Most UK insurers now run AI, but execution gap widens

TL;DR:

  • More than half of UK insurers (55%) have embedded AI in core business functions, a sharp shift from the proof-of-concept caution of 2024.
  • A widening gap between ambition and operational scale is now the sector’s pressing challenge, with governance, data quality and talent the main constraints.
  • The findings mirror a broader UK pattern: AI adoption is accelerating faster than the capacity to run it reliably at scale.

UK insurance has crossed a threshold. New research from Earnix finds 55% of UK insurers now use AI in some core business functions — claims, underwriting, pricing and customer decisions — a decisive move beyond the isolated pilots that defined the market two years ago. The harder question, the report argues, is no longer whether to adopt AI but whether firms can make it work consistently in production.

Ambition outpacing execution

The numbers point to momentum and strain in equal measure. Some 98% of UK insurers either use or plan to use generative AI to process unstructured data, well ahead of the global average. Yet 30% admit they significantly lag customer expectations on personalisation, only 28% of leaders strongly agree their governance cadence is adequate, and around 80% remain worried about the quality of data feeding their models. Talent compounds the problem: many carriers run older core systems and lack the AI engineering depth to move pilots into live use.

The execution gap is not unique to insurance. It rhymes with evidence that half of enterprise AI projects stall before scale, and with regulators’ warnings to banks over frontier AI and cyber resilience. Parliamentary scrutiny is rising too: the Treasury Committee has found more than 75% of UK financial services firms now use AI, and warned that a wait-and-see approach to risk could expose the system to serious harm.

Looking forward

The firms pulling ahead, Earnix suggests, treat AI as a governance challenge rather than a technology project. For UK insurers, that reframing is the practical lesson — scaling depends less on buying more models than on the data, oversight and skills to deploy them safely where decisions are actually made.

Share this article