Background—Industry view
The state of the industry—A fragmented landscape of maturity
Until the industry aligns its architectural readiness with its strategic aspirations, AI will remain a promise partially fulfilled. The call to action is not just to innovate—but to prepare, to govern and to lead with purpose. Autonomous AI agent control Most organizations aren’t ready yet - even the most advanced players, those who have invested heavily in modernization, established robust governance frameworks, embedded ethical safeguards and deployed human-in-the-loop models, are not ready to let go of control. Because the truth is, no organization today is ready to hand over complete operational control to an autonomous AI agent. AI may assist. It may recommend. But autonomous action, without any human checkpoint, remains a leap too far. Having said that, if true intelligent autonomy is our goal, we can’t simply preserve human-in-the- loop approach indefinitely—we must ensure that safeguards remain. In our own experience, this pattern holds true. AI is widely used in claims fraud detection to identify anomalies and surface suspicious patterns. However, they do not make the final determination. Flagged cases are still referred to experienced claims professionals who understand the nuances of policy language, client history and regulatory context. The same is true in underwriting. AI can help surface risk indicators and suggest pricing based on historical loss data or exposure trends. But ultimately, it is the licensed underwriter, often working closely with the broker, who weighs those insights against market conditions, client relationships and the judgment developed over years of experience in the field. The technology supports the decision but does not replace the human element.
The insurance sector, long defined by fiduciary duty and regulatory conservatism, now faces a crossroads. A growing body of research reveals a stark truth: While a handful of global insurers are exploring AI as a transformative tool, most companies remain early in their journey. According to BCG, only 4% of companies have matured to the point of running AI as a full- fledged value engine, while 22% are scaling AI to generate impact. The remaining 74% are still in the early phases—either doing very little with AI or limited to proof-of-concept efforts. Meanwhile, research from Forrester indicates that fewer than 5% of insurers will realize significant, direct AI- driven revenue gains—highlighting widespread experimentation overshadowed by legacy systems and fragmented governance models. Our own journey reflects this broader reality. Legacy platforms, built for a different era, have inhibited data fluidity, delayed decision-making and limited our ability to generate insights at the speed today’s market demands. Before activating meaningful AI capabilities, we had to modernize the enterprise data landscape—integrating disparate systems into a unified, trusted and accessible foundation. This isn’t merely a technical upgrade; it is a strategic imperative. The lesson is clear: Ambition alone does not unlock AI’s potential. Enterprises must first re-architect their core—data, infrastructure and governance—before AI can scale with impact. Even among insurers who have streamlined legacy systems and begun experimenting with foundational models, few have embedded the transparency, auditability and governance required to inspire confidence in regulated environments. Perhaps because, in a domain where risk is existential, the margin for error is virtually zero.
3 | Reinforcement before autonomy: Engineering trustworthy autonomy in insurance AI
Powered by FlippingBook