Insurance AI Deployments Jump 87%
AI deployments in the insurance sector have surged 87% year-over-year, driven by generative AI and agentic systems. Despite the boom, a recent podcast revealed that 90% of AI pilots stall in 'pilot purgatory' due to workflow integration issues and tech debt, not technology failures.
The surge in AI adoption is most prominent among Property and Casualty (P&C) insurers, who accounted for 50% of all deployments in the last quarter of 2025. The overall AI in Insurance market is projected to skyrocket from approximately $15 billion in 2025 to over $246 billion by 2035. This growth is fueled by tangible results, with about 40% of insurers now reporting measurable business benefits from their AI initiatives. Insurers are moving beyond basic automation to full process ownership with agentic AI, which represented 21% of new deployments in late 2025. These systems excel in complex areas like claims, where they handle multi-step workflows autonomously. As an example, Allianz's "Project Nemo" utilizes a central AI agent to coordinate seven specialist agents, cutting the processing time for certain claims by 80%. The impact on underwriting is similarly transformative, as AI can automate up to 70% of underwriting tasks, according to McKinsey. Generative AI specifically helps underwriters by extracting key data from submission documents and augmenting risk analysis, freeing them from manual work that consumes an estimated 40% of their time. Insurers using AI are seeing an 18.6% reduction in claims processing times and getting new products to market 15.4% faster. The "pilot purgatory" issue is deeply rooted in the industry's estimated $200 billion of technical and process debt. Many insurers operate on legacy systems from the 1980s or 90s, with up to 70% of IT budgets spent just on maintenance. This foundational weakness is a key reason why only 7% of insurance companies have successfully scaled AI systems across their organizations, despite high adoption rates. Beyond technology, stalled pilots often result from a failure to treat AI as a core business capability rather than an isolated experiment. Successful implementation requires a shift away from focusing on model accuracy in a sandbox to building the MLOps infrastructure needed for the real world. This includes establishing clear, measurable business outcomes and ensuring alignment between technical and business teams from the start.
