AI for insurance brokers: what UK brokers can automate in 2026

Renewals and quote requests are the best first builds, because most of the work is collecting the same facts, and the broker system already holds the client record.

Renewals

A renewal build starts at a window the brokerage sets, for example 60 days before renewal. It sends the client a summary of what is on cover, asks what has changed, and chases until the answers arrive. It then drafts the renewal invitation for the account handler, who checks the terms and the price before anything goes out.

The price stays outside the AI. Since 1 January 2022, the FCA's pricing rules for home and motor insurance say a renewal price should be no greater than the equivalent new business price (ENBP) for a new customer. The insurer's rating and the broker system produce that figure. An AI that writes renewal emails must copy the price from the system, never calculate or round it.

Quote requests

A quote intake build reads each new enquiry, fills the brokerage's fact-find, asks the client for anything missing, such as turnover or claims history, and opens a draft record in the broker system. The account handler chooses the markets and specifies the client's demands and needs.

Worked example for renewals

These figures are assumptions, so swap in your own.

1. Assume a brokerage renews 6,000 policies a year and staff spend 15 minutes on each one chasing changes and drafting the invitation.
2. That is 6,000 × 15 = 90,000 minutes, or 1,500 hours a year.
3. Assume the AI drafts and chases halve that time, saving 750 hours.
4. At an assumed staff cost of £20 an hour, 750 hours is £15,000 a year.
5. Model usage at an assumed 5,000 input and 1,000 output tokens per renewal on Claude Sonnet 5.5 ($2 and $10 per million tokens) is about $0.02 a renewal, or $120 for 6,000.

AI can compare two policy wordings faster than a person can, but the broker still reads every difference before advising. The FCA's rules on demands and needs (ICOBS 5.2) say that any policy a broker proposes must be consistent with the customer's insurance demands and needs. An AI comparison is an input to that judgement, not the judgement.

A comparison build takes the expiring wording and the proposed one, lines up each section and lists every difference in cover, exclusions, conditions, limits and excesses, with clause numbers in both documents. It marks clauses it could not match. The broker opens each flagged clause in the original PDF before writing the recommendation.

Model usage is small. Two wordings of about 30,000 tokens each (an assumption) cost about $0.12 to read on Claude Sonnet 5.5 at $2 per million input tokens, and a 3,000-token comparison costs $0.03 at $10 per million output tokens. One comparison costs about $0.15. The custom LLM vs RAG guide explains how a tool searches a brokerage's own documents.

The FCA has written no AI-specific rules for brokers, but it holds firms responsible for every outcome an AI tool touches. Four FCA documents set out the position, alongside the pricing rules above.

• AI and the FCA, our approach (first published 8 September 2025, updated 2 October 2026). "We do not plan to introduce extra regulations for AI. Instead, we'll rely on existing frameworks, which mitigate many of the risks associated with AI." It says the Senior Managers and Certification Regime emphasises accountability for senior managers and is relevant to the safe use of AI.
• AI in financial services, good practice and areas for improvement (19 October 2026). "Where firms are deploying AI-supported tools, they remain responsible for consumer outcomes." It warns that "Effective oversight is not simply about reviewing AI outputs" and that "Human review may mask weaknesses in the model". It says "Firms can outsource AI services, but remain responsible for them", and calls for auditable records of AI interactions, decisions and rationale.
• The Mills Review (6 July 2026). It found the existing framework, including the Consumer Duty, "remains sound". It also said "it will not be enough to say that a person remains 'in the loop'". Firms need to be clear what the person is expected to do, what information they receive and when they can intervene. Firms responding to its engagement paper expected "significant effects on intermediary markets, with disruption to brokers."
• The Consumer Duty (PS22/9). Its four outcomes cover products and services, price and value, consumer understanding and consumer support. It applied from 31 July 2023 to products open to sale or renewal, and from 31 July 2024 to closed products.

What this means for a broker's AI build

• Log every draft, the model version, who approved it and what was sent, so a complaint can be investigated.
• Give the approver the source documents next to the AI's draft, so the check is real.
• Keep a person on every recommendation, every statement about cover and every claims decision.
• Test client emails with different customer groups, including customers in vulnerable circumstances.

Data protection

Claims files often hold health and injury details, which UK GDPR treats as special category data. The ICO's "Guidance on AI and data protection" covers data protection impact assessments, lawfulness and transparency, and the ICO says it is under review after the Data (Use and Access) Act. Client data should only go to AI providers on paid tiers whose terms keep it out of model training.

Yarigai is a done-for-you AI agency in London. It builds AI agents, document reading and filing, inbox triage, support agents that answer from a client's own documents and internal tools that replace spreadsheets. For a brokerage, it builds them inside the firm's own email, document store and broker system. The brokerage owns the code, the data and the accounts from day one.

A project starts with a free 30-minute scope call on Google Meet, and the brokerage keeps the notes whether it hires Yarigai or not. A blueprint of about a week, credited to the build, sets out a written plan, a clickable prototype, the monthly running cost, a fixed price and the payback maths. The build takes 2 to 6 weeks in the firm's accounts with a demo of working software every week. A single workflow, such as renewal chasing, can go live in about two weeks.

A person approves anything that sends money, emails a customer or deletes data. AI and hosting bills are paid by the brokerage at cost, and the monthly running cost is on the quote. Yarigai's guarantee is one sentence: "If the build isn't profitable for you, we won't build it."

Yarigai has no published insurance broker case study. Its document and operations work is described on the services pages, and the software it built and runs is listed on the about page.