Xceedance wants to stop charging insurance clients for hours worked and start charging for each policy issued or submission cleared. Asked where that leaves an Australian broker whose licence sits above the work, Dan Miller (pictured), associate vice president and client partner for Australia and New Zealand at the insurance services firm, said the position does not move.
“Same contract, same service level, same liability, same working relationship,” Miller told Insurance Business.
The pricing change is already underway. Xceedance said in a statement in late September that processes moved onto IAN, its agent studio for insurance, delivered around 35% efficiency improvement from the point of go-live in agency and broker operations, with some implementations now approaching 50%. The company said it is moving away from effort-based contracts towards transaction- and outcome-based ones, and has built a catalogue of more than 120 agents spanning claims, underwriting, catastrophe modelling, finance and accounting and data operations.
Asked what sits beneath the 35% to 50% range, Miller described a time-based measure captured at the level of the individual transaction, with system-captured timestamps recorded across each stage of the work: document read, extraction, comparison, checking against the draft, and human verification. That produces an effort figure in minutes per transaction, which Xceedance compares against a documented baseline.
His own gloss was simpler. Work that once took him 10 minutes is timed, run through the agent and timed again, and the gap between the two becomes the percentage.
Two qualifications sit around the number. The figures are Xceedance's own, drawn from its own deployments, and the September statement named no Australian client and no Asia-Pacific deployment using IAN. Miller also said the model is not AI automation alone, but a combination of agents, human judgment and core systems in proportions appropriate to the process, with the human in the loop critical to decision-making and review as the model learns.
Miller said an auditability feature records what the agent looked at and how it reached a particular decision, with hyperlinks through to the underlying information. The information presented to the human reviewer, and the decision that person then made, is captured in the same trail. He described three layers: audit history, the work behind each agent decision, and the sign-off recording who approved it and why.
Read next: This broker's AI answers the phone and admits it is not a person
Asked whether that would satisfy a regulator examining a decision made on a client file, he was direct about where the obligation lands.
“You can't just say, 'Oh, the AI did it',” Miller said.
Asked separately whether using the agents has any bearing on an Australian broker's own licence obligations, Miller said it should not, and that he did not see it being an issue.
The Australian Securities and Investments Commission (ASIC) has set out a firmer position on third-party AI. In Report 798, the regulator analysed 624 artificial intelligence use cases across 23 Australian financial services and credit licensees as at December 2023 and found the potential for a governance gap, with some licensees adopting AI faster than they were updating risk and governance arrangements.
Third-party reliance was its own finding. ASIC reported that 30% of all use cases in the review ran on models developed by third parties, that 13 of the 23 licensees sourced half or more of their models externally and that four sourced all of them that way. Not all had governance arrangements appropriate to that reliance. The report states that the regulatory framework for financial services and credit is technology neutral, that licensees remain responsible for outsourced functions, and that they should consider how that expectation applies when third-party providers are used at any stage of the AI lifecycle.
Read next: OpenAI admits its AI models have learned to cheat, hide their own mistakes
The review has limits worth noting. Its scope excluded back-office functions, which is where much outsourced policy checking and submission work sits, and its data predates the current generation of agent deployments. ASIC returned to the subject in May 2026 with an open letter to all Australian financial services licensees on frontier AI and cyber risk.
Insurer clients sit under a second regime. The Australian Prudential Regulation Authority's Prudential Standard CPS 230 Operational Risk Management commenced on 1 July 2025, replacing the previous outsourcing and business continuity standards and extending obligations to material service provider arrangements, a shift APRA board member Therese McCarthy Hockey discussed in an interview on the biggest change for insurers under CPS 230. Whether an AI-enabled operations provider running live policy and claims volumes meets the material service provider test is a judgment for the regulated entity, not the vendor.
The coverage side of the question is no more settled, as Insurance Business reporting on AI liability and client policy wordings has set out, with lawyers expecting a dispute over whether professional indemnity or product liability responds when an AI tool goes wrong.
Miller's position throughout was that the arrangement between Xceedance and a client does not change because an agent did part of a task, and that the teammate who approves the output owns it.
“Accountability ultimately relies on you as a human, not the AI agents itself,” he said.
