A real-estate chatbot picks up messages from people interested in a listing, on your website, WhatsApp or Facebook, including evenings and weekends. It asks the qualifying questions (budget, area, rooms, financing), answers from your own listings data, proposes viewings in the right agent's calendar and hands the lead to your CRM. It does not negotiate prices and does not give legal or financial advice.
Real-estate enquiries come in waves, after campaigns and outside office hours, while agents are at viewings, not at their phones. An AI chatbot can take over the repetitive part of that flow: triage, routine answers, scheduling. Below you see how it connects to listings and the CRM, what a qualification scorecard looks like, what to know about GDPR and when it is not worth building.
Where it helps and where it does not: volume decides
A chatbot makes sense when you receive more enquiries than a person can answer on the same day, especially across several channels and outside office hours. Under a few dozen enquiries a month, a good form and a fast reply from you cost less and work just as well.
Signs you have crossed the line: the same questions (floor, year, parking, VAT, mortgage) typed dozens of times, serious buyers mixed with browsers in one queue, portal notifications, Facebook messages and missed calls with no overall picture, viewings negotiated over ten messages.
The qualification scorecard: what the bot asks and what the agent receives
Qualification means finding out, in a few natural questions, whether the person wants something you have and can buy or rent it. The table below is an illustrative scorecard; the criteria and weights are yours to set, based on what a "good lead" means for your agency or development.
| Criterion | How the bot asks | What reaches the agent |
|---|---|---|
| Purpose | "Are you looking for yourself or as an investment?" | Own home / investment / rental |
| Budget | "Between which values are you looking?" | A range, not an exact figure |
| Area and type | "Which neighbourhoods interest you? How many rooms?" | Areas, rooms, preferences (floor, new/older building) |
| Financing | "Paying cash or taking a mortgage? Any pre-approval yet?" | Cash / mortgage needed / approved / not started (recorded only, no eligibility assessment) |
| Timeline | "When would you like to move?" | Under 2 months / 2–6 months / over 6 months |
| Contact | "What number can our colleague call?" | Name, phone, preferred channel, consent to be contacted |
| Summary | (generated from the conversation) | Two lines + link to the conversation + the listings discussed |
The practical rule: do not ask everything at once. A chat that feels like a 12-field form gets abandoned as fast as the form. The bot alternates questions with useful information (for example, two listings that already match), so the person gets something in return.
On top of these fields you can define priority rules: a lead with a matching budget, approved credit and a short timeline can trigger an immediate notification, while a "just looking" lead enters a slower follow-up sequence.
Want a scorecard adapted to your portfolio? Write to us and we build it together from your listings and your process.
Answers from your listings, not from the model's memory
A chatbot answers correctly about an apartment only if it reads that apartment's data. The technique is called RAG (retrieval-augmented generation): the bot searches your listings database for the relevant information and phrases the answer from it. The mechanism is also explained in our guide to what an AI agent is.
Structured fields (price, usable area, layout, floor, year, heating, parking, availability), free-text descriptions and documents such as a residential complex's amenities can be indexed. A buyer asks "Do you have a 3-room in Floreasca with parking?" and gets matching listings with links. When the information is not in the data, the bot says "I don't have that, a colleague will get in touch" and does not guess.
For critical details (price, area, availability) the bot should show the exact value from the data, not a paraphrase, and point to the source listing. Even with RAG, a model can phrase a detail wrongly.
Viewings in the calendar and routing to the right agent
The bot can read agents' availability from their calendar, propose two or three free slots, confirm and create the event with the address and client details. The logic resembles the booking chatbot for clinics.
In real estate, depending on the data and integrations, scheduling rules can be configured to account for travel time between viewings and slots when the agent is already busy. Compatible calendars (for example Google Calendar or Outlook) are settled during Discovery.
Routing follows your rules: area, property type (residential, commercial, land), budget band, language, or whichever agent owns that listing. The handoff goes into the CRM you have, with the conversation history and summary. If you have no CRM or the current one does not fit, see custom CRM vs off-the-shelf CRM.
Leads from portals, WhatsApp and Facebook
The chatbot does not replace the portals: it can pick up enquiries arriving through them and other channels and run them through the same qualification flow. How automatic that is depends on what each source offers.
- WhatsApp. This uses the WhatsApp Business API, not the regular app. The bot replies freely within the 24-hour window after the customer's last message; for messages you start outside it, Meta requires approved templates and the person's consent. Meta charges for template messages by category and country, and rates change periodically, so check its official price list. Details in our guide to WhatsApp Business automation.
- Facebook and Instagram. Click-to-message ads open a conversation in Messenger or WhatsApp, where the bot can continue. Lead Ads forms can be pulled into the CRM, depending on the integration.
- Real-estate portals. Enquiries usually arrive as email notifications or in the agent account. If the portal offers an export or API for agencies, it can be integrated; otherwise email notifications can feed the same flow. A portal's terms may forbid extracting data, so we check them first.
- Your own website. The widget on a listing page already knows which property the person means.
