From pilot to P&L: what Australian insurers need to get right on AI

For Australian insurers, technology is no longer a constraint on the adoption of AI. Frontier capability is broadly accessible, and many insurers already have pilots running across claims, underwriting support, document handling and customer service. The real constraint is the ability to convert those experiments into economic outcomes. AI value realisation remains uneven, and the insurers that scale successfully are distinguished less by model choice than by how they build capability, focus investment, industrialise delivery and govern deployment.

In our work with insurers and other financial institutions, we have observed four dimensions that consistently determine whether organisations realise value from their AI efforts: fluency, strategy, foundry and trust. These are not sequential stages, and they are not independent workstreams. Any one can become the rate limiter on the others, which is why partial progress feels like motion but produces little value.

Fluency is the organisational ability to understand AI, adopt it effectively, rethink work around it and continually evolve the capabilities and workforce required. It is a shared language that lets a board and executive team separate signal from hype, understand where AI changes economics, and set risk appetite and investment criteria without deferring to whoever is loudest in the room. The evidence that this judgement pays is now hard to ignore: PwC's 2026 AI Performance study found that almost three-quarters (74%) of AI's economic value is being captured by just one-fifth (20%) of organisations – and that the leaders pulling ahead are two to three times more likely to use AI to pursue growth and reinvent their business model, and roughly twice as likely to redesign workflows rather than simply bolt AI onto existing ones.

Because so many insurance tasks fall within AI's current capabilities, the sector is more exposed than most other industries. PwC's 2026 AI Jobs Barometer found that the skills needed for the most AI-exposed roles are changing more than twice as fast as those for the least AI-exposed roles, and that the traditional career ladder is compressing. AI-exposed junior roles are far more likely to demand traditionally senior skills such as judgement and leadership. If expertise concentrates in small, experienced groups while automation absorbs the foundational work, the result is a brittle organisation that struggles when models err or conditions shift. Role expectations should now include the ability to interpret and challenge model outputs.

Some insurers are encouraging broader adoption to lift organisational fluency. One large insurer reports that more than 60% of its workforce uses AI regularly, with more than 600 "activators" who have deployed more than 90 AI agents across customer service, operations and corporate functions. Our research shows that real impact comes when leaders are equipped to understand AI and redesign work accordingly. Leaders aren't expected to become AI developers, but understanding the tools at their disposal and using them well is no longer optional.

Most AI strategies fail on problem definition, not ambition. A focus on optimising existing tasks or delivering incremental productivity gains is likely to produce limited returns. These challenges manifest in two gaps we see recur: a paper-to-practice gap, where leaders admire the problem and nothing ships, and a pilot-to-P&L gap, where initiatives multiply without economics or a funded path to scale. PwC's 29th Global CEO Survey found that close to a third (30%) of CEOs reported AI had already increased revenue in the past 12 months, and a quarter reported lower costs, confirming that the value is available but not automatic. Just one in eight reported both at once.

For general insurers, the value lies where differentiating capabilities sit: risk selection and pricing sophistication, claims cost and cycle time, leakage and fraud, cost-to-serve, and retention under affordability pressure. These areas are also broadly in line with the five prioritised AI-adoption use cases identified by CSIRO and the ICA in 2025. In these areas, AI can move beyond internal efficiency to deliver tangible customer value. The simplest frame is to group opportunities by whether they protect value, create value, or multiply it across the enterprise. Value protection covers the activities required to respond to AI-enabled competition, such as margin compression and materially lower cost envelopes. Value creation is the next step in adoption, and involves redesigning work around AI. Value multiplication is where AI helps insurers leapfrog by enabling a step-change in capability, such as dynamic personalisation.

PwC's Reinventing insurance  research highlights the opportunities in the value multiplication space. These include using technology to shift from compensating customers after an event towards helping them avoid loss in the first place and, for life insurers, developing propositions at the nexus of life, health and wealth as an ageing population confronts retirement affordability and coverage gaps. Identifying these opportunities requires a consistent approach: name the value pool, isolate the lever inside it, define the enterprise conditions for scale, and progressively turn those choices into proven solutions and reusable assets through a foundry model.

Pilots stall because there is no repeatable route from proof to production. A foundry is that route: a studio-to-scale engine that proves value in weeks, proves return with credible attribution, and then proves the solution can survive enterprise complexity in policy administration, claims and finance systems, and in regulated processes. Owners, service levels, benefits ownership and stop-go rules exist before the build starts, and winning use cases are industrialised as reusable patterns rather than rediscovered by the next team.

