levelpartner.ai
Real estate

AI for real estate agents: how it works and how to set it up

AI for real estate agents earns its keep fastest on the phone. Property calls arrive while agents are in viewings, on the road or with a seller, and a buyer who reaches voicemail on a Saturday rings the next agency on the portal. This article answers one question: how should an office of 10 to 250 people set up a Voice AI agent that picks up property calls, books viewings and valuations into the right calendar, and logs every call to the contact record so the agent who rings back knows what was said?

An AI real estate agent is not an agent

The phrase "AI real estate agent" gets used loosely. In most markets, the activities that make someone an agent, such as advising on price, negotiating and acting for a client, carry licensing or professional duties. Software cannot hold those duties. What a voice agent can do is the work around them: answer, ask the standard questions, book, confirm and record.

That distinction shapes the whole design. A voice agent told to "handle calls like one of our negotiators" will be asked about price expectations and offers, and it should not answer. A voice agent told to answer listing questions from approved details, book viewings and valuations, and pass everything else to a person will be trusted by your licensed staff, because it never strays onto their ground.

Where the phone still loses business

Chat and SMS enquiries wait politely. Phone calls do not. A caller who gets no answer rarely leaves a message, and the ones who do leave a name and number with no context. Most offices have three phone gaps: evenings and weekends, the hour after a new listing goes live when the line rings constantly, and lunchtime when the front desk is covering viewings.

Illustration: a 50-person sales and lettings office counting missed calls

Imagine the office logs every inbound call for a month and counts about 1,400. Say 310 go unanswered: 140 after six in the evening or at weekends, 110 in the two days after each new listing launches, and 60 at lunchtime. If fewer than a third of those callers leave a message, more than 200 enquiries a month would vanish without a trace. The numbers are invented to show the method.

Those are the calls a voice agent should take. Not all calls, and not the call from a seller who wants to discuss a price reduction with the agent they know.

What the voice agent should and should not do

Write the job down before you open the platform. A good brief for a property office fits on one page.

The agent answers questions about listings from approved details: price, bedrooms, parking, availability for viewing. It asks a short set of qualification questions: buying or renting, chain position or move date, whether they have a property to sell. It books a viewing into the listing agent's calendar, or a valuation into the valuer's calendar. It takes a message for anything else and says who will call back and when.

It does not discuss offers, give opinions on value, comment on other buyers, or promise anything about a sale. Any of those triggers a transfer to a person in working hours, or a message and a callback task outside them.

Before you start

You need the View & Manage Voice AI Agents permission; without it, Voice AI does not appear in the menu. You need calendars set up for the people who take viewings and valuations, with real availability and travel time between appointments. You need a phone number you can assign to the agent, either an existing one or a new number bought from the deployment screen. And you need your one-page brief, because the prompt is where most of the work happens.

How to set up a Voice AI agent for property calls

  1. Go to AI Agents > Voice AI, then open the Agent List tab.
  2. Click + Create Agent, choose Create Custom Agent, then click Continue.
  3. Enter an Agent Name, then select the Voice, Model and Language.
  4. Set the Timezone using the clock icon in the toolbar.
  5. Write or review the prompt, then set the Welcome Message for inbound calls.
  6. Choose whether the agent or the caller speaks first, if the option is shown, and set the pause before the agent speaks.
  7. Click Save.
  8. Open Actions, click + New Action, select the Appointment Booking type, name it, define when it should run, connect the calendar and save.

Then put it on the phone line:

  1. Open the agent's Deploy tab.
  2. In the Phone section, select an eligible phone number or a configured Number Pool.
  3. Under Call Routing, choose Answer Calls Directly or Use as Backup.
  4. Under Working Hours, choose All Hours or Custom Schedule.
  5. Review the inbound call flow, forwarding, timeout, voicemail and fallback settings for the number.
  6. Save the deployment settings.

