Global commercial insurance rates fell 4% in Q2 2025, marking the fourth consecutive quarterly decline after seven years of increases. Property rates are down 7% globally, cyber down 7%, and reinsurance prices softened by 2.5% at July renewals. As top-line growth stalls, the industry inevitably turns to operational efficiency. We’ve seen this movie before: increasing rate masks core profitability challenges for years. When the smoke suddenly fades, a reckoning is forced and the C-Suite looks around the room for expense levers and increased competitive advantage. But here’s what’s different this time, AI offers the most significant opportunity to reduce expenses I’ve seen in three decades in this industry.
According to Accenture research, (re)insurance will be the second most impacted industry in the AI era, with 62% of all working hours potentially transformed. Companies moving decisively are already seeing expense ratio reductions up to 65% and underwriting capacity increases of 40%. The early movers are pulling ahead fast.
Meanwhile, everyone else remains stuck in familiar traps: endless internal builds, builds with vendors that never quite seem to work, siloed systems, and that maddening paradox of drowning in data while unable to use any of it effectively.
The Billion-Dollar Problem Hidden in Plain Sight
Walk through any underwriting floor or broker office and you’ll see the same scene: talented professionals drowning in PDFs, manually entering data into multiple systems, waiting for responses from other departments. Yes, data extraction from documents is the critical foundation as very little moves forward without accurate, structured data. But stopping there misses the real opportunity. Think about a typical commercial underwriting process: submission triage, risk assessment, pricing, quote generation, negotiation, binding, policy issuance. Each step involves multiple systems, manual handoffs, and unnecessary delays.
Here’s what modern AI can actually do: Start with rock-solid document processing, then build to complete process automation. I’m talking about AI that autonomously handles broker queries at 2 AM, manages underwriting workbenches, conducts risk assessments, generates quotes, and processes routine claims without human intervention. Yet most (re)insurers treat AI like a fancy OCR tool. This limited thinking leaves massive efficiency gains on the table precisely when the industry needs them most.
Why Traditional Technology Approaches Have Failed
Let me share a number that should terrify every insurance CEO: One former CIO said his company maintained 2,500 different systems. Twenty-five hundred. Each requiring its own team, its own budget, its own integration nightmares. This is the “system spaghetti” that’s strangling the industry as technology that was supposed to streamline operations has become a operational burden.
The problem isn’t just the number of systems; it’s the thinking behind them. Companies buy point solutions like they’re collecting trading cards. Essential document classification here, critical data extraction there. But, rarely are these systemsconnected into something transformative. Each addresses real needs, sure, but in complete isolation. The real opportunity? Deploy AI agents that handle entire workflows autonomously, from first customer contact through claims payment. Even better AI agents that are pre-built for this task.
Lessons from Early Adopters
After watching dozens of AI implementations, the pattern is clear. The proven solution path takes a fundamentally different approach: use AI built specifically for insurance from day one. These aren’t generic models with insurance training wheels they’re domain-specific language models (dsLMs) that already understand our industry’s complexity.
The Hartford just proved this works, processing 100,000 submissions with immediate accuracy. Why? Because the AI already knew what a dec page was, understood coverage types, and could interpret endorsements without months of training. Twenty-minute time savings per submission, right out of the gate. And that’s just the beginning as this foundation enables true workflow automation that transforms entire departments.
Critical Success Factors for AI Implementation
First, prioritize domain-specific models over generic AI or in collaboration with mass market LLM vendors. Insurance-specific language models trained on millions of policies, claims, and submissions understand the industry’s nuances without customization. When combined with deep insurance knowledge graphs that map relationships between coverages, perils, and exclusions, these specialized models can accurately process any insurance document from day one. This accurate data extraction forms the essential foundation and without it, downstream automation cannot function. But with it, the possibilities expand exponentially.
Second, move from document processing to complete workflow automation. Once AI can reliably extract and structure data, it can then power autonomous agents designed for insurance workflows. The most advanced platforms deploy multiple AI agents working in concert—each understanding specific insurance processes like underwriting guidelines, claims adjudication rules, or regulatory requirements. This layered approach, starting with rock-solid data extraction and building to orchestrated automation, delivers compounding efficiency gains.
Third, measure success progressively: first by data extraction accuracy and speed, then by workflow completion rates. Solutions built on insurance-specific language models should demonstrate immediate accuracy without training for document processing. From there, they should expand to handle complete workflows—any class of business, any document type, any process variation. If a vendor needs weeks to achieve basic extraction accuracy, they’ll never reach the level of process automation the industry needs when 100s of thousands of these documents need processing everyday.
The Window Is Open … for Now
Here’s the brutal truth: The companies deploying comprehensive AI automation today will own tomorrow’s market. They’ll operate at expense ratios their competitors can’t match, respond instantly while others take days, and handle double the volume without adding headcount driving their competitive edge. Their underwriters will actually underwrite instead of doing data entry. Their claims teams will focus on complex cases while AI handles the routine ones.
This isn’t theoretical. Leading carriers are already deploying AI agents that autonomously manage entire processes. Not pilot projects. Not proof of concepts. Real, production systems transforming how insurance works—from submission to binding, FNOL to payment, customer inquiry to resolution.
In a softening market, 5% efficiency improvements are worthless. You need 50% improvements. The winners will be those who stop tinkering with point solutions and embrace comprehensive transformation. The window to gain competitive advantage through AI is measured in months, not years. Companies that move decisively now will define the industry’s future. Everyone else will spend the next decade trying to catch up.
