Real estate is the cleanest ROI case in voice AI because the cost of a missed call is unusually high and unusually measurable. A voice agent answers every enquiry instantly, qualifies the caller, books the viewing and logs it to your CRM, for roughly $200–$800 a month off the shelf or $20k–$60k built to your systems.
A buyer enquiring about a property is in market right now and calling several agencies. Whoever answers first, and books the viewing, usually wins. That is the entire business case, and it does not require anyone to believe anything about AI.
Where the leak is
Agents are with clients, at viewings, or driving. Calls go to voicemail, and enquiry callers largely do not leave voicemails; they call the next listing. Evenings and weekends are when buyers browse and call, which is exactly when agencies are least staffed.
Model it for your own numbers rather than trusting a vendor's: multiply monthly missed calls by a realistic enquiry-to-viewing rate, a viewing-to-offer rate and your average commission. For most agencies the result is uncomfortable, and it is larger than any plausible cost of fixing it.
What the agent handles
- Answering instantly, every time. Including at 9pm on a Sunday, and including when five people call at once.
- Property details. Price, bedrooms, availability, features, pulled live from your listing system so it is never quoting a withdrawn property.
- Qualification. Budget, timeframe, finance status, whether they have a property to sell. The questions your agents ask, asked consistently every time.
- Booking viewings. Straight into the right agent's calendar, with confirmation and reminders.
- Callback capture. When the enquiry needs a person, it takes the details and books the callback rather than leaving a message nobody returns.
- CRM logging. Every call recorded, transcribed and attached to the lead. This alone changes what management can see.
- Outbound follow-up. Post-viewing feedback calls, which agents consistently skip and which materially affect vendor relationships.
What it should not handle
Offer negotiation, vendor management, anything where the caller is upset, and legal or contractual questions. The agent's job is to make sure the human conversation happens, and happens with someone qualified. An agency that lets AI negotiate is optimising the wrong step.
Cost and payback
| Option | Cost | Suits |
|---|---|---|
| Off-the-shelf voice receptionist | $200–$800 / mo | Small agencies, common CRM, standard flow |
| Custom voice agent | $20k–$60k build | Proprietary systems, outbound, multi-office routing |
| Enterprise multi-office | $60k–$120k | Multi-brand, multi-language, compliance requirements |
Payback rarely needs modelling. One additional transaction a quarter covers the annual cost of most configurations. The reason to build rather than buy is usually integration: if your listing data and calendars sit in a system the products do not support, or you want outbound follow-up on top of inbound, a build pays for itself in capability rather than price.
Compliance details that matter
- Disclose the AI. Required in a growing number of jurisdictions and better practice everywhere. It costs nothing in conversion when done confidently at the start.
- Recording consent. Two-party consent rules vary by state and country. Get the announcement right.
- Outbound calling rules. TCPA in the US, do-not-call registers elsewhere. Outbound is where the regulatory risk concentrates, and it needs opt-out detection built in.
- Fair housing and equivalent rules. The agent must not steer, and must not ask or infer questions about protected characteristics. Script and test this explicitly.
Why latency decides quality
A property enquiry is a conversation with interruptions: callers change their mind mid-sentence, ask two things at once, cut in. An agent that cannot be interrupted reads as a phone tree, and callers hang up on phone trees. The engineering that makes this work is barge-in handling and end-to-end latency under roughly 300ms, which we have written up in sub-700ms voice agents and barge-in detection.
When you evaluate vendors, call the demo line and interrupt it. That single test separates the products faster than any feature comparison.
The same template resells
Property is the clearest version of a pattern that repeats: high-value enquiries, staff who cannot answer phones, and a booking as the conversion event. Clinics, dealerships, trades and field service are the same system with different qualification questions. If you are in one of those, the maths above transfers with minor changes.
Our voice AI development page covers how we build these, and voice AI ROI has the measurement framework.
Common questions
What does an AI voice agent for real estate do?
It answers every enquiry call instantly including out of hours, quotes live property details from your listing system, qualifies the caller on budget, timeframe and finance, books viewings directly into the right agent’s calendar, logs everything to your CRM, and can make outbound post-viewing feedback calls.
How much does an AI voice agent cost for a real estate agency?
Off-the-shelf voice receptionists run $200–$800 a month and suit small agencies on common CRMs. A custom agent integrated with proprietary listing systems and calendars, including outbound follow-up, costs $20,000–$60,000 to build. One additional transaction a quarter typically covers the annual cost.
Should an AI voice agent negotiate offers?
No. Restrict it to answering, qualifying, booking and logging. Offer negotiation, vendor management, upset callers and any legal or contractual question should route to a person. The agent’s value is making sure the human conversation happens quickly and with someone qualified, not replacing it.
What compliance rules apply to AI calls in real estate?
Disclose that the caller is speaking to an AI, handle call-recording consent according to local one-party or two-party rules, follow outbound calling regulations such as TCPA and do-not-call registers with automatic opt-out detection, and explicitly script and test against fair housing rules so the agent never steers or asks about protected characteristics.