TL;DR: Real estate runs on speed-to-lead, and most leads arrive as a phone call nobody picks up. I built production voice and chat agents for real estate at Fortell AI in the UK, and this is the honest playbook: what the agent should own (qualifying leads, booking viewings, answering listing questions, following up), the compliance line it must never cross, and the Retell plus n8n stack I ship it on. The line that matters most in real estate is not an emergency, it is Fair Housing. Get that wired in explicitly or do not ship the agent at all.
Every "AI receptionist for realtors" pitch talks about booking more showings. Fewer of them talk about the call where a prospect asks whether the neighborhood is "family-friendly" or "safe," and the agent, trying to be helpful, says something that just became a Fair Housing violation. Booking is the easy part. Knowing which questions to answer, which to route, and which to refuse politely is the whole job. Here is how I build one that does all three.
Why real estate is a strong fit for voice AI
Real estate has a brutal, well-documented truth: the agent who responds first usually wins the lead. A buyer who fills in a form or calls about a listing is, in that moment, the most interested they will ever be. An hour later they have called two other agents. A day later they have booked with one of them. And most agents are showing a property, driving, or in a closing when the phone rings, so the call goes to voicemail and the lead goes cold.
The call mix is also unusually repetitive, which is exactly what voice agents are good at:
- Inbound listing enquiries: price, availability, square footage, is it still on the market, can I see it.
- Qualifying leads: buyer or seller or renter, timeline, budget, pre-approved or not, area of interest.
- Booking and rescheduling viewings against the agent's live calendar.
- Outbound follow-up and nurture on leads that went quiet.
- Basic FAQs: open house times, application process for a rental, what documents a viewing needs.
None of that needs a licensed human in real time. All of it decides whether a lead survives the first hour. When I built agents for real estate clients at Fortell AI, the goal was never to replace the agent's relationship with a client. It was to make sure no inbound lead ever hits a voicemail, and every one arrives qualified and booked.
What a real estate voice agent should actually do
I draw a hard line between what the agent owns and what it must escalate. In real estate, getting that line right is not just about customer experience, it is about legal exposure.
The agent owns these
Lead capture and qualification. The agent answers every inbound call, captures name and contact, and qualifies: are they buying, selling, or renting, what is their timeline, what area, and are they financing or cash. It writes a clean, qualified lead into the CRM so the human agent picks up a warm, sorted contact instead of a voicemail with no context.
Booking and rescheduling viewings. The agent checks the calendar live, offers real slots, books the showing, and sends a confirmation text. This is the core value and it works reliably. I keep the booking logic out of the voice prompt and in the automation layer, for reasons I will get to.
Factual listing questions. Price, address, number of bedrooms and bathrooms, square footage, is it still available, HOA fee, when it was listed. These are objective facts pulled from the listing data. The agent reads them out or texts a link. It never editorializes about them.
Outbound follow-up. The reverse direction: the agent calls back leads that went quiet, confirms tomorrow's viewings to cut no-shows, and re-engages an old list. Speed-to-lead is the headline, but consistent follow-up is where a lot of the recovered revenue actually hides.
The agent must escalate these
Anything that touches Fair Housing. This is the real estate equivalent of the emergency line in a vet-clinic agent. If a caller asks whether a neighborhood is "safe," "family-friendly," "the right kind of area," what the schools or demographics are like, or anything that steers by a protected class, the agent must not answer. Under the Fair Housing Act, steering is illegal, and an AI that helpfully describes "the kind of people" in an area has created liability with a straight face. I wire this as an explicit branch: the agent gives a neutral, scripted response ("I want to make sure you get accurate information on that, let me have one of our agents follow up with you directly") and routes to a human. No opinion, no guess.
Price negotiation, offers, and commitments. The agent never negotiates price, accepts or conveys an offer, or commits the seller to anything. A number spoken by an automated agent is a number a client will hold you to. It captures the interest and hands the negotiation to the licensed human, full stop.
Legal, contract, and financing advice. Contingencies, contract terms, whether an appraisal will come in, mortgage specifics. These need a licensed professional. The agent collects the question and routes it rather than improvising something that sounds authoritative and is wrong.
Speed of handoff on all three is the safety feature. The agent's competence boundary has to be a hard, explicit branch, not a hope that the language model behaves.
