How Should Utilities Prioritize AI Investment?

Teri Viswanath

April 24, 2025

Teri Viswanath, CoBank’s lead economist for power, energy and water, recently interviewed Casey Werth, the global leader of the energy industry at IBM. His team works with energy and utility companies to drive energy sector sustainability and energy transition by improving the usage of data and AI, automation, enterprise systems optimization, asset performance management, and cognitive customer interaction.

Teri Viswanath: Casey, I am so pleased that you’ve agreed to sit down with me and rehash that terrific discussion you had with our electric co-op executives at the NRECA CEO Close-up earlier this year. Can you reflect a little bit about the race for making AI investments from a perspective of electric utilities and how do they fit amongst other industries making similar investments?

Casey Werth: You are likely referring to the IBM study that came out a year ago, ahead of Distributech 2024. It was fascinating in that of all the industry segments that had the highest expectations of positive impact from AI, the energy sector was in the highest, I believe top two or three. However, the pace of adoption and the pace of progress was, of course, in the lowest two or three. So, we have this divergence in our utility of high expectations versus a relative slow pace of investment or adoption.

Viswanath: Interesting, so utilities expect to travel far with this technology but haven’t really made it off the starting blocks? Can you talk about the different areas where AI is actively being or could be deployed?

Werth: Generative AI is very hot right now as a topic, but AI and ML (machine learning) algorithms and capabilities are in many of the packages, the products, the vertical products we already use. When you look at an EMS (energy management system) or an ADMS (advanced distribution management system), the capabilities are there in terms of modeling, forecasting and running models to make sure that we are within the margins – where we need to be from a supply as well as within our standards safety standards. This is not insignificant.

Viswanath: What are some other examples of AI deployment in utilities?

Viswanath: Ok, that is helpful. As I recall you mentioned a broader array of areas, where we might expect AI to be deployed on a go-forward basis, can you help frame that discussion?

The first three of the nine are all related to power flow modeling that enable us to plan, forecast and orchestrate grid capabilities – a foundational model that we could use to change the way we fundamentally view the grid. Areas four and five are tightly coupled to use-cases around large language models (existing tech) and are tied to ticket analysis, trying to get better prediction around operational programs. Seven and eight are geared towards bringing chat into capabilities and improving training and customer interfaces. Six is about changing trading and improving market making – right now high frequency trading and market making is really is still just brute force compute. Nine is about resiliency in cyber and physical security. One of the first AI use-cases is security – there is always threat hunting and threat analysis. On any given day, security operation software is going to get thousands, tens of thousands or even hundreds of thousands of alerts and attempts of intrusion. Having the ability to sort through this volume and surface the handful of viable threats to hand over to security analysts is mission critical.

With the ideation and the path towards definable, quantifiable value of these technologies defined, however, we’re still going to continue to struggle with adoption. So, if we can pick an operational area like asset management (which for every single utility is the core of what we do) and apply technology to improve the ROA or operational procedure, it becomes a no brainers as the outcome is going to prove so powerful.

The point being, if you build a chatbot that allows you to search your website better, terrific.  But if it’s going to cost you $1,000,000 to do that, this will be a difficult investment decision.

Viswanath: I want to take a step back and understand some of the foundational elements for AI technology application. How can AI help with utility system planning?

Viswanath: There was the part of your conference discussion that I was really struck by…you mentioned that it is important to align an organization’s AI investment with the largest capital outlay – having a practical way to pencil in broad savings across the utility. So, how can AI improve asset management?

Werth: I asked the executives of one of the large utilities we have been working with on where we could find the largest organizational impact. This is an important lesson I’ve learned, personally having gone through several proof of concepts, or POCs.

Sometimes these technology discussions are challenging to start with as they require a base level understanding of the scope that that technology could feasibly impact. And if you don’t have that, then the use cases are so narrow that they will ultimately not make that large of an organizational impact. Having that right mix of corporate leadership vision with an awareness of some of the bounded can make the brainstorming more effective.

Chris VanLokeren, North Carolina’s Electric Cooperative’s CIO, and I spoke about the peak prediction model at the conference, which is one of the areas where we landed on that would be an effective path for investment after multiple hour discussions. To find the use case, you need a couple of characteristics. The project shouldn’t be so complex that it’s impossible to do, but you also need it to be a certain level of complexity so that when you apply automation to it, you really are compressing a process that has meaningful impact. This is what we found for this project.

With the peak prediction model we went from POC to production seemingly overnight because there was an immediate recognition that this effort was fundamentally changing and improving an existing process that would impact returns. There was a lot of considerations around software lifecycle management, scaling support, high availability etc. but we didn’t really get all of those things done upfront because there was the recognition that this model works…so, use it. With these POCs, you might land on a goldmine but then need to still pump the brakes a little bit to say, “OK we got it.”

Viswanath: You mentioned that you wanted to showcase solutions that are grounded, right? How has IBM applied AI?

Werth: We have a legitimate answer here. We have roughly 180 corporate-level use cases that we’ve brought generative AI workflows in what we do. And can share all the things we’ve learned about it.

Viswanath: In particular, I’m curious about how AI might improve enterprise performance management (EPM)?

Viswanath: Casey, you mentioned that the investment in AI might come sooner than planned but many not in the way that we saw the future unfolding. So many critical utility tools or resources have either embraced this technology or will soon incorporate the technology into their primary offerings, can you expand on this?

Werth: Yes, we will all be shipping products with AI capabilities in the box. If you look at GE, Schneider, Siemens, Hitachi – really most of the large operational technology vendors – are going to start providing AI capabilities more so than they have already. We will now be in a moment where utilities are going to have to figure out how do we orchestrate AI because it will be in most of the products we use.

Do we have agents talking to agents? Because now we’re in the age of agentic AI. And if our organizations are still at the starting blocks, this might leave us at a conceptual disadvantage as we’re now going to be on the third lap of the mile with these products being dropped in our shop.

AI will be embedded in all the new technology purchased or even when you do an upgrade, it will now likely come with its own chatbot or other AI capability. So, I see this as a potentially missed opportunity of not getting more value out of your investment because there’s not a level of awareness.

Once again, I just see that with our sector, there are high aspirations but low adoption. Fully leveraging the tools we have that are now embedded with new capabilities is just another potential hurdle we’re going need to be prepared for.

Viswanath: Switching gears and thinking about AI from our electric consumers perspective, what is the potential impact of AI on load growth?

Werth: If we could get to 20% improved utilization assuming a 20% load growth, just think about this… We could do a lot more with what we have if we had better insight and better knowledge of real-time conditions so that we can make more informed decisions.

Viswanath: I once heard it said that if data centers represent a total of 4% of electricity used, imagine if that use could enable the 96% of us to become more efficient.

Werth: That’s groundbreaking for our sector.

Viswanath: Casey, I want to thank you for sharing your insights with our CoBank customers. Appreciate your time and insights.

Werth: My pleasure.