An AI voice agent for real estate is software that phones prospects, speaks with them in natural language, asks the qualifying questions a telecaller would ask (purpose, budget, configuration, location, timeline, visit willingness), answers factual project questions from material you give it, and then books a site visit, fixes a callback or hands a hot buyer to a salesperson. It is best used as a first-touch qualifier on fresh enquiries and old lists. It does not replace the negotiation, the site visit or the closing conversation.
Key takeaways
- Give the AI the calls with a few known outcomes: first response, requalifying old leads, visit confirmations. Keep negotiation and objections with people.
- The agent is only as good as its script and price sheet. Six questions, eight turns, current data.
- A hot buyer should reach a human while still on the line, with a fallback if the salesperson misses the transfer.
- DND scrubbing, calling hours, recording disclosure and honesty about being an AI are not optional.
What does an AI voice agent actually do on a real estate call?
The job is qualification, not closing
Treat the agent as a disciplined telecaller who never skips a question and never forgets to write notes. It introduces itself, says why it is calling (the portal enquiry yesterday, the ad form, last month's walk-in), works through a short script, and answers factual questions (“2 BHK ka starting price kya hai?”, “Is it RERA registered?”, “How far is the metro?”) from documents you have loaded. It ends the call one of four ways: visit booked, callback fixed, marked cold with a reason, or transferred to a salesperson.
It should not quote discounts, promise a specific unit or argue with someone who says no. Those are human decisions, and a script that permits them will cause problems.
A realistic call flow (illustrative)
The shape of a two-minute qualification call for a Gurugram 2/3 BHK project. The dialogue is illustrative, not a recording.
Agent: Namaste, main Riya bol rahi hoon ABC Developers ki taraf se. Aapne kal 99acres par Sector 79 project ke liye enquiry ki thi. Do minute baat kar sakte hain? Yeh call quality ke liye record ho rahi hai. Prospect: Haan, boliye. Agent: Aap khud rehne ke liye dekh rahe hain ya investment ke liye? Prospect: Self use. Agent: Budget approximately kitna soch rahe hain? 2 BHK yahan 1.4 crore se shuru hota hai. Prospect: 1.5 tak theek hai. Possession kab hai? Agent: Tower B mein December 2027. Ready-to-move chahiye to Tower A mein kuch units hain. Kya is weekend site visit kar sakte hain? Prospect: Saturday 11 baje. Agent: Done. Main abhi aapko sales manager Arjun se connect karti hoon jo visit confirm karenge. Ek second.
Budget fits, purpose is self-use, timeline is now, visit agreed: this is a hot lead, handed to Arjun while the prospect is still warm. The CRM record shows every answer, the slot and the transfer. A Dubai version runs the same way in English or Arabic: off-plan or ready, budget around AED 1.5 million, payment plan or mortgage, viewing at the sales centre on Saturday.
Which calls should the AI handle and which should stay human?
| Call type | AI | Human | Why |
|---|---|---|---|
| First response to a fresh portal or ad lead | Yes, within minutes | Follows after | Speed beats polish on first touch |
| Requalifying leads older than 60–90 days | Yes | Only if hot | Low hit rate, costly for telecallers |
| Site visit confirmation the evening before | Yes | – | Yes / no / reschedule |
| Cold list from an exhibition | Yes, with consent and DND checks | – | Volume work |
| Price or payment plan negotiation | No | Yes | Needs authority |
| Objection handling (“too far”, “overpriced”) | Light touch | Yes | Needs judgement |
| Channel partner coordination | No | Yes | Relationship, not qualification |
The pattern: the AI takes calls where the outcome is one of a few known options and the information is factual. Humans take calls needing authority, judgement or a relationship.
How should an AI caller qualify a property buyer?
The six questions that decide temperature
- Purpose: self-use, investment or rental yield.
- Budget: a range, and whether it includes stamp duty and registration (India) or DLD fees (UAE).
