Suncorp turns to multi-agent AI for business transformation

Suncorp Group is planning a multi-agent approach for its AI strategy, and is building out a library of reusable components that will help it to scale the technology across its brands.

The insurer, which is behind brands including AAMI, Gio Insurance and API, revealed that it had commenced with plans to transform its customer service and lodgement processes in October.

More recently, at Databricks’ 2025 Sydney event, the group’s chief machine learning engineer Touraj Varaee said that, having already used AI to save thousands of work hours and generate more than a million words in case summaries, the business case for expanding its AI use was solid.

“All of this shows us we are very good at building the AI enterprise … but to be honest, our perspective for the future is [that it’s] not just about having a single high value and individual AI application.

“It's mostly about of the how these applications or AIs and agents are working together and able to collaborate with each other and how we'll be able to build a complex multi-agent systems,” Varaee said.

Accordingly, the group has adopted a system architecture comprised of Databricks suite of products that Varaee said is tailored to its multi-agent approach.

The suite consists of Databricks’ Lakehouse to manage the data and Unity Catalog for governance.

The architecture provides the underpinning for Databricks’ enterprise suite for building, serving and unleashing generative AI applications, named Mosaic AI.

Varaee said that starting all over again building new agents for each business case from scratch was not a feasible given regulatory and governance hurdles that the prudential authorities impose on insurers and, by extension, the technology the sector uses.

“The next layer, I would say, is a scalable engine for us. You can imagine if you want to build the agent into some enterprise companies, you cannot start everything from the beginning and build everything from scratch.

“This is not scalable. This is not regulated. Our approach [to] this achievement is to build a layer that consists of all reusable components and services, so that all agents [will] be able to use them [in a] unified and integrated way.

“That consists of the observability layer, reusable agents, components and the agent context memory that provides the stateful capability of … our agents.

"And all of these stacks [will] be able to empower the agents that we are building in the future. This is the beauty of this architecture,” he said.

“Here you'll be able to have the plug-and-play capability. You'll be able to use agents that previously have been built."

Varaee provided an update on Suncorp’s achievements with AI since the insurer revealed its program of work in October.

He said that the insurer’s claims division had used AI to automatically generate over a million words in case summaries

He also revealed that the group’s Retrieval-Augmented Generation-based (RAG) Generative AI Smart Knowledge application had saved its customer service staff more than 15,000 hours of work.

“For us, AI is not just interesting R&D projects. It's the main engine for scaling our responses to be able to process the claimer in [an] accurate way [while] also managing risk,” Varaee said.