Illustration by James O—Brien
About four years ago, when NiSource began pushing into using artificial intelligence in its operations, the endeavors seemed like an advance into a “bleeding-edge” technology.
“People usually say ‘leading edge,’ but we were ‘bleeding edge’ to the extent that few utilities had performed this work, and it required a significant commitment of resources and effort,” said William Mojica, NiSource senior vice president of technical services and asset risk management for the company’s six-state territory. “We had to fix a lot of the legacy data and information stretching across multiple states. There were a lot of process improvements and technology enhancements, which helped us to better govern data, and we continuously worked to enhance such data.”
Once the process was complete, company officials better understood the quality of data, the proper steps to govern it and the strategy to drive risk-modeling capabilities. Those efforts now allow the company to react more quickly, efficiently and safely in its dayto-day operations, thanks to AI, Mojica added.
“Think about a neural network that is constantly functioning behind the scenes to learn where the highest risks are and to input that information and then return a model that paints a picture of where to go to address the highest risks,” Mojica said. “I think that is the greatest opportunity we are seeing.”
AI is expected to revolutionize numerous industries, and although some companies have been slow to adopt it, they likely will start to see the benefits as others succeed, Mojica said.
Like NiSource, other stakeholders in the natural gas industry have been seizing on the technology. On the operations side, everything from prioritizing infrastructure replacement to interactions with contractors submitting 811 “know what’s below” requests can be streamlined through AI modeling. According to Richard Meyer, vice president, energy markets, analysis and standards for the American Gas Association, members are widely using AI for streamlining data analysis to assist with issues such as damage prevention, routing work orders, safety predictions, corrosion predictions and leak quantifications.
Most consumers and businesses are already using AI without knowing it—generative AI is the technology beneath the chats and voice commands they rely on online and through automated phone systems. Natural gas utilities are exploring gen AI’s potential too—“with cautious optimism,” noted Meyer.
“AI includes machine learning and data analytics, which we’ve had three to five years as a mature discipline, and it is still ever-growing from an analytic standpoint,” said Cullen M. Hale, enterprise architect for AI strategy at Consumers Energy based in Michigan. “From gen AI and the recent upsurge in AI capabilities, I think we’re still probably with the rest of the organizations trying to understand how best to utilize it. It is getting more mature, and we are finding more uses.”
For utilities across the board, acquiring and analyzing massive amounts of consumer and other data gathered from advanced metering infrastructure and other internet-connected devices has been a top priority for at least the past three years, according to Artificial Intelligence for Natural Gas Utilities: A Primer, a report by the National Association of Regulatory Utility Commissioners. Some utilities are now at a point where that data is helping them direct resources more efficiently.
Consumers, for example, leverages AI so crews can better understand a job in the field and benefit from detailed insights to help them prepare. “They could then reduce incidents and the lead time for getting a job done,” said Hale. “But it’s also trying to think how we uniquely solve problems with this toolset.”
At NiSource, company leaders have been working specifically on the problem of incident reduction, using vendors to develop proprietary software that assists its workers, contractors and customers to interact more seamlessly and to improve response times to address the highest-risk 811 tickets, Mojica said.
While the company’s AI model considers more than 300 risk factors, the four main drivers of risk include excavator risk (frequent damager); work type (water, sewer, cable); ticket type (normal or emergency); and location complexity, said Nancy Maynard, who manages NiSource’s ticket management system as director of damage prevention strategy and centralized operations. For example, water and sewer lines are often located below gas lines, so incident risk increases when contractors are doing water and sewer work, Maynard said. “AI takes all of that into account,” she added.
Logically, deeper pipelines might mean a lower chance that a contractor will strike one, but they also could mean that a contractor assumes it might be clear to dig, Mojica said. “The model learns from where the pipeline is, and it takes information—not only from our inputs but the inputs from other companies—and makes determinations,” he said, such as whether a contractor is a “heavy hitter” who has a history of damaging pipelines. “It makes a correlation between the placement of the asset and the person calling in the 811 ticket, and it starts learning to tell you that you have a higher risk on this ticket than you do on another one. And that ends up being where we should put our time and attention, oftentimes leading to a field visit, a call or both.”
The calculations are even more complex because each state has its own regulations, and NiSource has a service territory that spans Indiana, Kentucky, Maryland, Ohio, Pennsylvania and Virginia. AI can account for those variations, along with all the other characteristics, and determine how to proceed within eight minutes of a ticket coming in, as opposed to the hours it would take a person to do the same task, Mojica and Maynard said.
“It takes an army of neural network components working together very quickly to deliver that in eight minutes for me,” Mojica said. “That’s where AI is powerful. I could probably come up with some of that. But how do I do that in eight minutes, and how do I do it in a way that can be trusted? That’s AI.”
However, the processes do not completely remove the human element, Maynard and Mojica pointed out. If, for example, NiSource gets two high-risk tickets, well-trained staff members must determine which one needs immediate attention and a field visit to reduce the risk of damage. “There’s a human element, but we want the starting point to be the model,” Mojica said.
