Carriers stuck in 'pilot purgatory' as AI fails to graduate – Sedgwick
Most insurance carriers have poured money into artificial intelligence tools for property claims, but barely any have figured out how to make them work at scale, a new report from claims management firm Sedgwick has found.
The gap is stark. While as many as 82% of carriers now use some form of AI in their operations, just 7% have reached what Sedgwick calls scalable success – the point where AI is embedded across the full claims lifecycle, not just running in a handful of pilot programs.
That disconnect has left the industry in what observers have described as "pilot purgatory," where carriers experiment with AI but fail to graduate to enterprise-wide deployment.
The report found that nearly two-thirds of carriers acknowledge a gap between their AI ambitions and actual capabilities, and 90% said AI needs to be coordinated across business processes to deliver meaningful returns.
The picture is more encouraging in pockets where AI has gained a foothold. Sedgwick pointed to intake automation cutting average property claims processing times from 10 days to 36 hours, and AI-driven photo analysis boosting claim handling efficiency by up to 54%.
On low-severity claims, some carriers have reported 80% faster processing and 50% productivity gains in documentation. Industry experts believe that as much as 85% of straightforward claims could eventually be processed end-to-end with minimal human involvement.
A previous Bain & Company study painted a similar picture, finding that 78% of P&C insurers had adopted generative AI in some form but only 4% had managed to scale it across their organizations.
The biggest drag on progress remains legacy technology. Research from Equisoft has shown that many claims systems still run on outdated languages such as COBOL and were never designed for cloud-native applications or real-time data exchange.
Read more: Why most agents are still stuck in AI pilot mode
Databricks has noted that legacy data warehouses built around batch processing cannot support the streaming data AI needs for tasks like fraud detection.
The cost of maintaining these aging systems is punishing. NCube has estimated that legacy upkeep can swallow as much as 80% of a carrier's IT budget, starving innovation of funding.
McKinsey has previously noted that many of these platforms are poorly documented, making modernization even harder – though it found one major global insurer used generative AI to improve code modernization efficiency by more than 50%.
Sedgwick stressed that AI should sharpen human expertise in property claims rather than replace it, especially in complex or emotionally charged cases. Three-quarters of claims professionals surveyed said AI still requires human oversight, and Sedgwick found that human-in-the-loop models quadruple trust in AI-generated outputs.
Regulators are also paying closer attention. Colorado's Artificial Intelligence Act has raised the bar for transparency and human oversight in AI-driven claims decisions, a trend likely to spread to other jurisdictions.
Looking ahead, the report projected that routine claims tasks will become increasingly automated through 2026, with more carriers expected to turn to agentic AI to drag their legacy systems into the modern era.
