AIG's August warning about AI data center saturation lands hard because it is honest. AI infrastructure demands comprehensive, interlocking coverage: construction phase property and cyber, operational liability, supply chain disruption, systems failure, third-party breach. A single facility can require dozen separate policies. UK P&C underwriters, already stretched by regulatory scrutiny (PRA SS1/23 on operational resilience, FCA Consumer Duty PS22/9 on fair value assessment), are now facing a queue of deals that demand specialist expertise they do not have enough of. This is not a story about AI generating opportunity. It is a story about the insurance industry revealing it cannot scale to meet real-world infrastructure demand.

The pattern is stark. Insurance technology has spent five years optimizing the wrong thing. We have invested in faster document review (Luminance, Everlaw) and smarter claims triage (various RPA vendors), all useful, but none of these tools address the core problem: underwriters lack the time and cognitive bandwidth to run through complex exposure scenarios, regulatory fit assessment, and concentration risk modelling on deals that arrive faster than they can be underwritten. The market has filled this gap with generic large language models wrapped in insurance UX (Microsoft Copilot for Enterprise, Harvey's industry modules) that promise to 'augment' underwriting teams. They do not. They automate the easy parts and hand back messy ambiguity to humans who are already overloaded.

Trovix's view is different. This crisis does not require better AI. It requires better information architecture. Underwriters drown because they are reassembling the same regulatory, technical and commercial intelligence on every deal. The solution is to build a retrieval-augmented knowledge layer (Trovix Aria) that sits between the underwriter and their institutional expertise—ISA UK 315 audit standards, FRC audit guidance, PRA expectations on cyber resilience, Lloyd's Blueprint Two requirements, sector-specific exposure precedent. Then automate what can be automated without false confidence: structured data extraction (Trovix Sift) from technical drawings, financial statements, and D&O disclosures. The rest remains human judgment. This is not about replacing underwriters. It is about giving them back the time to think.

For a mid-market P&C insurer operating under FCA and PRA oversight, the immediate action is not to license another AI platform. It is to audit your underwriting workflow as it actually exists, not as your tech vendor describes it. Where do underwriters lose time to context-switching? Where do they repeat the same regulatory interpretation across deals? Where does information live in siloed systems? Then systematically layer in AI where it genuinely compresses work (intake, data standardization, regulatory monitoring via Trovix Watch) rather than where vendors promise it will. The capacity crisis is real. But the answer is not fairy dust. It is disciplined process redesign with AI as a tool, not a solution.

Source: Bloomberg