The 2026 Insurity study has delivered the headline the insurance industry wanted to hear. Nearly 40% of consumers now support AI use in claims, underwriting and service delivery — a jump of 95% year-on-year. For mid-market insurers operating under FCA Consumer Duty PS22/9, this data is being read as permission to accelerate. But it is being misread. Consumer comfort with AI existing does not mean consumer trust in how that AI is being used. Familiarity — 84% of consumers use AI daily — has bred acceptance, not understanding. Insurers are confusing these two things, and that confusion is a regulatory and reputational liability.

This pattern repeats across financial services. Organisations see rising consumer tolerance for AI and interpret it as cover for implementation at pace. They deploy large language models like GPT-4 or proprietary systems without the documentation, testing and explainability controls that regulators — the FCA, PRA in SS1/23, and increasingly the ICO on algorithmic bias — now expect. The result is a widening gap between what boards think they can do and what they can actually defend. In law and accountancy, we see the same cycle: Harvey and Legora brought generative AI to legal research fast, but they succeeded because they stayed narrowly focused on genuinely safe use cases and published their limitations. Most mid-market firms adopting AI are doing neither.

Trovix's view is this: consumer permission is not compliance. Deploying AI in insurance — whether for claims triage, pricing, or fraud detection — requires three things most insurers are skipping. First, documented risk mapping that shows what can go wrong and how you tested for it. Second, explainability that actually works: not tooltips explaining the output, but evidence you can show the FCA that the model is not discriminating against protected characteristics in breach of Equality Act 2010. Third, human oversight that is real, not ceremonial. Tools like Trovix Sift exist because they solve the first two: they extract and classify data with auditability baked in, not bolted on. When your underwriting AI makes a decision, you need to show why. Not assume the consumer trusts you.

If you run a mid-market insurance firm, do this now. Map which AI systems are touching customer decisions — claims acceptance, premium setting, fraud flagging — and which are touching only internal efficiency. The first category needs full governance; the second can move faster. Second, audit your current AI vendors. If they cannot show you their testing methodology, their fairness metrics, or their error rates by demographic cohort, they are not compliant with FCA expectations, and neither are you by proxy. Third, document your human override process in writing. Consumer trust rises when people know a human is checking the machine, not when they are told it.

Source: Repairer Driven News