct correlation in coding, but there is a lot of overlap, and certainly enough that it makes logical sense why you would do this.
[00:17:37] So that was, our kind of first test with taking someone through that journey. And so it was like drinking from a fire hose, for sure, initially, like, all of this new stuff. But she was able to achieve it, and then she became the person in the organization that is now doing that for them because she's gone through it.
[00:17:55] Now, the thing is, others in her space in that organization think that's out of reach for them still. At that time, they were like, "Oh, she's a genius." She's smart, for sure, but the leap for her was actually just seeing how to do it, just seeing that it's possible. You can do this.
[00:18:12] Dr Genevieve Hayes: So you took a Tableau developer and you used AI to help coach her into becoming a software developer as well.
[00:18:22] Nicholas Kelly: Yes
[00:18:23] Dr Genevieve Hayes: And now to people who haven't seen behind the curtain, she looks like a genius and whereas it's just using AI as support staff. Is that right?
[00:18:36] Nicholas Kelly: 100%. So Dr. Hayes I think the thing is so someone who's in BI, they already do two really hard things. They work with the data, clean it make it ready for consumption, and in theory, they work with the business. So which would be harder, is to take that person and add another sort of skill set to them that's using AI and AI enhanced, or take someone who's got the development skill set and teach them how data works and how to work with the business?
[00:19:06] So if you just look at the gaps there which gap is easier to solve, and for me, the data person already has a lot of those skills. So we're certainly adding a new skill, a new capability. But it is not like a four-year degree capability. Certainly to understand it, it's gonna take you time, but you can almost fake it until you make it right now with the capability of the tools that are out there.
[00:19:27] And don't get me wrong, you definitely need to understand how it's all working. 100% you need it. You need oversight. You need a process in place. But you can do that with process. You don't have to do it with four years of knowledge, and I'm saying this speaking as a computer scientist,
[00:19:41] if I was gonna be selfish about this, I would say, "No, no one can do that." You can only be doing this if you have a computer engineering, software engineering degree. It's not the case, this is accessible to people who have a good problem-solving background and If you're capable of working with data and you're capable of working with the business, then you're definitely capable of this jump this kind of leap in your career
[00:20:03] Dr Genevieve Hayes: So you still need to know that whole stakeholder engagement piece and you still need to have the data fundamentals. So I'm guessing you'd still need to know things like the fundamentals of statistics how databases work things like being mindful of data security, privacy, governance, all those things, because otherwise it's just gonna be a ticking time bomb.
[00:20:26] But it's just a different tool that people are learning, which leads to them being more productive, and the output is a piece of software rather than a BI dashboard, for example.
[00:20:38] Nicholas Kelly: Yeah, you're spot on. And also , that keyword you mentioned, governance, data professionals should be used to data governance. We have to extend that into the AI field. Now, the other thing here is as well, when we're using AI, it is producing data, so it becomes itself a way to reinforce that a data professional should be the one who's working with this,
[00:20:58] 'cause now you're seeing, let's say you've got a bunch of business users and they're all typing into the, teams chat agent, " can you show me this metric and this one? " That's all requirements. It's all data. It goes back into your iterative cycles
[00:21:12] and it's okay, people are always asking for this. We don't have the data for that let's plan for that. Let's figure out what we need to get in there and how do we do it so that we can answer those questions for them. If anything, it's a very natural evolution, and you could argue it is the most natural place for someone to move more into AI from than any other field apart from directly AI,
[00:21:35] but any of these tangential roles to AI, the data people make the most sense to move into it
[00:21:40] Dr Genevieve Hayes: I remember 10 years ago when the organization I worked for first got Tableau, and the promise of Tableau at that time was anyone can build a dashboard using Tableau 'cause it's just drag and drop, so even someone without data experience can use it, and, you had that thing about citizen data scientists at that time.
[00:22:00] But the reality was you didn't have the person in marketing going out and building their own dashboard. You might have had one or two, but they were usually pretty into data to begin with, or the person in, HR building a dashboard.
[00:22:15] After people had a little go with it and decided, "Yeah I've seen it," it just ended up going back to the data team because they knew what they were doing, and they enjoyed doing it, and it was the best place for it. So it's probably gonna be the same with this. You'll have some people in other areas of the business who give it a go, but they've got their jobs to do.
[00:22:38] They don't have time to spend building AI apps all day. So it'll go to the people who this is what they do all day. And I don't think we're gonna have citizen AI engineers all over the place.