- Configuration or type: 2 BHK, 3 BHK, villa, plot, shop, office.
- Location: the specific micro-market, and how flexible.
- Timeline: this month, this quarter, or “just looking”.
- Site visit: will they come, when, and who else decides (spouse, parents, business partner).
An optional seventh is financing: loan pre-approved, cash, or undecided. In the UAE, end-user versus investor and residency status change the conversation enough to ask early. Keep the script to eight turns or fewer; every extra question costs completion rate.
What “hot” should mean
Decide the rule before the agent goes live and write it down. A workable version: budget within range, timeline inside 90 days, and either a visit agreed or a specific unit question asked. Budget mismatch or “not for a year” is warm or cold. The agent applies the rule mechanically and your team sees the reason, not just the label. See how AI lead categorisation works in a real estate CRM for how temperature and lead score fit together.
How does the handover to a human work?
Warm transfer, for hot buyers. The agent says it is connecting the prospect to a named colleague, dials them and bridges the call. The prospect does not hang up, wait or repeat themselves if the salesperson can see the answers on screen. The risk is a missed pickup, so the agent needs a fallback: book the visit itself, send a WhatsApp confirmation and raise a high-priority follow-up.
Callback, for warm leads. The agent fixes a time, writes the summary and the follow-up lands in the salesperson's queue. Enough when the buyer is six months out.
Either way, three things must happen in the CRM: the lead is assigned to a named person, that person is notified on their phone, and the transcript and summary are on the record before they call back. A lead follow-up system with reminders and SLAs already in place makes this routine.
What guardrails do you need before switching it on?
- DND scrubbing. In India, promotional calls to numbers on the National Customer Preference Register are restricted; check current rules on the TRAI website. In the UAE follow the telecom regulator and your brokerage's compliance policy. Keep your own suppression list too: anyone who says “don't call” goes on it the same day.
- Calling hours. Stay inside reasonable local hours (many teams use 9 am to 9 pm) and respect Friday prayer time and public holidays in the UAE.
- Recording disclosure. If you record, say so at the start, every call.
- Honesty. If asked “are you a real person?”, the agent must say it is an AI assistant. Pretending otherwise damages trust and may create legal exposure.
- Consent for cold lists. Exhibition sign-ups with a consent line are fine; a bought database is not.
- Spending limits. Per-minute billing means a badly targeted 5,000-number campaign burns money fast. Set a daily budget and watch the first day.
How do you set up and run an AI voice agent well?
- Write the persona. Name, voice, language (Hindi, Hinglish, English, Arabic), tone, what it may and may not say.
- Load the knowledge base. Dated price sheet, payment plan, possession dates per tower, RERA number, distances, amenities, bank tie-ups. Remove anything stale; an agent quoting last quarter's price is worse than no call.
- Define the script in turns. Opening, disclosure line, six questions, close, with exact phrases for visit booked, callback fixed and not interested.
- Set the hot rule and the transfer target. Who receives transfers, by project or rotation, and what happens if they do not answer.
- Use a consistent business number that prospects can call back, with the provider's KYC done.
- Dry-run with your own team. Five colleagues call in as the haggler, the “send on WhatsApp” person, the one who switches to English midway, the one who asks for a discount.
- Pilot on 200 old leads before any fresh lead. Old leads cost nothing to lose and expose the script's weak points.
- Review transcripts daily for two weeks. Look for unanswered questions, talking over the prospect, and hot labels your team disagreed with. Fix, then widen to fresh leads.
Connect the outcome to the rest of the funnel. A booked visit should flow into your site visit management process with a WhatsApp confirmation and a reminder to the salesperson; a prospect who prefers chat should continue in your WhatsApp CRM rather than going cold.
What are the common mistakes?
- A 20-question script: completion drops and people hang up.
- Calling fresh leads before the knowledge base is ready: the first impression becomes “the bot didn't know the price”.
- No fallback when the salesperson misses the transfer: the hot buyer evaporates.