For most offices, Use as Backup is the right first choice. Your team answers as usual, and the voice agent picks up only the calls nobody reaches. A Custom Schedule that covers evenings and weekends is the other common pattern, and schedules can run past midnight.

Booking viewings and valuations

Use one booking action per kind of appointment, each with its own calendar and its own condition. A viewing books into the listing agent's calendar; a valuation books into the valuer's. If the voice agent offers a single generic appointment, the wrong person turns up or the right person gets no preparation time.

Illustration: what the same office could expect from a backup voice agent

Suppose the office deploys the voice agent as backup in working hours and on a custom schedule for evenings and weekends, and it picks up about 300 of the calls the team would have missed in a month. If 70 of those callers book viewings, 18 book valuations and 95 ask about offers, price or a sale in progress, the agent would leave 95 callback tasks for people. The 18 valuations are the line to watch, because each one is a chance at an instruction. These are assumed figures, not measured ones.

Logging the call to the contact

Every call the voice agent takes leaves a call log with the transcript, a summary, the recording and the result of any action, such as a booking. The summary can also be saved to the contact's notes, which is the part your agents will use. An agent who opens the contact before ringing back sees what the caller asked, which property it was about and what was booked. Keep that summary option switched on; if you turn it off, the summary stays in the call logs but no longer reaches the contact notes.

Once calls land on contacts, the rest of your system can work with them: a new caller becomes a lead in the pipeline, a booked valuation starts your pre-valuation emails, and a callback task lands with the right person. For how those leads are assigned and reviewed across a team, see real estate lead management, and for keeping each caller's record complete, see keeping track of real estate leads.

What most articles about AI for agents get wrong

They treat every channel as the same conversation. A chat bot and a voice agent need different scripts. On the phone, callers interrupt, give their name once and quickly, and expect short answers. A prompt written for chat, with long answers and three questions in one turn, sounds robotic out loud. Write the voice prompt as spoken English: one question at a time, short confirmations, and the property address read back before booking. For the written channels, conversational AI for real estate covers how chat and SMS conversations should be designed.

The trade-off

A voice agent is the most visible AI you can put in front of clients. A chat bot that misreads a question is mildly irritating; a voice that mishears an address is memorable. You give up some control over the first impression in exchange for answering calls you currently miss entirely. That trade is worth it for after-hours and overflow calls, and usually not for your best sellers ringing their own agent. Keep known clients routed to people.

Who owns the voice agent

Give it a named owner, usually the sales or lettings manager. They listen to a sample of calls each week, update the prompt when listings or policies change, and read every call that ended in a complaint. Change one setting at a time and retest, so you know what made the difference.

Troubleshooting

A number is assigned but the agent does not answer. Working Hours, Call Routing, forwarding and timeout all affect whether a call reaches it. Check that the schedule covers the time of the test call and that forwarding is not sending the call elsewhere first.

Call transfer does nothing in testing. Transfers do not run in a web call test. Use a phone call test or ring the deployed number.

Viewings book into the wrong diary. Each booking action needs its own calendar and a clear condition. Split viewing and valuation bookings into separate actions.

Agents say the summaries are missing. The call summary option has been turned off for contact notes. Switch it on so summaries reach the contact again.

Frequently asked questions

Can an AI real estate agent conduct viewings?

No. It can book them, confirm them and take a message afterwards. The viewing itself, and any advice given during it, stays with a person.

Should the voice agent answer every call?

Rarely. Start as backup for missed calls and after hours. Move to answering directly only for lines that carry mostly routine questions, such as a dedicated listings line.

Can it make outbound calls to old leads?

Outbound AI calls carry their own consent and disclosure rules, which differ by country and, in the United States, by state. Expect to disclose that the caller is an AI and to honour a spoken opt-out. Get inbound working first, then check whether outbound fits the rules where you operate.

What about rental and property management calls?

The same approach works with a separate agent and separate calendars, and with stricter limits on what it says about tenancies. Read AI for property management for that side.