For portal enquiries, a fast first reply (minutes, not hours) is probably the most valuable thing the bot can do. There is no Romanian study we can cite on this, so treat it as a practical principle, not a figure.
GDPR for lead data
A real-estate chatbot processes personal data (name, phone, email, sometimes information about income or credit), so GDPR applies from the first conversation. What follows is a general overview, not legal advice; have the implementation checked by a lawyer or your data protection officer.
- Information. People must learn at the start who processes their data, for what and for how long. The bot shows a short notice and a link to the privacy policy.
- Minimisation. A viewing needs a name and phone number, not a personal ID number or ID card. Do not ask for exact income if a budget range is enough.
- Financial data. "I need a mortgage" concerns a person's financial situation. It is not a special category under GDPR Art. 9, but it deserves limited access and retention only as long as useful.
- Marketing and follow-up. Commercial messages by WhatsApp, SMS or email generally require prior consent. Replying to an enquiry the person sent is different from a campaign; keep the two flows clearly apart.
- Retention and deletion. Decide how long you keep leads that did not become clients and delete them afterwards. The system must be able to handle an access or deletion request.
- Suppliers. If conversations pass through an AI model API, check the data processing agreement (DPA), where processing takes place and whether data is used for training. EU hosting can simplify compliance but does not guarantee it.
- Transparency about AI. European AI rules introduce transparency duties for systems people interact with directly. Whatever the timeline, say upfront that this is a virtual assistant and always offer "I want to talk to a colleague".
What it does not do and where it breaks
The bot gives no legal or financial advice, does not negotiate price and is no better than the data it reads. These limits are set from the first quote.
- No legal advice. Questions like "can I buy with a preliminary contract without a notary?" or "what taxes do I pay on sale?" go to a person (agent, notary, lawyer).
- No financial advice. It does not say "the bank will approve your loan" and does not present an instalment as a promise. It can note that the person wants a mortgage and refer them to a partner broker, if you work with one.
- No negotiation. It states the listing price. To "do you offer a discount?" it hands over to the agent, who knows the owner's margin.
- Data decides everything. If an apartment sold yesterday still shows "available", the bot will book viewings for it. Answers are as accurate as your listings base, so we connect the bot to a single source of truth, and listings not updated for a long time can be flagged automatically, depending on data and integrations.
- It does not replace the agent. Buying a home is a big decision and people want to talk to someone at some point. The bot takes the repetitive stages; the relationship stays with the agent.
Cost and timeline
Cost depends on the number of channels, the state of the data (listings in a structured system or in spreadsheets and PDFs), the number of calendars and routing rules, the CRM and conversation volume. A simple chatbot with FAQ and lead capture usually costs 2,000–5,000 EUR and launches in 3–5 weeks; an AI agent with RAG over listings plus CRM and calendar integration costs 8,000–15,000 EUR, in 6–10 weeks. Monthly operation (API, monitoring, tuning) is 100–500 EUR and grows with message volume, not the number of listings. The full breakdown is in the AI chatbot cost guide; the final quote follows the Discovery stage, where we look at the data and channels. More on the AI chatbot service page.
The most common cause of delay is data preparation: who maintains the listings, in what format they are, and who approves the routing rules.
When it is not worth it
It is not worth it if listings are not kept up to date or nobody will read the qualified leads. A chatbot amplifies existing processes; it does not repair them.
- Messy data. If prices and availability live in WhatsApp messages and in agents' heads, you first need a clean listings base.
- No follow-up. A lead qualified at 11 pm and not called the next day is worse than an unqualified one. Decide beforehand who calls whom and within what time.
- Ultra-premium segment. If you sell a few properties a year, every client expects personal attention from the first message. The bot may stay limited to booking viewings.
A robotic tone can put customers off; our guide to chatbots in Romanian explains why and how to avoid it.
If you want to see what a chatbot would involve for your agency or development, contact us with your approximate enquiry volume and the channels you receive them on.
Frequently asked questions
Can the chatbot negotiate the price or promise a discount?
No. It states the list price from the listing and hands any discount question to an agent. Negotiation depends on the owner's margin and context.
How does it know which apartments are available?
It reads from your listings database (CRM, listings system or periodic export). Answer accuracy is data accuracy: a sold apartment left marked "available" will still be offered, which is why frequent syncing matters.
Can it check whether the client will get a mortgage?
No. It can record whether the person pays cash, needs a mortgage or has started the process, so you can prioritise the lead. Eligibility is set by the bank or a broker.
Can I store conversations with prospects under GDPR?
Generally yes, if you inform the person at the start, collect only the data you need, set a retention period and can delete data on request. The details depend on your situation; check them with a lawyer.
Does it connect to real-estate portals?
It depends on what your portal offers: an export or API for agencies, or only email notifications. We do not extract data in ways the portal's terms forbid.
Does it work only on WhatsApp or on the website too?
Both, with the same knowledge base. On the website the widget knows which listing is on the page; on WhatsApp the conversation continues on the channel many buyers prefer.