Why Schneider Electric Integrated OpenAI LLM for Sustainability Management

Enterprises know they need more intelligent and accurate methods of measuring and reporting greenhouse gas (GHG) emissions, but a good chunk of executives admit they don’t understand what’s meant by scope 1, scope 2 or scope 3 emissions. In a move to help organizations improve understanding of sustainability issues and meet environmental targets on time, Schneider Electric integrated generative artificial intelligence (AI) with its cloud-based Resource Advisor platform to simplify emissions data management and provide sustainability context to decision makers.

Recent research conducted by Reuters and GE surveyed more than 550 industry experts and found 59% agree they could improve their emissions measurement and reporting processes. And more than half of those surveyed noted precise data is necessary to progress toward net-zero emissions targets.

Schneider Electric’s Resource Advisor Copilot, based on an OpenAI large language model (LLM), took just four months to move from an idea to a tool ready for beta testing. “It’s been a whirlwind,” Schneider Electric Senior Data Manager Jeffrey Willert said. Copilot will be available later this year or early next year for the 2,500 customers currently using the vendor’s Resource Advisor platform, on top of which the genAI tool sits.

Jessica Kipper, senior director of software project management at Schneider Electric, told SDxCentral  that the platform is unique in the way it collects data from customers and pulls it together to “drive deeper analytics and deeper insights around the consolidation of that information,” she said.

“We have all kinds of ways for data to enter the system,” including smart meters at facilities, utility invoices and client data submissions, Willert added. But “the real heart of this is being able to understand the user’s query and then rely on other tools we built internally to get access to the data,” he said.

Questions that Copilot was built to answer include the following:

• What were my total emissions across all facilities in 2022?
• How much natural gas did we use last month?
• How much did my energy usage change from June to July?
• How much have I spent on energy so far this month?

Given Schneider Electric’s global operating status, “we have the data and the information that we can integrate with those LLMs to tailor that answer to the industry application,” Willert said. “Having an emissions calculator is great, but it’s meaningless unless you can act.”

With the Copilot tool, the sustainability and energy information a company needs to take meaningful action is “all right there,” Willert added. “Not everybody is an expert at staring at dashboards, but being able to extract that data and then being able to do so many different things in such a short period of time” is what makes this solution stand out.

Securing customer data

Copilot is based on a Microsoft Azure OpenAI integration, and Schneider Electric claims it didn’t do any fine-tuning or training of the LLM. That might “sound lazy, but it’s actually very intentional,” Willert noted. To secure both the company’s proprietary information and client data, Schneider Electric handles that data “behind the scenes, away from the LLM,” he explained.

The genAI tool also has built-in guardrails to ensure clients only ask questions about energy and sustainability topics and that the model cannot be exploited, Willert said. Plus, Copilot uses a technique called retrieval augmented generation (RAG) – an increasingly popular method in the genAI scene – to feed data to the model at runtime. This ensures client data “always stays within our control,” he said.

To view the original interview posted in SDX Central's newsroom, click here.