Suncorp will put five AI agents into production for insurance claims handling by the end of June, making parts of its claims process agentic and far more automated than the insurer has disclosed before. The agents will take on distinct sub-tasks inside the existing claims workflow, from initial report capture to settlement calculations.
Kranthi Nekkalapu, AI practice executive at Suncorp, described the move at the Databricks Data+AI Summit in San Francisco. "Claims processes are inherently very complex, but at a high level, they are workflows," he said. "The workflow remains the same, but within that, there could be several tasks. These tasks themselves could be done by agents."
The first five agents are set to handle intake validation, claim classification and routing, coverage checking, vendor dispatch for repairs, and settlement calculations. Two of them work in the first-notice-of-loss stage where a customer first reports an incident. At least one is a voice agent that has already gone live. "They do the initial intake of the claim in some cases, and if it becomes more complex, then pass on to humans," Nekkalapu said.
How the agentic workflow is stitched together
Suncorp models the end-to-end process with business process model and notation (BPMN) and orchestrates everything through UiPath. Each task inside the workflow can be handled by a single agent, a system of agents, or a nested workflow. This design keeps the stages familiar - intake, triage, assessment, fulfilment, settlement - while delegating repetitive or rule-based decisions to software.
This approach aligns with broader AI Agents & Automation strategies now entering regulated industries. Insurers that master such orchestration can shorten cycle times without giving up human oversight on complex or borderline cases.
Watching every agent move
Suncorp built a central observability platform on Databricks that ingests OpenTelemetry-based traces from every agent and sub-system in real time. Nekkalapu said observability is "really, really key" because without it, a "very fancy system" counts for nothing. The company also assembled a library of roughly 100 controls tied to different AI applications and risk types.
The dashboard serves multiple audiences. Executives see program-level metrics. The safety team checks whether risk controls are working. Business users can drill into why a claim decision was made. Engineers get debugging access and a Databricks Genie-powered chatbot that lets them ask questions and receive immediate analysis. "A chatbot could be the first level [of support] where they would ask questions, and they would get the answers and all the analysis in real time," Nekkalapu said.
Coverage decisions at 99 percent accuracy
One of the harder problems Suncorp is tackling is AI for Insurance coverage determination - figuring out exactly what caused a loss when multiple events interact. Nekkalapu described a scenario where a storm fells a tree, the impact damages a pipe, water leaks into a garage, and carpets and rugs are ruined. "It's not an easy problem," he said. Claims handlers typically spend significant time untangling such chains.
Suncorp has built a workflow with agents that handle this classification and coverage analysis. "We've developed a workflow with agents in there which can do all these jobs with 99 percent accuracy," Nekkalapu said. Human reviewers still stand as a mandatory gate for any claim an agent recommends rejecting, and intake agents hand off to people when a claim gets too complex.
Why this matters for insurance professionals
Five agents moving into production this month signals that claims automation is moving from pilot to real workload. For claims handlers, team leads and operations managers, the immediate impact will be a shift in the type of work that reaches their desks: fewer routine validations, more judgment-heavy cases and oversight of automated decisions. Familiarity with how agentic workflows are monitored and controlled will become a practical skill, not just a technology topic.
The Suncorp design also shows that even partial automation requires serious investment in observability and controls - an alert for any insurer planning similar moves without a central way to trace every agent decision in real time.
Your membership also unlocks:
