(Image credit: IIR/Gemini.)
The insurance industry has spent decades modernizing from the outside in. Sleek front ends and digital channels have kept pace with customer expectations, while core policy, claims and underwriting systems have lagged behind. Agentic AI is now exposing the limits of that approach.
Deploying AI agents into operating models designed for a pre-AI era risks compounding complexity rather than reducing it—or breaching governance requirements. While 75 percent of boards have approved major AI investments, 78 percent lack strong confidence that their organizations could pass an independent AI-governance audit within 90 days. The challenge is to move beyond pilots and embed AI agents safely in business-critical processes where they can deliver meaningful value.
Why pilots fail to scale
When pilots fail to scale, the instinct is often to blame the AI. In practice, three organizational factors frequently stand in the way.
First, insurance processes—even within digitally native InsurTechs—were built around human execution. Agentic AI depends on workflows that coordinate human and machine activity, yet few insurance processes were designed with that collaboration in mind.
Second, compliance often enters the process late, typically at the proof-of-concept stage, when it should be shaping design decisions from the outset.
Finally, without enterprise-wide support, pilots remain isolated initiatives serving a single team or function, with no credible route into the wider business.
Insurance also brings sector-specific challenges. Claims and underwriting workflows carry decades of institutional knowledge, while expectations for model-risk management continue to rise. Regulatory scrutiny increasingly extends beyond explainability to the full model-development lifecycle, including design, backtesting, regression testing, governance reviews and drift assessment. Insurers are consequently reassessing risk and control frameworks that predate AI.
These pressures point to a broader problem. Agents introduced into fragmented environments cannot coordinate effectively across workflows, increasing complexity, cost and compliance risk. Insurers therefore need mechanisms that embed coordination, control and auditability into processes rather than adding them only after problems emerge.
Three routes through the transformation dilemma
Insurers cannot rebuild their operations overnight without putting balance sheets, customers and business continuity at risk. Most will instead pursue some combination of three approaches.
The first is to connect existing platforms and applications, extracting greater value from previous investments. The second is selective replacement. Large-scale rip-and-replace programs remain expensive and risky, but replacement may be justified in particular functions or new product lines where legacy constraints have become unsustainable.
A third option is an ecosystem model in which insurers combine their brand, distribution and underwriting capabilities with specialized technology or service providers.
Each approach requires insurers to coordinate systems, agents and human decisions while maintaining consistent controls and auditability. That coordination can help existing platforms deliver greater value, support controlled migration from legacy infrastructure and make external operating models viable in areas such as customer onboarding.
Redesign before deployment
Coordination alone is not enough. Many insurance processes in use today were designed before AI agents became a practical consideration, making process redesign a prerequisite for deployment. Yet insurers often implement agents before determining how end-to-end processes should operate, where bottlenecks exist and where human judgment remains essential.
Governance must be embedded from the outset. Process logic determines when an agent is engaged and when human review is required, while decision logic defines what can be automatically accepted, referred or escalated. Building those decisions into the design stage helps ensure that agents operate within enforceable guardrails rather than the gray areas that cause many pilots to stall.
One example of a process designed around specialized agents and retained human authority is Allianz’s Project Nemo. Launched in Australia in 2025, the system handles low-value food-spoilage claims arising from weather events. Seven specialized agents address planning, coverage verification, weather validation, fraud screening, payout calculation and auditing, while a human retains the final payout decision. Allianz has reported an 80 percent reduction in processing and settlement times for eligible claims below $500 and plans to extend the system gradually into other low-complexity, high-frequency use cases.
The decade ahead
Over the next decade, some major insurers may concentrate more heavily on brand, customer relationships and risk-bearing while relying on specialized providers for selected claims-management and servicing functions.
That model will require flexible, composable architecture. As models, vendors and regulations evolve, insurers will need to adapt AI components, update governance frameworks and onboard providers without losing operational visibility or control. Consistent coordination and oversight across participating systems and organizations will be necessary for such models to operate safely at scale.
The insurers that make the greatest progress will not necessarily be those running the most agents. They will be those that redesign their operating models around effective collaboration among people, agents and systems—and govern how that collaboration works.