Real estate does not have a lead-generation problem. It has a lead-answering problem. Brokerages pay for portals, ads and signs that make the phone ring, then let most of those calls end without a conversation with an agent. An AI voice agent for real estate is worth buying for one reason: every caller gets an answer, a next step and a named person to follow up, within the same minute.
This guide is for brokerage owners, team leaders, developers' sales offices and property managers: what the latest data shows, where a phone inquiry leaks, which calls to automate, the rules that belong in code, a break-even model and a vendor checklist.
What the 2026 lead-handling data shows
- Consumers still hire agents. In the National Association of Realtors' 2026 Profile of Home Buyers and Sellers, covering transactions from July 2025 to June 2026, 90% of buyers used an agent or broker and 92% of sellers used an agent, an all-time high (NAR, November 5, 2026).
- Most phone inquiries never reach one. In June and July 2026, the customer-experience research firm Pied Piper submitted 2,198 phone and website inquiries to local offices of 21 large real estate brands and scored the response over the following 24 hours. 97% of calls connected to the office, but only 53% of callers reached a representative quickly and only 33% spoke with an agent during the call. 73% received no follow-up by phone, text or email within 24 hours (Real Estate News, October 6, 2026; Florida Realtors, October 13, 2026; study by Pied Piper).
- The web is worse. 68% of website inquiries got no response within 24 hours and roughly one form in ten did not work. Average scores were 30 out of 100 for phone and 11 for web.
Read together: nine buyers in ten will work with an agent, so the question is which one. Pied Piper's own commentary makes the point that new clients weigh several options, and a slow response gives them time to connect with someone else.
A brokerage that answers 97% of calls but connects only a third of callers to an agent does not have a staffing problem. It has a routing problem, and routing is exactly what software does well.
Caveat: the study covers large US brands, not independent teams or developers. Use it as a reason to measure your own office, not as your baseline.
Where a real estate phone inquiry leaks
The study's figures map onto a funnel, and each stage needs a different fix.
| Stage | 2026 average | Common causes (not measured by the study) | What fixes it |
|---|---|---|---|
| Call connects to the office | 97% | Rarely a problem. | Nothing to fix. |
| Caller reaches a representative quickly | 53% | Hold queues, phone menus, a front desk busy with walk-ins. | Answer on the first ring, every time, including evenings and weekends. |
| Caller speaks with an agent | 33% | The office hands out an agent's name and number and leaves the next call to the consumer. | Qualify the caller, then book a time or warm-transfer to the right agent, instead of passing a number. |
| Follow-up within 24 hours | 27% | Nobody owns the lead once the call ends. | Log every call in the CRM with a summary, an owner and a deadline the manager can see. |
The 27% is the inverse of the 73% who heard nothing. Pied Piper found that some offices simply handed callers an agent’s name and number, while stronger performers were more likely to make the connection themselves or follow up proactively. That is the behaviour to automate.
Which calls an AI voice agent for real estate should handle
Rank each call type by how often it arrives and how much harm a wrong answer causes. A starting matrix:
| Call type | AI role | Why |
|---|---|---|
| Listing questions: price, beds and baths, status, open-house times | Resolve | Only if the data is read live from your listing feed on each call. A listing that went under contract this morning must not be offered this afternoon. |
| Showing and viewing requests | Resolve within rules | Book against the agent's real calendar, and run your buyer-agreement step before any tour is confirmed (see the rules below). |
| New buyer and renter inquiries | Assist | Capture area, budget, timeline and whether the buyer already has financing, then book a call with an agent. No financial advice and no judgement on whether they qualify. |
| Seller and valuation requests | Assist | High-value leads. Capture the address and motivation, then put a listing appointment in a senior agent's diary the same day. |
| New-development and off-plan inquiries | Resolve within rules | Availability and price lists come from the developer's live inventory, never from a brochure in the prompt. Show-home visits are booked directly. |
| Tenant maintenance and repair calls | Resolve | Log the request, confirm access details and give a reference. Emergencies such as gas, flooding or no heat go straight to the on-call line. |
| Offers, negotiation, commission and contract questions | Keep human | Licensed judgement and legal terms. The AI routes it to the agent of record. |
| Questions about who lives in a neighbourhood, or whether an area is “safe” or “good for families” | Keep human, with a set answer | Fair-housing risk. One approved neutral answer, never an opinion. |
“Keep human” still means the AI answers on the first ring and hands a complete case to the right person. The cost of unanswered calls is mostly lost revenue, not handling time, as we show in the hidden cost of call center overflow.
