AI for insurance brokers: automate transactions, not relationships

Insurance brokers ask us one question more than any other: where does agentic AI actually help, and where does it get in the way? We spent days on that question at the American Agents Alliance Expo in Palm Desert. This post sets out our answer.

Where agentic AI fits: quoting to servicing

Insurance work is a chain of transactions. A risk is described, documents are collected, a quote is requested from several carriers, the quotes are compared, the client signs, and then the policy is serviced for years. Most of that chain is repetitive: the same fields typed into a different portal, the same follow-up e-mail sent for a missing registration document.

Agentic AI is well suited to this kind of work because the steps are known and the result can be checked. It can:

• collect the documents and chase the ones still missing, as our intake agent does;
• fill a carrier portal from a single risk file, screen by screen, as our quoting agent does for extranets;
• compare quotes against a fixed grid, without inventing coverage that a quote does not state (see how to compare insurance quotes with AI);
• prepare the servicing requests that follow a bound policy.

Automate transactions, not relationships

One keynote point stayed with us: AI should automate transactions, not relationships. A client who calls about a claim, or asks whether a coverage still fits after a move, is not waiting for a faster form. They are waiting for an agent who knows their file and picks up the phone.

So we draw the line in the same place. The AI handles the work between the client and the carrier. The agent keeps the conversation with the client, the choice of coverage, and the final review. Nothing is submitted without a person checking it.

In practice, the line looks like this:

• AI does: read the documents, build the risk profile, fill each carrier portal, line up the quotes side by side, draft the follow-up e-mail.
• The agent does: talk to the client, recommend a coverage, explain the exclusions, decide what to send, and submit.

Every value the AI writes points back to the document it came from. When a field has no source, it stays empty, and the agent sees what is missing before anything leaves the agency.

What we take from the Expo

• Start with the quoting workflow. It has the most repetition and the clearest checks.
• Measure time saved per file, not the number of tools deployed.
• Keep a human at each step where the client is affected.
• Make every value traceable to the document it came from, so a reviewer can check it in seconds.

Agencies keep their clients. The software should give them back the hours they now spend on re-typing.