AI, Utilities, and Solving Key Business Problems: Part 2 – AI by Any Other Name

Utilities have been doing AI for years, they just didn’t call it that

Before everyone started talking about AI, utilities were already running it. We just called it load forecasting, predictive maintenance, and demand planning.

The electric grid is one of the most complex systems humans have ever built. Managing it has always required predicting the unpredictable: equipment failures, weather impacts, customer demand, and grid stability. Utilities have been making the right decisions fast enough to keep the lights on for years.

Here’s where AI comes in: the math and concepts aren’t new, but the compute power and size of the data we can process in real time changes the game.

The data advantage utilities have — and the problem in leveraging it

Utilities have datasets most industries would envy: decades of operational history tied to assets, customer behavior, and regulatory performance.

At UDC, we work with utilities managing decades of operational intelligence: SCADA tracking real-time grid conditions. Smart meters capturing consumption at 15-minute intervals. Work order histories documenting every repair, every failure, every crew dispatch. Inspection records. GIS networks modeling assets for operations and insight. Outage logs going back decades.

For us, the challenge isn’t having enough data- it’s data being ready.

Maybe because the data lives in different systems. Or one department might own inspection records, while another owns maintenance scheduling. Often, data standards vary by region, by vintage, by vendor. Asset IDs might not match between GIS and work order systems. Historical records have gaps, inconsistencies, or migrations that left out some things. We have data, but it isn’t in the shape necessary to help AI work at scale.

And it’s not just legacy data. The volume of new data coming in is astronomical: LiDAR, aerial imagery, 3D models, real-time sensor feeds filling datastores faster than we can organize them. How do you match 30-year-old work order records with this week’s drone footage? Without governance and cataloging, more data becomes more noise

At UDC, we are working to help organizations leverage their investment in AI capabilities by addressing the data foundations that make those capabilities useful. We are already acquiring new data. We need to pair that with making the data we have usable. 

Where AI creates value today — and why data quality is the key

When utilities get the data foundation right, the results are genuinely impressive. The AI applications delivering value aren’t the flashy ones getting headlines. They’re the operational ones improving reliability, reducing cost, and keeping people safe. And we’re seeing them work. 

Predictive asset management is a strong example. (PS. My colleague, Kelley Rodriguez, has written a series of articles about how this can impact Transmission^[1]). AI models that have been trained with the right data (load patterns, failure history, etc.) can accurately predict issues with assets in the near term. Research shows AI-powered predictive maintenance can reduce equipment failures by up to 73% and cut unplanned downtime by 40%.^[2] Some utilities are documenting 20 to 30% reductions in emergency maintenance costs.^[3] 

How are they succeeding? Integrating sensor data across systems, aligning information, and standardizing. They have spent years cleaning and organizing their data to enable the transformation. And UDC is working with utilities to build this data governance directly into transformation projects: metadata cataloging, data quality standards, integration architecture.

Outage prediction and response, a critical component of our work, is also getting real traction. Aggregating patterns across weather data, grid layout, and historical information allows utilities to prepare crews before storms arrive.^[4] When 80% of outages in the US are caused by weather, this makes a huge impact on utilities.

Again, success here requires tying data together across many sources: outage logs, weather services, and dispatch systems. Work that organizations like Dominion Energy have made big strides to enable.^[5]

The message is real: AI doesn’t fail because the algorithms aren’t cool enough. Research shows that 70 to 85% of AI project failures are due to poor data quality, not algorithmic problems.^[6] RAND Corporation found that over 80% of AI projects fail to reach production, twice the failure rate of traditional IT projects.^[7] We see this across the industry as the models are advancing every day. What holds back value isn’t algorithmic- it’s the quality, accessibility, and governance of the data that powers those models.

The real edge

This is not a sexy message. Not many people get excited about data cataloging, master data management, or API governance. But that is where the opportunity is!

We are focused on this foundational work every day at UDC because we’ve seen what happens when utilities treat data as a strategic asset, with ownership, standards, and quality controls. The models that were already running (load forecasting, predictive maintenance, outage analytics) can start delivering results that weren’t possible before.

The grid is a data problem. We in the utility industry have more data than almost any other sector. The question isn’t whether AI can help. It’s about whether we’ve made our data usable, cataloged, governed, connected, and reliable enough that AI can actually amplify what we’re already doing.

Because the algorithms are ready, the question is whether our data is.

Next in this series: One of the most powerful — and underutilized — datasets for utilities isn’t in your SCADA system or your smart meters. It’s in your maps. And most utilities have no idea how valuable it is.

Missed part one of TJ’s AI series? Read it here: AI Isn’t a Magic Wand