Rules to put in code, not in the prompt
A real estate AI that answers confidently and wrongly is worse than voicemail. A wrong price becomes a complaint; a careless sentence about a neighbourhood becomes a fair-housing problem. These rules belong in deterministic logic around the model, for the reasons we set out in Confidently Wrong. This is buyer guidance, not legal advice; confirm each point with your broker of record and counsel.
- Fair housing is a hard boundary. The federal Fair Housing Act makes it unlawful to make any statement about the sale or rental of a dwelling that indicates a preference, limitation or discrimination based on race, color, religion, sex, handicap, familial status or national origin (42 U.S.C. § 3604(c)). State and local laws often add more protected classes. Detect these topics, answer with approved neutral wording and log the call for review.
- No tour without the buyer-agreement step. Since August 17, 2024, an MLS participant working with a buyer must have a written agreement with that buyer before touring a home, in person or by live virtual tour (NAR, Written Buyer Agreements 101). If the AI books showings, it should check whether an agreement exists and, if not, trigger your process for the agent to send one before the appointment is confirmed.
- Listing data is live or it is not said. Price and status come from your listing feed on every call. If the lookup fails, the agent books a call-back; it never improvises.
- Say it is an automated assistant, and offer a person. “Can I speak to an agent?” is honoured at once, with the context passed along.
- Inbound first, outbound later. In February 2024 the FCC confirmed that AI-generated voices count as “artificial” voices under the Telephone Consumer Protection Act, so automated outbound calls generally need the called party's prior consent (FCC Declaratory Ruling 24-17). Calling old leads is a separate project, scoped with counsel.
How many closings pay for it: a break-even model
For a brokerage, the honest business case is not cost per call. It is how many extra transactions you need for the service to pay for itself, and whether that number is believable. Every input below is illustrative except the Pied Piper average; replace each one with your own phone-system and CRM data.
If your break-even needs more closings than the recovered calls could plausibly produce, the case does not hold, and you should know that before you sign. Two cautions: some unanswered callers call back or reach another of your agents, so measure call-backs first; and the AI does not close deals. It creates conversations; your agents still convert them.
For reference, averages across Unicall's live enterprise deployments, which are not real estate results, are a 100% answer rate, about 60% of calls resolved autonomously and cost per ticket down 52% within 90 days, with sample sizes and periods varying by client. A reason to test on your own calls, not a brokerage forecast.
A 30-day pilot scorecard
Mystery-shop your office on the funnel stages above before go-live and again after 30 days. Then track:
- Answer rate on the first ring, in and out of business hours.
- Speed to agent: minutes from the first call to a live conversation with a licensed agent.
- Appointments booked per 100 lead calls, and the share that actually took place.
- Follow-up inside 24 hours, measured from the CRM, not from agent self-reporting.
- Wrong answers found in a weekly transcript review: wrong price or status, wrong booking, any fair-housing flag.
Buyer checklist: what to ask any vendor
A good vendor answers each point with a recording or a log, not a slide.
If you are comparing specialist voice platforms, our comparison hub applies the same ten buyer criteria to each vendor, using public information only. Dealerships face the same leak; see our guide for car dealerships.
FAQ
What does an AI voice agent for real estate do?
It answers every buyer, seller, renter and tenant call on the first ring, 24/7: listing questions from your live feed, showings and valuation appointments booked in agents' calendars, new inquiries qualified on area, budget and timeline, maintenance requests logged. Offers, negotiation, complaints and sensitive topics go to a licensed person with the context captured.
How many real estate phone inquiries go unanswered?
In Pied Piper's 2026 Lead Handling Effectiveness study, which tested 2,198 phone and web inquiries to offices of 21 large US real estate brands in June and July 2026, only 33% of callers spoke with an agent during the call and 73% received no follow-up within 24 hours.
Is an AI phone agent compatible with fair housing rules?
It can be, if fair-housing topics are handled by rules rather than left to the model. The Fair Housing Act prohibits statements indicating a preference, limitation or discrimination based on protected characteristics, and state laws often add more. Define the triggering topics, give the agent one approved neutral answer, log every flagged call for review and confirm the approach with your broker of record and counsel.
Can an AI voice agent book property showings?
Yes, against agents' real calendars and your routing rules. Since August 17, 2024, an MLS participant working with a buyer must have a written agreement with that buyer before touring a home, so the booking flow should check for an agreement and trigger your process to send one before a tour is confirmed.
Should a brokerage use AI to call old leads?
Start with inbound calls, where the caller is already engaged. The FCC confirmed in February 2024 that AI-generated voices are artificial voices under the TCPA, so automated outbound calls generally need prior consent. Treat outbound follow-up of portal or past leads as a separate project, scoped with counsel.
Unicall.ai builds and operates done-for-you AI voice agents that answer every call 24/7, take real actions in your systems, and hand over to your people, with full context, the moment they should.