Part of our guide: AI in financial services
Insurance is one of the most regulated, process-heavy sectors in the UK, and that also makes it one of the best positioned to benefit from AI, provided you target the opportunities correctly.
Most of the noise around insurance and AI focuses on large-scale automation, and the vendor pitch usually arrives in a single line: "Cut claims processing costs by 60%." That pitch is not wrong, and it is not the strategic picture, because real insurance AI is more nuanced: reducing friction, improving underwriting quality and staying ahead of regulatory expectations.
Where AI actually works in insurance:
Claims processing. This is where you will hear most vendor pitches, and they have a point, since claims triage and assessment are genuinely repetitive. AI can categorise claims, flag high-risk or complex ones for human handlers and provide preliminary assessments, and claims handlers are expensive for a reason that matters here. They make judgment calls, spot fraud and know when something does not add up, so you are not replacing them and are buying back their time for the decisions that actually need their expertise.
Underwriting assessment. This is where AI gets more interesting, and historical underwriting data is rich, telling you what happened to policies that met certain criteria, what claims came through and what the actual loss ratios were. You can train models on that data to assess new applications, and the regulatory question is whether your model is explainable. The test is whether you can tell a regulator why you declined an application or charged a premium. The answer is usually yes where you choose the right approach, and you have to think about explainability from day one.
Customer onboarding. Insurance onboarding is document-heavy and repetitive, and the work runs to KYC checks, ID verification, proof of address and beneficial ownership. These tasks are ideal for AI-assisted processing, where the aim is not full automation: flagging exceptions and making the compliant cases invisible. That speeds up good customers and keeps your compliance team focused on the hard cases.
Regulatory reporting. This work places a heavy load on insurance operations teams, covering monthly, quarterly and annual returns to the FCA or PRA, regulatory capital calculations and stress testing inputs. All of it requires data from multiple systems, validation, reconciliation and commentary, and AI can automate much of the data plumbing and flag inconsistencies before they become compliance problems.
Where AI doesn't work in insurance:
AI does not work for anything core to claims decision-making where a significant pay-out is involved, and your insurer's reputation is built on fair claims handling. An AI system that declines a legitimate claim because it did not understand the context is not efficiency, and it becomes a brand disaster and a regulatory problem.
The regulatory backdrop matters enormously
Insurance in the UK is dual-regulated: the FCA covers conduct, distribution and governance, and the PRA (part of the Bank of England) covers prudential risk and capital. Both regulators hold strong opinions on operational resilience and AI, and the PRA published expectations on AI governance in November 2024.
Those expectations are worth reading, and they are not hostile to AI, though they expect you to understand model risk and have clear governance. They expect you to validate your models and understand what happens when they fail, and that is not bureaucratic friction, since it is exactly what responsible AI looks like.
The FCA cares about consumer outcomes, and an AI system that systematically treats some customer groups worse than others is a problem regardless of whether you intended it. Where you cannot explain a decision, that is a problem as well, and senior managers are accountable for it without qualification.
This is not a technology problem
The biggest insurance firms that are moving AI forward are not the ones with the cleverest models, and they are the ones with governance. They have a Chief Risk Officer who understands machine learning, clear owners for model validation and documented decisions, retaining a clear route for challenging a model that is not working.
Where your board treats AI as a technology question, you will end up building systems that regulators question and customers distrust.
How to sequence insurance AI
Start with the friction points where AI is obvious and low-risk: onboarding, regulatory reporting and triage, getting the governance framework right, since you will need it anyway. Build internal confidence that your models work, and then move into the more complex territory of underwriting, claims assessment and pricing. By then you will have people who understand the technology, a governance process that actually works and a board that trusts your approach.
For insurance businesses with 100-2,000 people, this is a 2-3 year programme and not a six-month vendor project. You need strategic clarity on the capabilities that matter to your business model, senior leadership alignment on governance and the discipline to walk away from shiny solutions that do not fit.
Get in touch to talk through where AI creates real value in your insurance business.