By Stefan Engelhardt, James McClelland, and Stacy Collett

Providing vital power to the world used to be a pretty simple business. One centralized company generated electricity. It delivered that electricity through wires. And then it sold it to millions of customers. Usage was predictable for the most part. There was very little interaction between utility and customer, except for occasional meter readings or customer phone calls to report a local power disruption. Customers had few options and little recourse if they were unhappy with the service or the cost.

No more. Take West Penn Power. It has been providing electricity to more than 720,000 customers in Western Pennsylvania for more than a century. To comply with the state’s Alternative Energy Portfolio Standards Act, which requires electric distribution companies to offer customers a percentage of electricity from alternative resources, West Penn Power now has contracts with about two dozen wind, solar, and hydropower energy suppliers—as well as consumers who generate their own solar power and sell it back to the company. That presents a lot of challenges for the traditional utility—from collecting energy from that growing fleet of suppliers, to determining how it’s consolidated and distributed, keeping the line voltage in balance, and then paying the supplier for the energy used. It needs good models able to forecast, predict consumption, and react in real time to adjust the energy supply when necessary.

Even interacting with customers has become more complicated, with so many energy options for them to choose from and various rates for each. Consumers now charge their electric vehicles from home or subsidize their power usage with solar panels that must be monitored. Businesses, too, may install their own solar arrays, or contract with third-party providers. All these solar power producers are now active market participants that affect the way energy is delivered. Multiply that by hundreds of thousands of customers, and that affects the utility’s scalability, network control, and capacity management. Utilities need to recapture that predictability, to understand the ebb and flow of energy and effectively manage their business now and in the future. And it turns out that artificial intelligence has the potential to address this need.

Generative AI as utility player

The next generation of AI, known as generative AI (GenAI), has the potential to address many of these challenges and more—and it’s already making inroads in the utilities industry.

The utilities industry has been using other forms of artificial intelligence for nearly a decade and will continue to do so. Computational AI, which creates algorithms that can learn from data, and machine learning, which focuses on interpreting patterns and structures of data to enable learning, are widely used for predictive maintenance and other utilities operations.

2024 Conference for Utilities Leaders

In a world that demands smart and sustainable practices, industry professionals will explore how their teams and business can enhance the consumer experience while delivering efficient and lean end-to-end operations at the SAP Energy and Utilities Conference, to be held April 23-25, 2024, in Seville, Spain.

Learn more about the event

Generative AI uses large language models (LLMs) to create new text, audio, images, or videos using myriad data sources and bring this understanding into a business context. The LLMs provide a revolutionary way for end users and customers to interact with technical systems using natural language.

A utility company in Germany, for example, is using generative AI to support its business transformation. It was able to take nearly 100 regulatory and business documents, asset standards documents, and system operating procedures; and then create business requirements, storyboards for orchestrating process changes, and business value charts for financial analysis—all without human interaction. A process that would have taken months could be completed in days.

Generative AI offers value in the field, too. A utility maintenance worker, for instance, could ask a generative AI bot how to fix a specific piece of equipment they are examining on site. After combing through data, the bot could provide instructions in natural language, easing access to expert knowledge.

Utilities are eager to turn generative AI’s potential into practice. In addition to field service improvements, training, and customer service applications, leaders see system management potential.

Combining generative AI with reliable data from IT systems across an enterprise could enable new capabilities for a range of roles: service agents who gather the information a customer needs about their energy use in real time, and maintenance workers who troubleshoot system problems in the field. It could also help customers who want fast, easy interactions with their utility provider.

Generative AI could also help solve broader challenges that utilities face—such as dealing with an aging workforce. Using LLMs and intelligent copilots or assistants, GenAI can play a key role in educating new employees, reducing training time, increasing efficiency, and making it easier to onboard a new generation of workers.

Utilities are eager to put generative AI’s potential into practice. In addition to field service improvements and training and customer service applications, leaders see system management potential, from making risk-based equipment replacement decisions and improving maintenance operations to orchestrating vegetation management and load balancing energy supply and demand.

And contrary to the industry’s traditionally careful approach to new technology, a large percentage of utilities are looking to become fast followers of AI adoption.

