How Public Power Is Putting AI to Work

Public power utilities big and small are already looking at what artificial intelligence can do for efficiency, forecasting, and more. Tasks range from supporting everyday interactions to conducting deep analyses and keeping up with changing regulations and market conditions. The aim of combining AI-based tools and technology with utility expertise is to improve safety, inform risk management, enhance reliability, and lower costs for utilities and customers alike.

While AI tools are still being honed and trained, customers are increasingly becoming accustomed to — and might even take for granted — how AI has become ingrained with daily interactions. However, these tools present risks if given automatic controls that exceed their capabilities.

As public power organizations explore and implement these tools, they are doing so with caution and consideration for utility and community security while working to keep up with customer expectations around efficient, personalized service.

Efficient, Effective Interactions

Often, a first place for AI adoption for utilities is applications that aim to enhance customer service.

In one example, Greenville Electric Utility System in Texas has managed 20% annual growth over the past six years by adding only two full-time call center workers and offering customers the option to use an AI-based chatbot.

“We didn’t feel that we had the staffing to manage a live chat in a way that a user might expect,” said Kami Williams, community connections specialist for GEUS. “Whenever you get on a chat, most people think they’ll ask a question and then have a person on the chat within a couple of minutes, but our [customer service representatives] are already busy helping people who walk into the lobby, call us, or send us an email.”

GEUS has 18,000 meter connections, and the utility manages billing for city services including water, sewer, and trash. The team picked a system that can deliver pre-trained answers to customer queries or point customers to a page on the utility’s website that answers the question in depth or allows them to take action.

“Our customer service manager takes the lead on writing and editing some responses, but I recommended we direct users to the website where we already house information, too,” Williams said. That recommendation means the team doesn’t need to constantly update the chatbot responses when things such as rates or program offerings change.

Williams said getting the chatbot started took minimal time. “We looked at responses daily for a few weeks,” she explained. Now that the system has been in place for a couple of years, the team reviews responses each quarter and makes changes as needed.

In the week before the interview for this article, Williams’ dashboard showed that the chatbot answered 117 queries from 77 unique users. The self-resolution rate was 78%, which she explained means the chatbot answered 78% of the questions asked in a way that provided customers with the information they needed.

In Colorado, Colorado Springs Utilities doesn’t use an AI chatbot with customers but does support call center staff with AI tools. Service agents save time with real-time access to a knowledge base, making them better informed and making responding faster. Those agents also save time with AI-generated wrap-up codes and call summaries, which have decreased after-call activities by 20 seconds per call.  

Springs Utilities has implemented Voice Analytics (Genesys), which provide insights from every customer conversation, automated and searchable transcription ability with topic and behavior intelligence, advanced customer sentiment, and agent empathy analysis. All of this provides the utility with better visibility into why customers contact them, improved coaching and quality assurance for their agents, enhanced customer experience, and increased operational efficiency.

Salt River Project, which serves 1.2 million electric customers in Arizona and fields over 1.5 million customer calls each year, is looking at similar uses of AI, such as generating call summaries or using prior interactions to help contact center agents understand why a person is calling. Kaitlyn Libby, director of corporate strategy and risk management at SRP, said an AI tool the utility is piloting provides real-time insights and assistance that enables contact center agents to focus on the discussion rather than taking notes or toggling between tabs to search for information. “We’re not planning on having AI interact with our customers directly,” Libby said. “We are really trying to figure out how it can assist our agents behind the scenes.”

Informed Decision-Making

Utilities are also finding that AI is suited for analyzing a complex array of factors for more accurate forecasting.

The procurement team at Springs Utilities has time-saving uses for AI, said Rich Norton, the utility’s general manager of supply chain. “We’ve used AI to assess supplier cost increases by analyzing large volumes of internal and external data to determine whether requested price changes are justified,” he explained. Among other things, the AI looks at real-time prices for commodities such as the metals used in transformers and other equipment to estimate a should-cost baseline.  