Consumers Energy encourages workers throughout its organization to consider how to put AI to use, said Hale. For example, he and others at Consumers Energy have recognized how AI can analyze massive amounts of data—such as complex regulatory documents—in a matter of minutes, highlighting relevant facts and information that might take hours for a person to comb through. “We can utilize these AI services to go through the document and find key points to focus our attention and then apply more of our ability to do some research and respond to those changes, versus trying to read through all of that,” he explained.
As NiSource started developing its models to improve operations about four years ago, Mojica and Maynard said that the utility immediately involved the employees who work in the field. NiSource has made significant investments in various tools such as geographic information system, or GIS, data and mapping software. These efforts, along with the utility’s data governance strategies, have resulted in the continuous improvement of records and maps, which took a commitment by employees throughout NiSource, Mojica and Maynard added.
“The biggest proponents of this tool have been the people who are closest to our facilities—the people that are closest to our work,” Mojica said. “We engaged them. They became the founders. They became the ones who understood it, and we let them speak up. If you get it in front of the people who do the work and you show them the facts, it’s hard to run from.”
Front-line workers also offered proof that the technology was working by using it to help them do their jobs, Maynard said. Worker feedback is shared to the software vendor to incorporate during periodic updates. “It is evolving. It is not static,” she added.
While the NiSource software is proprietary, Mojica said companies should seek the right software vendors to create strong partnerships to develop the systems that work best for them.
“I think utilities are often resistant to go down this path,” he added. “But we switched because we said, ‘We have data that is consistent and continues to tell us more, so why don’t we better use and leverage that data?’ That provides the next level of thinking and input we need.”
In terms of talking to vendors, Hale said, various companies will pitch different ideas and solutions. His goal is to ensure that decisions to use AI at Consumers Energy are thoughtful, so that technologies are not deployed unnecessarily. Decision-makers must also focus on how to truly understand a problem and the processes before jumping in with AI. “These tools are very powerful, and we have to find the best ways to be more efficient in our day-to-day activity,” Hale said. “We can make the right choice from there.”
Another issue that industries are grappling with across the board is the responsible use of AI. “A big piece of that is the governance portion,” Hale said, with AI deployments necessarily factoring in benefits to both the business and the customer. “It’s very important that we as an organization—and even as technology experts—understand those impacts and that we have mitigation steps in place to reduce the risks.”
He continued, “AI is a great tool to use, but it doesn’t always have to be the tool. We want to proceed, but proceed with caution, and try to take the time to understand exactly what’s being put in place and truly understand what the AI technology is behind the scenes, what model it may be using and how it is interpreting the data.”
AGA’s Meyer agrees. It is critical to understand the technology’s potential. “We really want to understand on a fundamental level what these tools are doing and what they are good for and also things we need to be watching out for because, like with any new opportunity, there also are risks,” Meyer said.
With ChatGPT, for example, he stresses that users must closely monitor the work for accuracy and clarity. But he said there is unlimited potential, analogous to the advent of the internet.
“There’s just an open world of possibility,” Meyer said. “We really don’t know how this is going to evolve. We’re at the frontier of it, and I’ve heard the use of the phrase ‘jagged frontier,’ where AI is actually really good at some things and not very good at others. We’re still learning about what and where those boundaries are. And of course, as we learn about it, the boundaries are shifting underneath us in the moment.”
Maynard and Mojica said they are aware that they need to ensure everyone is using good data.
“One of the things we’ve realized is that the better our data gets, the better we are informing the model,” Mojica said. “It’s the old adage of ‘junk in, junk out.’ We make very conservative assumptions in the model. But as we get better, and the models continue to learn, the sensitivity continues to increase. And we believe that sensitivity is going to allow us to make better, more informed decisions.
“And the data has proven itself,” he added. “We are making better decisions, and it’s showing better results to reduce risk.”
Reducing risks, after all, is a primary goal.
“For a utility that has to understand safety, quality and customer affordability, it’s essential that every dollar we spend goes into the best place where we are going to reduce the most amount of risk,” Mojica said. “We have built an organization that really targets that.”
NiSource knows that the number of 811 tickets will continue to increase, with requests rising 4% over the past year alone and into double-digits in some states. Some of the increase is due to growth in industries such as the telecom sector but also because of public awareness campaigns to get people to call 811 before they dig.
For companies that haven’t fully explored AI, Mojica suggests that they take the leap.
“I would tell them that this was bleeding-edge technology when we started, and what we found was it’s not that anymore,” he said. “People can come and look at what NiSource has done and build upon those solutions. A rising tide lifts all boats. We want every utility to rise with us.
“The only thing we ask is, when you make it better, come back and tell us how you did it, so we can learn from you,” he said. “Our goal has been to place the right tools and technology at the fingertips of our employees, so we can drive safety in damage prevention throughout our industry.”