Multi-Agent AI Voice System for a Final Expense Brokerage
Zargham Haider
A US final expense insurance brokerage was growing faster than its phone team could handle. Everything ran through one number, and people answered every call by hand.
The Problem
One line for everyone. Buyers, policyholders, the brokerage's own agents and job applicants all called the same number, and staff had to work out who was who.
Licensed agents spent most of each sales call collecting details. Selling got whatever time was left.
When every licensed agent was busy, a buyer who was ready to pay had to wait or call back later.
Recruiting had a ceiling. Recruiters could reach about 200 applicants a day, which produced 6 or 7 licensed hires.
New agents had to learn the sales script and prove they could handle difficult buyers before taking real calls.
The Solution
I built five AI voice agents that share the line. Each one was designed to remove one of these bottlenecks, and they hand calls to each other and to the human team.
Every call answered and sorted
The hotline agent picks up every call. It says it is an AI, asks for recording consent, takes a name and callback number, and works out why the person is calling. Buyers and job applicants are passed to the right agent. Policy and general questions are answered from the company knowledge base.
It also listens for a death in the family, a cancellation, a complaint, a legal contact or a request for a human. When it hears one, it drops the script. A grieving caller gets a slower pace, no sales talk and a personal callback.
Staff no longer sort calls. They receive 26 fields of clean data after each one.
Front desk Customer support Agents
Customer is qualified before a human joins
The sales intake agent takes over when a caller wants to buy. I built it as a 23-step conversation flow. It confirms the caller is ready, then collects what the application needs: coverage, budget, identity, health, medications and payment method. It can give a quote on the call, for example $10,000 in coverage for $139 a month.
It then checks whether a licensed agent is free. If not, it keeps the caller talking and checks again, up to 10 times, or books a callback. The buyer is not left on hold or lost.
When an agent is free, the call transfers with all 70 fields already filled in. The licensed agent confirms and submits.
Pre Qualification Agents
Hiring without the call limit
The recruiting agent calls each applicant about two hours after they apply and runs the full first interview: 26 questions in 12 to 17 minutes, covering licensing, experience, availability, compliance history and pay. It ends with a hire or pass decision and a signing date.
An n8n workflow then analyzes the recording, emails the candidate, updates the dashboard and books a human recruiter for the next day. Recruiters now speak only to people who are already interviewed and scored.
New agents trained and tested
The training coach plays a buyer through all 10 sections of the sales script and gives feedback after each one. Three or more mistakes and the section is repeated.
The test caller then phones in as one of five difficult buyers, such as a skeptic who was burned before or someone grieving who shuts down if rushed. A workflow grades the call, emails the report and switches the persona for the next test.
Recruitment and Training Agents
Results
The hotline agent covers the work of three people and responds about 10 times faster.
Licensed agents talk about 90% less on each sale call.
Sales doubled in the intake agent's first three weeks.
Hiring Licensed agents went from 6-7 a day to 40-45 a day.
Recruiting reach went from about 200 calls a day to more than 2,000.
How the system was built
Retell AI for the voice agents, using single-prompt and conversation-flow builds with agent-to-agent handoff and warm transfer. GPT and Claude models. n8n for the workflows. Webhooks into a custom admin CRM. An underwriting and quoting API for conditions, medications and quotes.
Every call includes AI disclosure and recording consent. The agents never give product advice and never state a price before a real quote comes back.
Have a team answering phones all day? I can build AI voice agents like these for your business. Check out my Never Miss a Call service or send me a message.
Like this project
Posted Oct 2, 2026
A brokerage's phone team couldn't keep up. I built five AI voice agents that answer, qualify, interview and train. Sales doubled in 3 weeks.