- Treating the agent's “hot” as final without reviewing the first few hundred.
- Measuring minutes instead of visits: talk time is a cost; visits booked and attended are the output.
How do you know whether it is working?
Track a small set of numbers per campaign and, where possible, compare the AI cohort with the same list worked by telecallers.
| Metric | What it tells you |
|---|---|
| Connect rate | List quality and calling-hour choice |
| Completion rate (reached the close) | Script length and tone |
| Hot rate, and hot leads your team agreed with | Whether the qualification rule is right |
| Visits booked per 100 connected calls | The real output |
| Visits attended per visit booked | Whether the agent over-promises |
| Transfers answered by a human mid-call | Handover discipline |
| Cost per visit booked (minutes × rate ÷ visits) | Economics versus a telecaller's day |
If visits booked look healthy but attendance is poor, the agent is booking people who were being polite; tighten the “when exactly” question. If the hot rate is high but your team disagrees, the budget question is too loose. Many of the reasons real estate leads don't convert show up in this table first.
How Homelead's voice agent qualifies and hands over
Homelead includes a voice AI agent inside its real-estate CRM, so the call, the lead record and the follow-up live in one place.
- Script and persona. You define the agent's persona and qualifying questions: budget, location, BHK or property type, timeline, purpose and site-visit willingness. Eight natural neural voices are available (four female, four male), speaking English, Hindi and Hinglish. If a prospect asks, the agent says honestly that it is an AI assistant.
- Answers from your own material. It answers from the Knowledge Center: brochures, price sheets, PDF, Word and Excel files, web links, plus property and inventory records that sync automatically when they change.
- Hot buyers reach a person mid-call. When a prospect qualifies, the agent warm-transfers the call to a colleague's phone, and every handoff is logged on the lead. Hot leads are assigned and the owner gets a push or Telegram alert. The agent can also book the site visit itself.
- Everything lands on the lead. Every voice-agent call and handoff is logged on the lead timeline, and the fixed-rubric lead score (0–100) moves with the answers: budget match, timeline, interest, site visit.
- Guardrails built in. The agent speaks a recording disclosure by default and marks consent only when it was actually said; campaign calls respect a DND list and a 09:00–21:00 IST calling window; a call is not placed unless the wallet can cover at least a minute.
- Numbers and billing. Calls go out from business numbers set up in AI Center → Phone Numbers (KYC, choose number and lines, pay from the wallet). There is no flat AI fee: voice minutes are pay-per-use from the usage wallet, with your account's rates shown on the Wallet screen, alongside the per-user plan.
To see it run on your own projects, book a demo and send your inventory sheet beforehand.
Frequently asked questions
Can an AI voice agent really qualify real estate leads?
Yes, for the factual part: purpose, budget, configuration, location, timeline and visit willingness. It asks consistently, records every answer and applies a fixed hot/warm/cold rule. It cannot negotiate or read hesitation the way an experienced salesperson can, which is why hot buyers should be handed over.
Will prospects hang up when they realise it is an AI?
Some will, and the agent should be honest if asked. Hang-ups fall when the opening names the specific enquiry the person made, the voice is natural, the script is short and the call offers something useful such as a price or a visit slot. Cold lists see more hang-ups than fresh enquiries.
How does the handover to a salesperson work?
The best pattern is a warm transfer while the prospect is still on the line, with the answers already visible to the salesperson. If the salesperson does not pick up, the agent should book the visit or callback itself, write the summary and alert the lead owner.
What does an AI voice agent cost to run?
Usually per minute of talk time plus a monthly charge for business numbers, rather than a per-seat fee. Judge it on cost per site visit booked against a telecaller's loaded cost for the same list, not on cost per minute.
Is AI calling allowed under India's DND rules?
Promotional calls to numbers on the National Customer Preference Register are restricted whether a human or an AI dials. Scrub lists against DND, keep to reasonable hours, honour your own opt-out list and check current TRAI rules before running campaigns.