Insurance people love to talk about underwriting. That makes sense. Underwriting is where capital meets risk, and carriers have been modeling risk for a long time.
But when I look at where AI can help fastest, I keep coming back to the brokerage layer.
Not because brokers are more important than carriers. Because the work around the broker is full of manual judgment support: reading documents, finding missing information, preparing renewals, comparing policy language, summarizing claim history, tracking open items, and helping a client understand what actually matters.
That's the work where good people lose hours.
The mess sits close to the client.
A commercial account doesn't arrive as a clean data object. It arrives as emails, PDFs, loss runs, schedules, contracts, expiring policies, spreadsheets, notes, and half-remembered context from the last meeting.
The broker and service team have to turn that into a usable account story. What changed? What coverage matters? What needs a market explanation? What's missing? What's a real issue versus background noise?
This is where AI can help if it's pointed at the actual work. Not by replacing judgment. By giving judgment a better starting point.
Brokerage work is full of repeatable friction.
Every brokerage has its own style, but the friction points rhyme.
- Submission intake takes too long because the source material is scattered.
- Renewal prep starts late because nobody has a clean account brief.
- Policy review depends on someone manually comparing forms and endorsements.
- Claims context gets pulled only when a client or market asks for it.
- Service follow-through lives across too many inboxes and notes.
These aren't glamorous problems. They're also not small problems. They shape margin, client experience, producer capacity, and retention.
The test is whether the work changes.
I'm not that interested in AI demos that produce a clever answer once. I'm interested in whether the daily workflow gets better.
Can the account manager find the policy answer faster? Can the producer walk into the renewal meeting with a cleaner view of the account? Can the team see what's missing before it becomes urgent? Can a claim issue be tied back to the policy language without three people losing an afternoon?
If yes, that matters. If the only answer is that the system sounds impressive, it probably doesn't.
The broker still owns the call.
Insurance has too much nuance, accountability, and regulation for the human layer to disappear. The good broker still has to decide what matters, what to ask, what to challenge, and how to explain the answer to a client.
But the broker shouldn't have to start from a blank page every time. A better system can pull the context forward. It can organize the mess. It can surface the issue. It can make the next human decision easier.
That's why brokerage is such an interesting place to build. The work is messy enough to matter, repeatable enough to improve, and close enough to the client that better workflow turns into better service.
That's a real opportunity. Not because the technology sounds impressive. Because the work needs help.