An internet lead from EverQuote, SmartFinancial or QuoteWizard lands in your agency inbox at 7:42 on a Tuesday evening. The producer who would normally call it went home at 5:30. The shopper who filled it out is still sitting at the kitchen table with the laptop open, and the same lead is on its way to several other agencies.
That minute is the whole job of an AI appointment setter. Here's what one does, step by step, and where the line is that it must never cross.
Step 1: it dials within seconds, inside calling hours
The setter receives the lead by webhook, vendor feed or form post and places a real phone call within seconds, about 10 seconds end to end in our case. It only dials inside permitted local calling windows. Federal rules bar telephone solicitations before 8 a.m. or after 9 p.m. at the called party's location (47 CFR 64.1200(c)(1), https://www.law.cornell.edu/cfr/text/47/64.1200, current as of 2026), and several states are tighter, so the window is set per state, not per office.
The reason the first minute matters so much for insurance specifically is that the lead is shared. A 2020 secret-shopper study by AgencyZoom, published by HawkSoft, submitted the same lead to 150 independent agencies. Only 6% responded within five minutes, 30% responded within an hour, and 34% never responded at all (2020). If the shopper's quote request went to four agencies, the first one on the phone usually gets the conversation. The setter's one real advantage is that it's never in a meeting and never at lunch.
Step 2: it confirms the person and says what it is
The call opens by confirming it's reached the person who asked for the quote, naming the agency, and saying it's an AI assistant. Whether disclosure is proactive or on request is a per-agent setting on our platform, but under either setting the agent never claims to be human. Utah's AI Policy Act, in force since May 2024, requires a clear disclosure when a consumer clearly asks whether they are talking to AI (2025), and the safer default for most agencies is to just say it up front.
Step 3: it qualifies on the four things a producer actually needs
A good setter isn't running a survey. It asks the handful of questions that tell a licensed agent whether this is a bindable opportunity:
- Coverage type: auto, home, bundle, renters, commercial, life.
- Currently insured, and with whom.
- Renewal date, which decides urgency.
- Homeowner or renter, which decides whether a bundle is on the table.
Those answers get written back to the agency management system or CRM with the recording and transcript. The producer picks up a lead that already has a shape, not just a name and a phone number.
Step 4: it hands off to a licensed agent, live or booked
If a licensed agent is available, the setter bridges the call on the spot. The agent's phone rings, they hear a short whisper describing who's on the line (auto and home, currently insured, renews in six weeks), and they press 1 to accept. The shopper never hears hold music or a transfer tone. If nobody is free, the setter books a call on the agent's calendar instead and confirms the time with the shopper before hanging up. More detail on the handoff is on our planned live transfer page.
What the setter must never do
This is where licensing rules and the technology meet, and I'll be blunt about it. The AI doesn't quote premiums, doesn't rate, doesn't bind, and doesn't advise on coverage. Those are licensed-producer activities in every state I know of, and no amount of model quality changes that. When a shopper asks "so what would my rate be?" the setter says it can't quote, and that the licensed agent it's connecting them to can. That's the scripted answer on our insurance agent, and you can't configure it away.
Same with lines of business. CMS marketing rules don't allow automated or AI-voice outreach for Medicare Advantage and Part D, even with permission to contact, so we don't run Medicare campaigns, full stop. Auto, home, commercial, life and final expense are all supported.
The consent question, in plain terms
In February 2024 the FCC ruled that AI-generated voices count as "artificial" voices under the TCPA (2024). Telemarketing calls with an artificial voice to a cell phone require prior express written consent, which the rule defines as a written agreement bearing the called person's signature that authorizes calls using an automated system or artificial voice (47 CFR 64.1200(a)(2) and (f)(9), https://www.law.cornell.edu/cfr/text/47/64.1200, 2026).
For purchased internet leads, that consent lives on the vendor's form. The major vendors' TCPA-compliant forms capture it, and the FCC's stricter one-to-one consent rule was vacated by the Eleventh Circuit on January 24, 2025, before it took effect (2025). What hasn't changed is the record-keeping. You need to be able to produce the consent language and the timestamp for every lead the AI dials. We check that with every agency before launch, and we don't dial a list that can't show it.
What this looks like at the agency level
A two-producer office buying 200 internet leads a month was never going to dial all of them inside five minutes. Not between walk-ins and a day that ends at 5:30. With a setter in front, every lead gets its first call within seconds, the producers only pick up the phone for qualified shoppers, and the lead that came in at 7:42 p.m. gets called at 7:42 p.m. The lead spend doesn't change. What changes is that you stop donating shared leads to whoever happened to be faster.