When AI Meets Infrastructure with Duke Energy SVP & CIO Richard Donaldson

So I went to college for engineering, civil and environmental engineering, and ended up doing that for a couple of years before this whole everyone regardless of their major was getting into it. So I followed suit. Actually worked at a software startup for three months before they laid the whole office off. So that's how I came running to the safest place I could find but to also continue to do it, which was Duke Energy. That was in 2001 and I've been here since. I've probably done just about everything there is to do in IT and enjoyed every minute of it. The nice thing about Duke Energy is if you're one of those people that has this fear of getting bored with what you do, which is very much my persona, we have so many different jobs at this company. So I always felt like there was no horizontal boundary to what I could go pursue. And that's been the case. I've been able to find everything I've been looking for within it, thankfully.

Yeah, we're in several different states. I can't even quote you the miles of transmission and distribution, but here's one that's interesting. Because everybody's talking about data centers these days and they're saying, well, a gigawatt data center. That's a big load. And that's like big news. The idea of someone talking about a gigawatt data center is a new thing. Well, at Duke, we're running 11 nuclear power units. Each of them are somewhere in the range of a gigawatt. So think about that. That's just on the generation side and the magnitude of the size of Duke Energy. And of course, we have a huge renewables portfolio. We have gas. We still have some coal. We've been around for a long time. We have a lot of hydro. So just from generation and then to transmit and distribute all that energy to all the places it needs to be, it is most of the folks at this company aren't like the three of us. They are putting on the protective equipment every day, getting in a truck, and going and climbing a pole or working in a substation or working in a power plant. It's a real source of pride for me even though I'm not on that side of the business.

Well, especially we're in North Carolina. The year we had last year with those horrible storms, we are critical responders in those cases. And we showed up very well, especially in western North Carolina to help those folks during that time.

I'm hesitant to call it a hype cycle because I think we're finding value a lot sooner than former hype cycles. And I really do believe that this is the real McCoy. This isn't, no offense to people that love blockchain, but we've all had those hype cycles where it either didn't go anywhere or even with AI where you had the AI winters of the previous decades. But I think we've got something. You see it right when you play with your favorite generative AI engine. It's a game changer. So I think we're in a good spot. I think that the value being sold and advertised and marketed, I think we're a ways from getting there because if you don't have your data and processes in the right place, you're just going to get surficial value versus deep value, especially in a big enterprise like ours.

Absolutely. In fact, just to go on the AI journey a little bit, 2024 was a year we spent a lot of time talking about AI, but in very unsexy ways. We were talking about how we're making sure we're protecting ourselves from our data going every which way but loose. What kind of governance are we going to put in place to make decisions about AI? What kind of platforms and security measures are we going to have so that when we go to develop a solution in AI or even leverage a commercial model, we're not going to get ourselves upside down? That was a lot of work. Platform work is hard and it takes a lot of time. To your accountant, it looks like you're spending money and not getting any return. You've got to pave the roads. So this is the year we're really excited to turn that page and say, okay, we're open for business and we think we can go pretty fast on the technical side. But to my point earlier, we can go as fast until we hit data that's not of high quality or that's not organized or that's in 800 different places. Specific use cases, by the way, that we just finished up. That's what'll ultimately become your hurdles and your slowdown in value.

I think so. Interesting is in the eye of the beholder. I was just telling one of my business partners the other day, everyone's waiting for me to say we did this AI thing and it sounds right out of a science fiction book. But the most valuable stuff sometimes doesn't have that spice to it. So I'm going to talk about two. One is we call it Duke Energy Explorer. We work in a world where we are continuously working with the regulators on rate cases, on our energy delivery plans, what we're going to do. There's a lot of asynchronous communications with that. So we submit large documents and all of a sudden we get a response back with 50 million questions and 16 hours to return the answers. I'm exaggerating grossly, but this was something where over time we have information on SharePoint sites, databases, cocktail napkins, you name it. And it was really hard to turn those around in a high quality fashion, in a repeatable fashion, and in the amount of time that we had. So we spent some time building a conversational AI tool to harvest the data that was in literally 800 different data sources and build this. It's not a fabric. We didn't go bananas with the architecture, though that might be a next progression. And now you get the questions and you can ask and get a quasi response back. You can go, hey, that makes sense. Or I don't know about this one, but it's got a little link to where the actual file was. Let me go pull that file. It really helps us move the needle with that, which is really important because every year we're working on getting regulatory approval for something. It doesn't matter what the administration is. If it's a big clean energy push or a big build big power plants push, you have to go through the regulatory agents in the states we operate. So that's one I'm really proud of.

