Powercor to tap into agentic AI across the organisation

Victorian electricity distributor Powercor is incorporating multiple AI agents into processes to improve access to information and the work of its contact centre and field force staff.

Digital R&D manager Peter McTaggart told the recent AWS Summit in Sydney that the company is 18 months into its adoption of AI, beginning with generative AI before moving more towards the agentic AI space.

Its initial work was to help non-technical users understand what they were looking at in asset images captured through drone-based inspections.

“We wanted to give non-technical users a natural language interface to some of the technical documentation to help them understand what they were seeing in the images,” McTaggart said.

“We were trying to reduce false positives that non-technical users were raising as defects on these images and that proved out at the end of last year.”

The interface that was developed is called Autofocus, and it is built predominantly on an AWS stack that includes Amazon Bedrock.

With similar information retrieval use cases popping up in other parts of the business, Autofocus was expanded “into a more general deployment where we could load up a lot more documents”, McTaggart said.

About the same time, the frontier of the enterprise AI wave was transitioning from generative AI to agentic AI – and Powercor was keen to make use of AI agents.

Powercor started with a “basic” call analytics agent that could check phone calls for compliance with call centre scripts, offer feedback to operators on the conversations, or perform some sentiment analysis.

It then turned its attention to implementing AI agents in “more complex workflows”.

McTaggart noted that prior to agentic AI, the company would typically use “some sort of business process tool or workflow tool to try and automate a lot of these tasks.”

One AI agent is being used to generate responses to email inquiries from customers, so that staff do not have to write each response “from scratch”.

Another use case for AI agents is to run compliance checks to ensure that field plans are created correctly.

The company creates around 16,000 field plans a year for construction, management or maintenance activities for its distribution network.

“Particularly around the field plans, there’s lots of low value work, and so we’re looking to take a lot of time out of these processes of people clicking around and looking at maps and looking at images and looking up documentation to really streamline these processes,” McTaggart said.

Just like generative AI before it, McTaggart said that there is considerable interest in other parts of Powercor to make use of AI agents in process-intensive workflows.

“There’s a huge amount of demand in the organisation for agentic AI,” he said.

“It has the potential to enable us to improve efficiency and productivity dramatically.

“While we started with it in a small corner of the organisation it is spreading, and [management] are keen to see us move this out right across the business, which is the focus for the next six to 12 months at least, if not longer.”