Briefing: How UKGI should respond as AI becomes more popular with consumers
Consumer sentiment is shifting. Research indicates the share of customers comfortable with AI deciding policy prices without human intervention rose from 31% in 2024 to 37% in 2025. Preference for phone support remains strong, but AI is earning trust where it proves speed, clarity, and outcomes.
What consumers are saying
In 2023, 56% of surveyed UK insurance customers preferred to speak on the phone about their insurance. That hasn't vanished. But the 2025 data matters: more UK customers now accept AI assisting human call handlers (39% vs 30% in 2024) and more are open to AI-only pricing decisions.
Signals from the market
Bridgetech's chief executive, Matthew Porter, reports strong growth in AI chatbot uptake, noting "communities that really want to talk on the phone and there are communities that really don't." His firm has rolled out an animated claims journey and an AI chatbot to meet both needs. The direction is clear: serve channel preference and reduce friction.
This sits under the FCA's Consumer Duty, with an explicit emphasis on fair outcomes and customers in vulnerable circumstances. Firms must offer real choice, clear explanations, and easy escalation paths. See the FCA overview of Consumer Duty for grounding on expectations and outcomes: FCA Consumer Duty.
Where adoption lags
There's a delivery gap. According to DCL's September 2025 Broker Barometer, 68% of brokers feel providers are not fully using technology to streamline claims, which drags on communication. Industry discussion also points to insurers deploying AI more than brokers, with one estimate citing 3.5x more insurer AI use cases than broker implementations.
Many brokers report small-scale usage (documents, marketing) rather than end-to-end processes. Meanwhile, confidence in AI's potential for customer benefit is high, suggesting the bottleneck is execution, not belief.
The playbook: a practical human + AI model
- Route by preference: offer phone, chat, and self-serve. Let customers choose. Bake in a clear "talk to a person" option at every step.
- Claims triage first: use AI for FNOL capture, status updates, and smart routing. Hand off to humans for complexity, vulnerability, or disputes. Keep full audit trails.
- Pricing with guardrails: treat AI pricing as decision support under human accountability. Implement pre-deployment testing, bias checks, challenger models, and periodic reviews.
- Vulnerability by design: detect signals (language, behavior, declared needs). Slow down, simplify, and switch channels when needed. Provide escalation to trained specialists.
- Broker enablement: offer APIs, enriched data, and AI co-pilots for quote prep, coverage comparisons, and correspondence. Share insights, not just PDFs.
- Front-line AI assist: give call handlers real-time guidance, knowledge retrieval, and summarisation to cut AHT and errors-while keeping the human in control.
- Data governance: align with UK GDPR and Consumer Duty. Run DPIAs, control data lineage, log explanations, and define model ownership and versioning.
- Security and privacy: redact PII in training data, enforce least privilege, encrypt in transit/at rest, and prepare an incident playbook.
- Measurement discipline: track FCR, NPS/CES, complaint rates, claim cycle time, leakage, and escalation rates. Publish service standards and hit them.
- Upskill the team: train underwriting, claims, and broker teams to use AI tools effectively and safely. A practical option by job role: AI courses by job.
12-month roadmap
- 0-90 days - Launch a claims-status chatbot with seamless human handoff. Deploy AI summaries and knowledge retrieval to assist call handlers. Start bias and performance baselines for any AI in pricing or fraud.
- 3-6 months - Automate document intake (OCR), fraud alerts, and claims triage rules. Ship a broker co-pilot for quote prep and renewal packs. Expand vulnerability prompts and scripts.
- 6-12 months - Pilot AI-influenced pricing under strict governance. Add subrogation and recovery suggestions. Integrate proactive outreach (delays, missing docs) via preferred channels.
Metrics that matter
- Customer: FCR, NPS/CES, complaint rate, average response time, abandonment rate, escalation-to-human rate.
- Claims: cycle time, touch count, settlement accuracy, leakage, indemnity spend variance, fraud hit rate.
- Risk and compliance: model fairness tests, explainability coverage, DPIA completion, audit findings, call QA scores.
- Commercial: quote-to-bind, renewal retention, cost-to-serve, and ROI by use case.
Risk, compliance, and trust
Disclose how AI is used, why decisions were made, and how to request human review. Provide channel choice, accessible formats, and extra support for vulnerable customers. Log every decision path for audit readiness under Consumer Duty.
Run DPIAs for material models and monitor drift, bias, and outcomes continuously-not annually. Practical guidance: ICO: Data Protection Impact Assessments.
Bottom line
Customers still value the phone, yet comfort with AI is rising fast. The winning model is hybrid: AI for speed and consistency, humans for context and care. Build for choice, govern tightly, measure relentlessly, and move now.
