A Practical Guide to AI adoption for insurance brokers

What AI adoption actually means in a broking firm

Whether a firm "uses AI" depends on four separate things: usage, who has access to a tool, the easiest thing to measure and the least useful; delegation, what work has actually moved from a person to a system; trust, built by a system proving its outputs are traceable back to source, not by a vendor's marketing; and responsibility, which never moves.

The FCA relies on existing frameworks, Consumer Duty, SM&CR and governance and controls, rather than AI-specific regulation, and using AI never transfers regulatory accountability to the tool or its provider. A firm's real stage of adoption is the answer to all four at once.

The five stages of AI adoption for insurance brokers

Stage 0 — Gated. Little or no approved AI use, usually over data or compliance concerns, or because nobody owns the decision. This often just pushes use underground, with staff experimenting on personal accounts instead of on anything the firm has actually agreed to.

Stage 1 — Assisted. One broker, one tool and one task at a time: summarising a wording, drafting an email or comparing a schedule against a proposal, with every line checked manually. This is likely where much of current adoption sits. The risk is mistaking speed for reliability under time pressure.

Stage 2 — Parallel. Several bounded checks run at once across a book: comparing insurer proposals, flagging wording differences, spotting endorsement changes. The bottleneck shifts from doing the work to prioritising what gets reviewed first.

Stage 3 — Supervised autonomy. Systems monitor documents and processes continuously. A wording change is compared automatically against the previous version, a discrepancy is sent directly to the account handler, and exceptions are routed to a person without waiting for someone to initiate the check. This only works with a genuine audit trail and clear escalation rules. Without them, a firm cannot account for what its own system did.

Stage 4 — AI-native. Processes are designed around what people and systems each do best, not AI bolted onto old workflows. The risk is accountability quietly drifting as automation grows, unless it's actively maintained.

The five stages at a glance

How to start safely

Begin with a usage policy: named ownership, a short list of approved tools, and explicit rules on what client data can go into them. Every personal-data processing operation needs a defined purpose and lawful basis, established before processing begins, with additional requirements where special category data is involved. Outputs need to be traceable back to their source document and logged: what was checked, when, and what a human did about it. Whatever the stage, final advice, material coverage recommendations and client negotiations stay human-owned.

A good first use case is frequent, easy to verify against a source document and low-consequence enough that an error will be caught before it reaches a client. Wording comparisons and schedule checks often fit those criteria; open-ended client conversations do not. Before choosing a vendor, ask whether client data is retained or used to train the model, whether access can be restricted by role, whether outputs link back to source clause by clause, and whether there's a complete, exportable audit log.

Days 1–30: Usage policy, a named owner, one or two bounded use cases.

Days 31–60: A controlled pilot, tracked against manual verification time.

Days 61–90: An honest review, and a documented decision to scale, adjust or stop.

That documented decision is worth as much as the pilot itself if the same question comes up again in a year.

The goal is not maximum automation

Reaching Stage 4 isn't the objective for every firm. A small brokerage can get real value sitting at Assisted or Parallel, depending on the work and how much supervision capacity it actually has. The honest way to measure progress isn't logins or prompt counts, but how much manual work the same result would have taken without the tool, and whether the firm can still explain how it got there months later.