Some 95% of utilities and energy companies surveyed by Capgemini Research Institute said they have discussed the use of generative AI in the past year, and one-third of utility and energy companies worldwide have begun to pilot generative artificial intelligence—algorithms capable of generating text, images, computer code, and other content—in their operations, according to the survey released in October 2023. Almost 40% of utility and energy companies have established a dedicated team and budget for generative AI.

But realizing value from GenAI requires first building an analytics strategy, putting systems in place to allow data sharing, and addressing the security concerns that come with GenAI.

Turning potential into practice

“Utilities are still experimenting with generative AI for the most part, and it will be some time before they adopt across the enterprise on this technology,” says Pankaj Guglani, strategic business unit leader, energy and utilities at Cognizant North America.

Still, utilities see an ocean of opportunities for GenAI to solve its many challenges, and they’re eager to wade in. Most are starting with quick tasks, applying GenAI to HR processes and business functions before moving toward key core processes. For instance, utilities are now procurement companies—dealing with hundreds or thousands of independent power producers. As these producers generate more energy, they’re signing purchasing agreements with the utility. These are very complex documents. GenAI interprets these documents and summarizes their meaning while simplifying the language. “GenAI is extremely productive in that use case,” says Somjyoti Mukherjee, global leader for energy and utilities at Cognizant.

GenAI researchers point to some other practical use cases, discussed below. It’s worth noting that some of these functions apply to other industries, like customer experience (in retail and financial services) and maintenance operations (in manufacturing). Others (like vegetation management and pipeline maintenance) are more relevant to power companies and other utilities like water suppliers.

Customer experience

Customers have energy choices today, and they want to save money. They also demand more real-time information about their power usage—how much it costs them to charge their EV fleet (or battery in the case of a single consumer) or how much energy their solar panels are producing. This is where GenAI can be particularly helpful by gathering this data and proactively advising customers on energy usage in predicted off-peak hours, so customers can charge those EV batteries at a lower rate and avoid sticker shock. This is important as utilities compete for customers.

Improved maintenance operations

Utilities can use GenAI to improve maintenance processes. “Plant engineers have been using artificial intelligence and machine learning in asset maintenance for several years, especially when they do load analysis and for predicting whether a piece of equipment will pass or fail a maintenance test—that’s already happening,” says Jing Wu, principal research director at Info-Tech Research Group in Canada.

Italian gas and electric company RetiPiù uses an intelligent predictive maintenance application that can predict abnormal performance … and then automatically create maintenance work orders.

For example, Italian gas and electric company RetiPiù uses an intelligent predictive maintenance application that can predict abnormal performance, such as leakages or malfunctions of gas cabins, and then automatically create maintenance work orders showing maintenance schedules and work order status.

“The game changer is when you add on generative AI. For example, once you’ve developed predictive asset maintenance, you can add on GenAI to do prescriptive maintenance—so the instruction and the recommendations on how to repair a piece of equipment could be more interactive and intuitive—as if you’re talking to a co-worker discussing what’s the best way to repair this transformer,” Wu says.

When it comes to risk-based replacement decisions, GenAI models can serve as intelligent companions, providing operators with comprehensive information to help with decision-making on how to prioritize network replacement projects based on customer risk, environment, and health and safety impacts, Guglani says.

Vegetation management

Vegetation management is “one of the biggest use cases of GenAI that I’ve seen with electric utilities—it’s a high operating cost function,” Wu says.

California’s Dixie Fire in 2021 that burned nearly a million acres, the state’s second largest wildfire to date, was caused by a power line striking a tree. Managing vegetation around powerlines and the above-ground transmission and distribution networks is a time-intensive and manual process. It’s also expensive. California’s Pacific Gas & Electric Co. (PG&E) spent approximately $2.5 billion over several years to trim vegetation near power lines. That costly method was not effective and the utility has shifted to investing in powerline equipment that can shut off power lines when it detects objects (like tree branches) touching them, the Wall Street Journal reported. The company will continue to prune trees in certain areas.

Burying cables underground is seen as another answer to the problem, but that fix is slow and expensive to carry out, while vegetation risk still needs handling in the interim.

