California public power utility SMUD has been implementing AI to advance both its customer service and internal business management.
In a conversation with APPA, SMUD Director of AI Oliver Daniels III and Director of Specialized Enterprise Initiatives Michael Champ detailed how the utility is exploring use cases that have enabled a newfound level of process efficiency.
Building on Innovation
SMUD had been exploring the use of machine learning and other data analytic capacities prior to the availability of tools like ChatGPT, with Daniels’ team having done foundational work supporting forecasting energy usage in the lead up to the wider commercial release of AI platforms.
GenAI didn't come out until 2022, “but we were looking at machine learning for quite a while because we have customer usage data and were always interested in customers being able to evaluate their energy usage and giving them tools to do that,” Daniels said.
This prior work exploring machine learning facilitated SMUD’s adoption of GenAI for customer data evaluation, which included internal personnel shifts such as Daniels being appointed the utility’s first director of AI.
“GenAI comes out in 2022, and the team starts exploring how we can use it from an innovation enterprise angle. After that happened, my position got created in August of 2024,” Daniels said.
SMUD quickly moved to create an AI governance council focused on how to apply these new capacities in ways that were both responsible and conducive to achieving the utility’s goals.
“One of the first things we did was establish an AI policy in mid-2024. We had a dual mandate to understand and manage the risk the AI brings on one hand and promote the usage and adoption of AI technologies on the other to ensure we get those benefits while mitigating the risks,” Champ said.
SMUD spent the latter half of 2024 mapping AI use cases, noting the areas of business operations and customer service that could benefit from a regimented use of large language models.
“It can be used for simple things like reviewing emails, but also for researching technical documents, drafting and reviewing project and procurement documentation and processes and procedures, making charts , and writing code,” Champ said.
Mapping Load Growth
Among other use cases, SMUD’s AI team has been deploying these tools to better forecast load growth.
The data processing and analytic capacities allowed by AI have refined and facilitated the utility’s grid management, particularly when evaluating complex data sets alongside each other – such as weather patterns against customer energy usage.
“Our load forecasting uses hourly meter data from our smart meters and even 15-minute meter data in some cases. They're creating load shapes and forecasts and using machine learning models like gradient boosted decision trees to understand customer behavior,” Champ said.
SMUD’s usage of AI for load forecasting has allowed the utility to incorporate a variety of data sources within its evaluation models, applying both smart grid inputs alongside economic and structural variables that are used to map an array of possible outcomes.
“We have to layer in expectations of economic growth – are we in a boom or bust cycle… We'll also forecast by customer types. What’s residential doing, what's small commercial doing, what's residential with solar doing versus without solar,” Champ said.
This has made it easier to create a general load growth model and append demand from individual sources, such as electric vehicles, resulting in a dynamic load mapping the utility continues to build upon.
“We can take that model and add layers for things like EVs to show what to expect based on historical evidence. So, you can say - here's where we think the EV adoption rates are headed and based off of the existing EVs that we have, here's how much load they have, and what shape they have. You can take your baseline and forecast with those on top,” Champ said.
Internal Process Improvements
In addition to increasing the accuracy and sophistication of SMUD’s load forecasting capacities, the utility has also begun using AI to streamline internal business processes.
Notable among these has been using AI tools for document review and information recall, automating in seconds what previously would have taken an exhaustive process of searching individual documents for regulatory statutes and other filings.
“RAG - retrieval augmented generation - where we're uploading a set of documents to a library and plugging the AI into that library allows us to easily look up policies and procedures. Instead of looking through all the policies, you just go ask Policybot,” Champ said.
AI-backed analytic tools have also made it possible to measure and identify the specifications of grid components by picture analysis alone.
“We have visual processing models as well. You can basically take a picture of the nameplate data from a transformer and feed it into the AI and it will identify parameters such as size, weight, and voltage,” Champ said.
These innovations ultimately tie into two foundational goals – providing the best possible services to SMUD’s customers and supporting staff while streamlining workflows. This has even taken the form of a sentiment analysis tool SMUD Is exploring that will gauge feedback from customers to evaluate their satisfaction with utility services.
“We had a demo with the AI team today on a sentiment analysis tool that was built to monitor customer sentiment from calls into our call center. We’ll be using that with GenAI to determine if we’re receiving positive, negative, or neutral sentiment. We’re going to continue refining that, and then see what value it can bring to the business.,” Daniels said.
