AI to Improve Affordability at Electric Utilities

Abstract

While much more T&D will need to be deployed in order to meet growing electric load, that utility CapEx will be increasingly borne by LIDAC customers as wealthier customers reduce load via DER and efficiency investments. The single best way to improve affordability for ratepayers is to provide the same energy services with the least amount of incremental T&D and the most advanced grid orchestration. Utility partnerships with AI and ML technology companies have started to squeeze more out of this growing infrastructure, but some AI solutions are window-dressing while others are transformational. Utilities must conduct a clear-eyed assessment of existing and burgeoning technologies, focusing primarily on those that address the biggest cost centers of the utility in coming decades, starting with the CapEx associated with the >100,000 miles of T&D that will need to be installed in coming years for a true net zero transition. AI can widen the path toward that future. And where solutions don’t exist today, utilities have a timely opportunity to partner with technology companies to co-develop those solutions.

The Electric System Will Be Getting Much More Expensive

Electric load is expected to increase 2-3X in the next 2-3 decades after remaining effectively flat for decades. Unfortunately, utilities and their supply chains are not in shape to meet this growing demand after such a long period of demand stability (case in point: the great transformer shortage). Need a new EV charging hub? That’ll be 5-7 years for a new substation. Need a new 100+ MW data center? That’ll be 10+ years for upgraded transmission. This increase in load is occurring at the same time Distributed Energy Resources (DERs) will stress a grid built for one-way power flow that now requires dynamic grid orchestration at the distribution level. This all will require a level of CapEx that hasn’t been seen in decades, if ever, to move power from where it’s generated to where it’s needed in real time.

Lower-Income Customers Will Disproportionately Bear Those Costs

One path to meeting the demands of the new load is to simply put more steel in the ground and triple the current transmission capacity, which most estimates suggest will be required if we rely significantly on getting wind and solar from sources of new generation to load. But as wealthy utility customers increasingly add solar on their homes, lower-income and disadvantaged community (LIDAC) customers will be stuck paying utility bills that are already at $0.45/kWh in California and elsewhere - and could increase to $0.70/kWh and beyond by the end of this decade if wealthy customers continue to consume fewer kWhs while relying on the grid for resilience. While we may need (much) more transmission and distribution (T&D) infrastructure, a brute force “put more steel in the ground” solution alone will exacerbate the growing inequities in utility costs.

Hundreds of Solution Providers Claim AI as The Answer

Hundreds of startups now claim to use AI to solve utility challenges ranging from T&D line optimization to predictive maintenance to customer support. 

Of the 150 companies in the Indigo Advisory Group GridTech landscape, Realize2050 estimates that over 90% claim to leverage AI in their processes. While we can be confident that nearly all of these companies are using AI in some fashion, a non-trivial set of their claims does not match the degree to which AI is central to their product offerings.

Indigo Advisory Grid Tech 150

Many utility executives are rightly asking today: Which of these technology companies (and the hundreds of others purporting to use AI to support utilities) are truly offering solutions capable of deferring infrastructure investment? To assess that question, utilities must conduct a rigorous evaluation of the landscape to determine which startups are using AI as window dressing rather than as a core solution to a demonstrated problem.

The Best Utilities Will Seek AI Solutions That Address the Biggest Cost Centers and Demonstrated Problem Statements

There are two other central questions that utilities should be asking:

1) Based on the utility's largest cost centers—particularly those expected to drive increases in customer rates—where could AI address those issues in ways that commercially available technologies currently do not?

2) Based on problem statements that forward-looking utilities have outlined, where could AI address those issues in ways that it currently is not? 

Once those gaps have been identified, forward-looking utilities should look to co-create solutions with innovators. If those solutions haven’t yet been developed, it may be because of the lack of coordination and collaboration between utilities and startups. Forward-thinking utilities could and should advance beyond simply providing specs for a desired solution, but actually working collaboratively to create a solution, testing it, and sharing in the upside from development. PG&E has shown the way with its R&D Strategy White Paper, written in collaboration with the Realize2050 leadership team, by outlining 68 problem statements it sought help from the innovation community in addressing. 

The opportunity here for a given utility in co-creating AI-based solutions is to guide true innovators towards solving the utility’s own version of a given problem, rather than a “generalized” version. These pseudo-customizations can be feature-based - e.g., integration compatibility with specific MDMS or GIS platforms - or even structural - e.g., training a computer vision model using images of the most relevant infrastructure samples - so that value can be derived quickly and with high confidence upon adoption. Importantly, what follows from this collaborative process is the sufficient internal knowledge and capabilities to continue to guide and derive value from any newly adopted solutions.

You’ve heard the nerdy adage “garbage in, garbage out”. This has a deep and complex meaning as it pertains to utility-oriented AI solutions; utilities sit on increasingly massive piles of data but have been slow to systematically utilize it meaningfully. Effective use of the shiniest new AI tools begins with a solid IT infrastructure posture. This includes secure, robust, and scalable database practices; exhaustive data quality and cataloging efforts; integration and data-sharing capabilities; and most importantly, all of the knowledge and human capital to manage and develop new IT capabilities. At Realize2050, we argue that these ingredients are prerequisites for sufficiently leveraging new state-of-the-art AI solutions.

Too often, utility innovation groups seek to solve “fun” problems rather than “big” problems. The robust landscape of AI (and faux-AI) solutions makes it all too easy for utility innovation groups to be solutions-focused rather than problem-focused. An analysis of the biggest cost centers—both those on the OpEx side and the CapEx side—should inform where to focus first. If steel in the ground is the single biggest rate-inflating cost over the next few decades, that seems to be the ideal first place to look for efficiencies from AI. 

Where AI solutions are insufficient today, utilities would be well-served by co-developing solutions with technologists while sharing in the upside of being that valuable thought partner and first customer. Even more ambitious still, utilities could partner with AI and ML startups to ideate the next technologies to lower the expense lines of those major cost categories that drive executive bonuses and LIDAC customer rates.