[00:22:53] Nicholas Kelly: I think that's a really good insight. Yeah, I think you're right. I think initially people will think, "Oh, this is easy. I can build an app." And then they'll realize actually no, because you still have to do user testing, you still have to do all these different things that you need to do when you're developing software.
[00:23:11] Totally agree with you. I think that's a very good similar journey that we're probably about to endeavor upon and see that, yeah, like last year, you could have foreseen there's gonna be a whole lot of new CRM apps out now, because everyone's going, "Hey, this is easy."
[00:23:26] And yeah, it is easy. It's easy to develop, and the reality is though, most of their apps aren't getting traction because there's so much other stuff you have to do in order for it to be successful, so like it's a good thing, but there's also, risks involved in that we are probably overconfident in, "Cool, I'm a coder now."
[00:23:47] But it doesn't mean you can avoid doing the change management and all the people things you need to do, which kind of, for many people, suck, but AI can help you with that too, it can help you come up with a change management plan. It can help you communicate with certain folks in teams.
[00:24:02] You could even, build agents to help you do some of that change management. So it does make the job easier and certainly more fulfilling, but we also just have to be aware of all the other things we can't ignore just because now, hey, we're building apps. This is great, but it doesn't mean this is the next tool to ignore people with, basically.
[00:24:20] Dr Genevieve Hayes: Or once the first catastrophic failure happens within an organization, everyone will be told that this functionality will be limited to the data team from now on.
[00:24:30] Nicholas Kelly: Yeah, you're spot on. Again, it's a great analogy with what happened with Tableau. It's gonna be very similar. It needs governance. It's gonna need, good proper guardrails around it, 100%.
[00:24:40] Dr Genevieve Hayes: So what's the single most important step our listeners can take tomorrow if they wanna remain relevant in the AI era?
[00:24:48] Nicholas Kelly: For me, it's experimentation. That's been by far, the biggest accelerator. You need to experiment. You need to build stuff. Okay here's a good one to do, go build a scheduler for your household, like an app that takes in, let's say, all your financial da-data, all your calendar data, all your, if you have smart devices.
[00:25:08] Get in all of that and get it to integrate into a single app that then sends you a ... let's say you have Signal or WhatsApp or whatever, and then so in the morning you get a briefing, "Hey, you got to pay this bill today, and, at 4:00 you gotta do this," and, turns on your lights for you automatically or, does all these cool things,
[00:25:28] I think it's very similar to what I used to advocate for people when they're building their first dashboard, is that it should be something that's personal. Like your fitness tracker or something like that. Something you actually care about. It shouldn't be something for the business, at least from my perspective because you don't really care.
[00:25:44] It's "Oh, that's my job. That's cool and all." But if you can get some of that passion, those things you're, like, you like doing, and build an app using, one of these tools then I think that's a great way to start, 'cause you're gonna start to see the stuff.
[00:25:59] You're gonna apply your own pressure to figure out how you do this, 'cause it's a passion project for you versus that probably you wouldn't try out all of this stuff if you were doing it for something in work, so that would be my advice. Start there.
[00:26:10] Dr Genevieve Hayes: Yeah. I'm experimenting with AI with a lot of the things regarding my podcast. Can I use AI to grab the transcript from Zoom and then transform it into a format that allows me to save it in my database and extract the key information from it and then send it to me, and things like that.
[00:26:29] It's not anything that's going to make or break my business, but it's something that I'm interested in and if it doesn't work it's not the end of the world
[00:26:39] Nicholas Kelly: 100%. 100%. Love it
[00:26:41] Dr Genevieve Hayes: So for listeners who wanna get in contact with you, Nick, what can they do?
[00:26:45] Nicholas Kelly: I'm pretty active on LinkedIn, so you can just go to LinkedIn, look for Nicholas Kelly. That's N-I-C-H-O-L-A-S K-E-L-L-Y, and you can find me there, and feel free to reach out, and you can message me on LinkedIn
[00:26:58] Dr Genevieve Hayes: Okay. And that's it for today's episode of Value-Driven Data Science. But if you want more from Nick, next week you can catch our Value Boost episode, where we explore why organizations shouldn't be building their AI future entirely on rented intelligence, and why understanding this puts you ahead as a data professional.
[00:27:20] And if you found today's episode useful and think others could benefit, please leave us a rating and review on your favorite podcast platform. That way, we'll be able to reach more data scientists just like you. Thanks for joining us today, Nick.
[00:27:34] Nicholas Kelly: Thanks for having me. It's been a pleasure
[00:27:36] Dr Genevieve Hayes: And for those in the audience, thanks for listening. I'm Dr.
[00:27:40] Genevieve Hayes, and this has been Value-Driven Data Science.