Brokers, AI, and the Next Decade: Guardrails, Governance, and the Human Advantage

I spent the last year watching brokers try AI tools. Some worked. Most didn’t. The ones that stuck had one thing in common — they didn’t replace the broker’s judgement, they evidenced it. I think the next decade in insurance advice won’t be defined by which firms adopt AI fastest, but by which ones build the guardrails to use it responsibly. Here’s why.

Early on, I watched a commercial broker trial a general-purpose AI assistant for policy comparison. It could summarise a wording in seconds. Impressive demo. But when a client asked why their business interruption limit was lower than last year, the AI had no answer — it had read the words but not the structure. The broker spent an hour reconstructing the comparison manually anyway. The tool that replaced it six months later was different: it read the clause structure, mapped endorsements against the base wording, and showed the broker exactly where the delta was — with the clause reference. The broker still made the call. But now they could show their working.

This is the position I’ve come to hold strongly: in regulated advice, the guardrails around an AI matter more than the AI itself. A model that reads a policy and produces a plausible summary is useful. A model that reads a policy, produces a summary, and points to the exact clause behind every statement — with a human reviewing before it reaches the client — is defensible. The difference isn’t speed. It’s accountability. When a regulator asks how you formed your advice, ‘the AI said so’ is not an answer. ‘The system read the wording, surfaced this clause, and I reviewed it before advising the client’ is.

I think we’ll see three phases. First, the current one: brokers experimenting with AI tools, most of which are general-purpose and poorly fitted to insurance. Second, a consolidation around tools that understand policy structure — not just words, but wordings, endorsements, and how clauses interact. Third, and this is the important one, regulation will catch up. When it does, the firms that already have traceability, audit trails, and human-in-the-loop workflows won’t need to retrofit them. They’ll just point to what they’ve been doing all along.

The brokers who win the next decade won’t be the ones who adopted AI first. They’ll be the ones who adopted it with the governance to prove it was used well. That’s the human advantage — not competing with the machine, but being the person accountable for what it produces.

What responsible AI in regulated advice looks like in practice — and why traceability is the guardrail that matters.

James Farrell

Blogs by James Farrell