The local benchmark for scale is becoming explicit. One established insurer, working with an affiliated venture business and a specialist technology partner, has embedded fully governed algorithmic lead underwriting directly into its own systems for a complex specialty insurance product. The end-to-end broker journey — from submission through underwriting, pricing, documentation and bind — now completes in under ten minutes for a complex risk that previously took days, with document generation cut from around five hours to seconds, all while the insurer retains full control of pricing, appetite and governance. That is what industrialised looks like: a governed capability that scales a specialty portfolio without adding operational headcount.

Legacy estates are the honest test of that ambition. Decades of claims histories, policy wordings and broker correspondence are a genuine data advantage, but they are largely unstructured and unfit for consumption today. PwC's Reinventing insurance research argues that IT must move from a maintenance function to a strategic enabler that co-designs with the business and is measured on business outcomes, not uptime. Insurers that treat data and platform readiness as a separate programme will keep meeting the same blocker at the same point in every business case.

Responsible AI is both the mechanism that gets initiatives approved and the discipline that ensures they deliver positive customer outcomes. Standard risk tiering, guardrails, rollback triggers, provenance checks and pre-baked evidence turn a bespoke approval argument into a repeatable one, which is what compresses time to production. PwC's 2025 Responsible AI Survey found that nearly six in ten executives (58%) said responsible AI initiatives improve return on investment and organisational efficiency, and a majority said they improve customer experience and innovation.

The regulatory posture has hardened in parallel. In August 2026, ASIC and APRA jointly urged financial entities to move beyond awareness of frontier-AI risk to action, with a "strong, tested plan to respond when the worst happens". ASIC Commissioner Simone Constanthas warned: "The urgency of this challenge cannot be overstated. Threat actors are exploiting frontier AI models to identify and exploit vulnerabilities that previously may have taken a team of professionals months to find." This followed the message in APRA's April 2026 letter to institutions that, while failing to embrace AI may put businesses at a strategic disadvantage, AI also has the potential to create new risks and escalate existing challenges.

The insurance-specific issue is accountability. AI blurs ownership of underwriting, pricing and claims decisions, and boards need a clear answer on who is responsible when a recommendation is wrong, and on how much deviation from model guidance is acceptable. That question changes character as agents move from advising to acting. Where an agent can request information from a customer, order an assessment or progress a claim without a person in the loop, accountability has to be designed before deployment rather than reconstructed afterwards. That means specifying what the agent may do, the limits on its autonomy, the events that force a human decision, and the record that shows which system took which action and on whose authority.

The rigour Australian insurers already apply to actuarial model governance provides a strong starting discipline, but AI extends the challenge well beyond the model: into data provenance and residency, workflow and decision rights, the security and resilience of the AI estate itself, the degree of autonomy granted to agents, and continuous human accountability for what those systems are allowed to do. Much of the AI an insurer runs will not be built by the insurer. It arrives inside policy administration, claims, pricing and distribution platforms, and increasingly as agents supplied by vendors and their subcontractors.

Boards should expect the same evidence from a provider's AI as from their own systems: what data it uses, how it was tested, how it is monitored, what it is allowed to do without a human, and how a failure is detected and escalated. Contractual assurance is not the same as control, and an outsourced decision is still the insurer's decision in the eyes of a customer and a regulator. The insurers furthest along describe this as execution, not overhead: one major insurer, whose approach to AI ethics and responsibility was recognised at the 2025 Financial Review AI Awards, grounds its programme in strong foundations for governance, risk management and AI safety that are embedded enterprise-wide rather than left to the IT department. Trust is not merely a means to gain approval; it is a practical enabler of successful outcomes.

Interdependence is the whole argument. Fluency without strategy produces curiosity and enthusiasm with no commercial edge. Strategy without a foundry produces shelfware and a credibility problem with the board. A foundry without trust produces solutions that queue indefinitely for approval or, worse, ship without the guardrails that protect customers and reputation. And trust without fluency produces compliance theatre, where propositions are waved through because nobody in the room can fully evaluate them. Treat AI as a technology programme and you get tools; treat it as work reinvention, governed properly and led by people who understand it, and you get scaled value.

For Australian general and life insurers, the implication is measured rather than dramatic. The next twelve months are unlikely to reorder the market, but they will separate the insurers building a reinvention capability from those still funding experiments. The ongoing questions for executives and boards are:

• Are we clear on how to mobilise our workforce to capture value from AI?
• Can we name the two or three value pools where AI will move our economics?
• Do we have a funded path from proof to production for each?
• How are we making sure we are building trusted, responsible and ethical AI?

PwC works with insurers across all four dimensions: equipping boards and executives to make informed AI decisions, translating ambition into quantified value pools, building a repeatable route from proof to production, and embedding governance that enables responsible scale. The priority is not to launch more pilots, but to build the enterprise capability that consistently converts AI investment into P&L outcomes.

With thanks to Scott Hadfield, Matthew Tutty, Matthew Benwell and Henry Briggs, all of whom contributed to this report.