Smart methods could be used for efficiency and productivity in this area. An AI-based classification service of point clouds, which are simple 3D models created from thousands or sometimes millions of individual measurements (known as points) of an object, can detect the extent of vegetation near power line conductors, towers, or poles.

This data can be used to analyze the distance between these to quickly identify where interventions are needed to remove vegetation within safe distances. And with mass capture of 3D data now readily available from aerial, drone, and helicopter light detection and ranging (LiDAR), as well as ground-based laser scanning and other point cloud sources, this data is ripe for smart analysis to aid vegetation management.

Vegetation management “is quite a complex, costly operation,” Wu says. “Utilities are already leveraging satellite imagery combined with machine learning predictions of growth patterns of some types of trees, then they do some smart scheduling—improving real-time resource planning and scheduling. Then they plan their resources to get the work done in the most cost-saving way. [Generative] AI optimizes this process by combining all the data from the imagery, maintenance, and scheduling systems.”

"Having a well-thought-out data strategy aligned with your organizational business strategy will help accelerate the deployment of AI."

—Jing Wu, principal research director at Info-Tech Research Group in Canada.

Risk detection through analysis of video and images

AI-enabled systems can be used to process a vast number of videos and images to detect potential defects in supply lines and create corresponding plans required to mitigate risks.

Utilities already take advantage of drones and high-definition videos, and they’ve collected so much unstructured data that it’s impossible for humans to process these in a reasonable amount of time, Wu says. An AI model can augment human expertise. It can be trained with GenAI to understand this unstructured data and identify some type of defect, such as cracks in an oil or gas pipeline. The trained drone does an initial assessment scan, and then humans do analysis and look at the image to identify whether it is truly a defect or will soon become one. This can reduce maintenance costs while maintaining reliability, Wu says.

Charting a path for generative AI initiatives

The first step in applying AI is to develop an analytics strategy and ensure that data is in good shape. Many utilities don’t have a proper strategy to deal with the data they have, and it’s one of the reasons many utilities will struggle to quickly benefit from AI. Inaccurate data leads to inaccurate results. Utilities should invest time in collecting, managing, cleaning, and maintaining truthful data. So, for instance, if the utility doesn’t have data on all its customers or has multiple entries for a single customer’s name, then that data needs to be fixed.

“Having a well-thought-out data strategy aligned with your organizational business strategy will help accelerate the deployment of AI,” says Wu, of the Info-Tech Research Group. “Also, lay out your road map of how to enhance your data capabilities, including both structured and unstructured data.”

AI applications need reliable and responsible access to this business data, as well as quality control for the results generated by the AI applications, to ensure data remains secure and meets any regulatory privacy requirements. This can be done using federated learning, and method to machine learning that uses decentralized data sources and allows models to be trained collaboratively across devices while keeping data localized. This way, utilities can build centralized AI models while meeting data privacy compliance regulations. It also generates insights from devices at the network’s edge, and because it doesn’t need to connect to a central server, the model is “trained” faster. Not residing on a central server also lowers the risk of data being targeted or compromised.

Even with such safeguards, it is important to acknowledge that generative AI is still in its early days of implementation. The world is still learning how to control the outcomes of these LLMs. That makes it important to install additional checks on system outputs. For instance, it pays to confirm that the results offered by a generative AI digital service assistant are possible to act on given current business practices and that they make sense given business strategy.

Still ahead: consensus with regulators

Utility regulators know the benefits of computational AI. For instance, AI optimizes operations on the power grid by analyzing real-time and historical data from advanced sensors, communication technologies, weather patterns, and trends to help predict demand. It can also make decisions exponentially faster than humans about when to switch lines to maximize power flow and help maintain grid stability.

Generative AI is a new and exciting advancement for utility regulators too, but their primary focus remains on reliability, safety, and operational excellence.

It is incumbent on utility leaders to educate regulatory bodies about the benefits and value of AI, Guglani says. “But change takes time,” he adds. Utilities have to influence the regulatory policies and work with the regulatory boards and help them understand what generative AI is, what it means for the industry, and how it will benefit utilities, Guglani says.

While its potential is impressive, generative AI is not a wizard; its potential must be tested. Start by looking at current roadblocks that hinder work from being as efficient as it should be, and then look at AI processes as a possible answer.