The stack I ship it on
The architecture is the same one I use for every production voice agent, tuned for a brokerage. I wrote the full version in how I build production AI voice agents with Retell, so here is the real-estate-specific shape of it.
- Voice platform: Retell AI. It handles speech-to-text, the conversation flow, and text-to-speech. The conversation-flow model makes the Fair Housing branch easy to express as an explicit, high-priority path and easy for a broker to read and sign off on later. I compared it against the alternatives in Retell vs Vapi vs Bland.
- Automation layer: n8n. Every action the agent triggers lives here, not in the voice prompt: qualifying and writing the lead, checking the calendar, booking the viewing, sending the confirmation SMS, kicking off the follow-up sequence. The full pattern is in how to connect Retell AI to n8n. Keeping logic in n8n means a brokerage's routing rules can change without anyone touching the agent.
- Telephony: Twilio, with A2P 10DLC registration so the confirmation and follow-up texts actually get delivered rather than silently filtered.
- CRM as the system of record. Leads land in the brokerage's CRM, not in the voice platform's logs, so a qualified lead is a real object the team can work. I built out exactly why that back half matters in the dashboard and CRM behind a voice agent.
At Fortell AI I also built a pipeline to spin up and test these agents in minutes using Claude Code and Comet browser automation, which meant a real estate agent could go from brief to a testable draft the same day, then get hardened against real edge cases, including the Fair Housing traps, over the following week. I wrote that build up in how I build and test voice agents with Claude Code and Comet.
The real-estate-specific gotchas I have hit
Every vertical has its own failure modes. These are the ones that bite in real estate.
The "helpful" Fair Housing slip. The single biggest risk. A model's instinct is to answer the question it was asked. Asked whether an area is "good for families," an unguarded agent will happily oblige and steer. The fix is not a softer prompt, it is a hard branch that recognizes the category of question and refuses to answer it, every time, regardless of phrasing.
A number spoken is a number owed. If the agent quotes a price, a rent, or a fee, it must come from the listing data, exactly, and never from an estimate. Anything about negotiation, "would they take," or "is there flexibility" is an instant route to a human. The agent states facts and hands off judgement.
Which listing are they even calling about. A brokerage runs many listings at once, and callers say "the house on Oak Street" or reference a photo they saw. The agent has to match the enquiry to the right property before it answers anything, or it confidently reads out the wrong address and price. I make it confirm the specific listing first.
After-hours is where the leads are. A lot of real estate calls come in evenings and weekends, exactly when the agent is off. The voice agent's highest-value hours are the ones a human is not working, so the after-hours flow has to be as sharp as the daytime one, capturing and booking rather than promising a callback.
Do not double-text an active client. If you bolt on a missed-call textback, filter it so a client you are mid-deal with does not get a cold "sorry we missed you, are you looking to buy or sell?" text. I covered that build and its filters in missed-call textback with GoHighLevel and Retell.
What it costs and what it returns
Running costs are usage-based and modest: the voice platform, the LLM, and telephony together land in the range of a few cents to around twenty cents per call minute, plus a platform or service fee. I broke the full economics down in how much an AI receptionist costs in 2026.
The return in real estate is stark because the ticket size is enormous. A single converted buyer or seller is a commission measured in thousands, sometimes tens of thousands. An agent that catches even one otherwise-missed lead a month, before you count the reduced no-shows from viewing reminders and the hours of qualifying handed back, has paid for itself many times over. The point is not that the agent is cheap. It is that in a business where the first responder wins, a missed call is a competitor's commission.
Where to start if you run a brokerage
Do not try to automate everything on day one. Start with the single highest-value, highest-frequency call: inbound lead capture and qualification, wired straight into your CRM. Get that reliable, wire the Fair Housing and negotiation branches so they are bulletproof, and only then layer on viewing bookings, outbound follow-up, and after-hours coverage. A voice agent for real estate earns its keep the same way a great inside-sales rep does, by making sure no lead ever hits a voicemail, arriving qualified, and knowing exactly which questions are not its to answer.
I build production AI voice agents for real estate and service businesses, including the real estate work I did at Fortell AI. If you want one built properly for your brokerage, with the compliance and escalation done right, book a call.