“AI quantifies reasonable portions of the price increase, highlights potential overcharges, and generates recommended actions for commercial negotiations,” Norton continued. “This enables my team to move from reactive manual validation to faster data-driven decision-making with stronger negotiating leverage.”

The Springs Utilities procurement team is in the early stages of using AI to compare supplier proposals. Responses to a complex RFP can be hundreds of pages long, which means that historically, evaluation teams had to spend considerable time to assess and score such proposals. Norton said with the use of AI, reviews of supplier proposals can be completed in a matter of minutes, although he warns that one still needs to carefully read the executive summary AI provides and ensure it aligns with the proposal.

Looking ahead, Norton’s team is developing an AI-driven tool for labor and materials cost trends to provide forward-looking market insights. The platform evaluates a broad range of external factors — including geopolitical developments, executive orders, and regulatory changes — to help anticipate cost fluctuations and support more informed decision making. “In today’s environment, we have to account for all those variables,” Norton said.

Another tool the Springs Utilities procurement team is developing with the technology team will analyze past statements of work to help employees write better SOWs.

Continuous Training

SRP has given staffers who write computer code access to GitHub, a developer platform where programmers can write, store, share, and manage code. Owned by Microsoft, GitHub has its own version of Copilot that helps developers write code, so it’s speeding up the development process. However, the utility’s initial testing team found that machine-written code needed human review and correction, Libby said.

The need for human intervention, correction, and training is common in AI. When SRP installed an AI-powered camera system to monitor remote transmission lines running through forested land, the camera vendor also provided 24/7 customer support, including vetting the images that train the AI to more accurately distinguish actual wildfires from weather phenomena or campfires. Now the system sends real-time alerts to SRP’s emergency services team so they can review the cameras’ footage before taking action, Libby said.

In addition to its remote-monitoring network, SRP has sensors along the transmission line that record temperature, wind, and precipitation. These data help the utility identify risks, such as high winds and low humidity, along specific corridors of the system, allowing grid operators to de-energize parts of the system rather than implement a disruptive, widespread public safety power shutoff. It also helps operators flag wind corridors so that maintenance workers can check for excessive conductor sway or prioritize tree-trimming operations.

Test First

All of the applications SRP and the other utilities are now using were piloted before being rolled out.

“AI is sort of a crawl-walk-run sort of thing,” said Travas Deal, CEO of Colorado Springs Utilities. “First, you have to have governance around it. Then, you look for those lower-risk projects and test them out. But I think the opportunities and utility are endless. AI is going to be transformative for our organization and others.”

Deal spoke more about how Springs Utilities is using AI at APPA's 2026 National Conference in Boston.

Both SRP and Springs Utilities are trying to apply consistency and guardrails in AI adoption. Both allow staff to use Microsoft Copilot, and both have implemented policies to ensure the basics are covered. At Springs Utilities, that means multiple departments vet all software and AI applications before they are made available for business use. The cybersecurity team has implemented controls that prevent users from accessing unapproved AI platforms while providing secure solutions such as Copilot, Deal said. “Nothing else can be downloaded with any Colorado Springs Utilities computer or cellphone.”

The Springs Utilities team continues to train all 2,300 employees on compliance issues related to data and AI use. Like Springs Utilities, the Salt River Project team is carefully training employees and working to protect sensitive data from AI exposure. “We’re not putting any sort of sensitive customer data or critical infrastructure information into AI tools,” Libby said.

“Whether the organization’s ready or not, you have employees who see AI and use it in their personal lives, and, of course, that starts intriguing them,” said Deal. “When it first started really taking off — before we even started looking at how to utilize it — we wanted to make sure that the AI applications our employees were using met our cybersecurity and privacy standards, especially when it comes to the protection of customer data.”

Data issues aren’t the only risks AI can pose. There’s a price tag, too.

Libby acknowledged a strong appetite for AI across the utility, encompassing requests for many different tools. “AI’s grassroots movement helped us build the early momentum, but could become a little disjointed over time if you don’t coordinate it,” she noted, adding that many AI tools are shifting to usage-based fee structures and could chalk up “very real” costs for the organization if not intentionally managed.