The other one, and look, to me a great AI use case is something that does something that I would find painfully boring to do. So another one is, if you've seen in the news, I mentioned those nuclear units. We've been doing a lot of relicensing. We're very proud of and very good at operating our nuclear fleet. So we want to relicense those and run them for another 20 years as long as we can operate them reliably and efficiently because they're clean energy sources and the cost to operate them is actually fairly competitive. So if getting rate cases done is a six out of 10 on the paperwork scale, imagine working on getting a relicensed nuclear plant anywhere in the world, much less in the United States. That's a 13 out of 10. So we've used generative AI to help us put those together, pull all the operational metrics and the data together. I heard something staggering like a $4,000 project saved us something like 200,000 man hours. It was something very, don't quote me on those, but yeah, great use case and another one we're really proud of.

I don't know that any come to mind, and not to get out of the AI space, but I think the game changer is going to come when, because if you think about how we use drones, it's very sensory based and collection of information and then having the AI engine do the fancy math to get insights and to get 99% accuracy. But the drone's not doing anything yet. So the simplest example I can think of is when we have a drone that flies a solar site, big solar sites. If you haven't seen a solar site, it's not like in your backyard. It's like seven football stadiums. And flying all over it to look for everything from thermographic imaging that says we have too much reflection or not enough reflection to a dead cell. Well, what happens when the drone can actually just lower down and maybe sweep off a panel, something totally simple like that, or change a part, or do something like that? So I think that's what we need to be thinking about sooner rather than later. Not just the confined AI, the stationary AI, but what happens when you actually give it action similar to a self-driving car? What is that going to look like? That's when I think it's going to get very interesting. And I don't think we're that far away from that.

It's a great question because you roll something like that out to the IT workforce and they're breaking the servers, they're hitting it so hard. But then you go into some of these business units and it's just the latest button they have on their workstation and they don't have time to use it. I don't know if you guys have ever seen this cartoon, but it's the guy pushing a wheelbarrow full of rocks, but the wheel is a square. And there's a guy on the side of the road with a sign saying wheels $15, come talk to me. He's like, I don't have time for that, I got to move these rocks over here. That's kind of that thing. So one of the, because you're describing a very real problem. When people start talking about the path to value, that's your barrier to value. Because everything you say, I'm going to save an hour and a half per task across a workforce of X. Well, if you can't get penetration in that whole workforce, you're not going to save the value that you advertise in your business case. The way we found is you've got to understand. So, Sam, Evan, what is your pain point? Not what do I want you to use this tool for so that you will become more efficient. It's where, what are your pain points and let me help you make your job easier that day. That gets people. They used to call it the killer app, like mobile check deposit on your banking app. That's what got people to start using your mobile banking app. So you've got to find that, and it's not one thing for everybody. In your HR department, it's not one thing, it might be many things. So that's the strategy we've used. You have to talk to people. Historically, IT projects was tell me what you want, I'll see you in 18 months and you can test it and see if you still like it. And we'll get it to where you dislike it at a level that we can do a lot of training and change management and tell you how to click the 17 buttons to get to what you want and then get it into production. I'm not trying to make this an agile commercial, but that whole side by side talking to folks and figuring out what do you need to help you do your job? And oh, by the way, there's things we're going to want you to do as part of your job because it's valuable to us holistically. But somewhere in the middle lies a real happy path.