The 12-Cent Renewal: How AI Is Rewiring the Independent Agency

Artificial intelligence has been the loudest word in insurance for two years running, and most of it has been noise. Here is what it actually looks like when you put it to work on the least glamorous, most expensive job in the agency: the renewal list.

I am an independent agent. Twenty years in. I did not come to AI as a technologist. I came to it the way most agency owners do, staring at a book of business that keeps growing while the team stays the same size, wondering where the hours are supposed to come from. What I found is not a robot that replaces agents. It is something much more useful, and much more boring: a way to write your process down once and have it run itself, correctly, every single time.

In our latest AI + Agency session I walked through exactly how we did that, with real numbers off our own book. If you were there, this is your recap. If you missed it, the full replay is at the bottom of this post.

AI in insurance, minus the hype

Insurance runs on process. A renewal, a certificate, a cancellation, a new business submission: every one of these is a sequence of steps that a trained person follows the same way, over and over. That is exactly the kind of work AI is good at, and exactly the kind of work most agencies are still doing entirely by hand.

The tools people reach for first tend to be chat assistants. They are genuinely useful for brainstorming, for a quick rewrite, for a one-off question. But an agency does not run on one-off questions. It runs on repeatable procedures that have to come out identical whether it is the first file of the day or the fortieth. When you type the same request into a chat window forty times, you get forty slightly different answers, because the tool is improvising each time. That is the ceiling most agents hit, and it is why AI so often feels like a toy that never quite makes it into the daily workflow.

I build on Claude for one reason above all others. It is built to follow a written procedure and run it the same way every time, and it can reach into the systems you already use to do the work, not just talk about it. That reliability is the difference between a novelty and a piece of infrastructure. Before real client data ever moves, the same questions any principal should ask still apply: check your plan's terms, your data handling, and your E&O policy. But the mechanism itself is sound, and it is what makes everything below possible.

What a skill actually is

Here is the single most important idea from the whole session, and it is the one that unlocks everything else. A prompt is a request. A skill is a procedure the AI can run the same way every time.

Put another way: a prompt is you asking. A skill is documentation that happens to run. It is the written version of how your best CSR does a task, stored where the AI reads it before it starts working, so the two hundredth renewal comes out exactly like the first. When we say we "built a skill," what we really mean is we wrote our process down carefully enough that it could be executed without a person re-explaining it each time.

Every skill has four pieces, and once you see them you start seeing skills everywhere in your agency:

That last piece matters more than it sounds. You do not learn a new syntax. You talk the way you already talk. "Shop the Joneses' renewal." "Does NowCerts have this policy?" "Send them something they can sign." "Remind me to follow up Thursday." The phrase itself does the routing, and the right skill picks up the work.

The real power: skills that link together

A single skill is handy. Skills that hand off to one another are how an agency actually changes. This is the part most people miss, so it is worth slowing down on.

Think about how any real task moves through your office. Nobody does one step and stops. The information from one desk becomes the starting point for the next. A quote feeds a proposal. A proposal feeds an email. An email feeds a follow-up task. Each handoff is a moment where something can get dropped, re-keyed, or delayed while a file sits in someone's queue.

Linked skills remove the handoff without removing the steps. The structured data one skill produces becomes the clean input the next skill starts from, so the reshop and the proposal begin without anyone re-reading and re-typing the file first. One request can set off a chain: pull the data, fill the gaps, check the coverage, build the document, write the outreach, log the task. Six skills, one sentence from you, and what lands is a finished package rather than a pile of half-done steps waiting on the next available person.

The skill is documentation that happens to run. Chain two of them and you have a process that runs itself.

Where it earns its keep: the renewal

We put this to the test on the highest-volume, most repetitive work in the agency: personal lines renewals. It is the best place to start precisely because it is so repetitive, and because the cost of doing it by hand is easy to measure once you actually look.

So we looked. We took one renewal and timed the prep, step by step, the way it happens by hand. Building the review form, launching the email sequence, chasing the client, pulling the current file, re-keying it into the rater, building the comparison, logging the outcome. Eight steps, all of them on a person.

Eighty-two minutes and about forty-three dollars in payroll, per file, just to get to the starting line. Run that across ten renewals a week per CSR and you are looking at roughly $22,000 a year, per person, in prep alone. Automated, the same run costs about twelve cents in tokens. Same eight steps. Same deliverables. Two human touches instead of eight.

The whole process hangs off one form. We prefill it through the URL so the client only answers what we do not already know, and their response comes back as structured data, so the reshop and the proposal have a clean start without anyone re-reading the file. Six skills carry it from there. What lands on the agent's desk is a proposal PDF, a drafted email, and a follow-up task, all sitting in the queue with a person's name on it. Nothing sends itself. An agent still presses send, every single time.

Better outcomes for the client, at a lower cost to serve

Here is where the math stops being about payroll and starts being about clients. The reason this matters is not that it saves you money. It is that it lets you do the right thing for the client on every file, which by hand you simply cannot afford to do.

Take a real example from the session. A client is up for renewal at $1,700 with their current carrier at a 15 percent commission. Left alone, it renews and the agency keeps $255. But the market moved, and there is a carrier who will write the same risk at $1,400. Reshopping it saves the client $300. It also drops the agency's commission to $140, and doing that remarket by hand costs another $42.71 in payroll. So the "right thing for the client" quietly costs the agency real money, twice. That is exactly why so many renewals never get reshopped. Not because agents do not care, but because the time is not there.

When the reshop costs twelve cents instead of eighty-two minutes, that calculation changes completely. The client gets the $300 savings. The agency keeps the relationship, keeps the trust, and keeps a renewal it might otherwise have lost to a competitor who did shop it. The only thing you gave up was the prep time you were bleeding anyway. Better coverage decisions, better pricing for the client, and a lower cost to deliver all of it. Those used to be in tension. Skills put them on the same side.

You can see it in our own book over a single quarter. Prep time per renewal fell from 82 minutes to 9. Renewals actually reshopped went from 69 percent to 100 percent, because nobody had to find the time anymore. Lead time on a rate increase shrank, so clients heard from us earlier with options instead of later with a surprise. And client conversations per CSR per week more than doubled. The hour we stopped spending on prep went straight back into talking to people.

What the agent still owns

None of this works if it turns into a black box that fires quotes at clients. It does not, and by design. The skill gathers the data, rates the whole market, formats the client-facing document, and logs the work. The agent decides what coverage is right, reviews every quote before it goes out, makes the call, and owns the advice. That rule is not a suggestion bolted on afterward. It is written into the skill itself. No quote reaches a client without an agent looking at it first.

We also built for where it can go wrong. Stale data flagged before it gets quoted as current. Limits and deductibles matched to the expiring policy so the comparison is apples to apples. Compliance timeframes pulled from the actual statute reference, not from memory. Drafts that sit in the queue until a person presses send. The goal was never to remove the agent from the work. It was to remove the busywork from the agent.

Watch the full session