AI for Real Estate Agencies: Portal Leads, Viewings, Follow-Up

AI for Real Estate Agencies: Portal Leads, Viewings, Follow-Up

AI for real estate agencies works best as a first layer of customer service. It answers the portal lead right away, confirms interest, collects what the broker needs to know and proposes viewing times. Negotiation, prices, payment terms and any promise about the property stay with the broker. The gain is in not letting the lead go cold and in handing the broker a conversation that is already organized.

Key takeaways

• The AI takes over what is repetitive and time-sensitive: the first reply, basic qualification, proposing a viewing time and reminders.
• The broker stays responsible for negotiation, prices, terms, valuation and any exception.
• Property availability and the broker's calendar must come from the agency's own systems, never from the model's memory.
• Follow-up needs rules: permission, approved templates, an attempt limit and an immediate stop when the lead replies or asks not to be contacted.
• The safe path is a pilot on one real flow, with conversation logs, an agreed metric and a decision to expand, adjust or stop.

What should the AI do and what stays with the broker?

The most useful split is not by technology but by type of decision. Anything that is already-registered information or an organizing question can be automated. Anything that involves commercial judgment, legal risk or relationship should go to a person.

A simple decision guide for your team:

• The AI answers: confirmation that the contact was received, listing data that is in the property record (floor area, neighborhood, bedrooms, parking spaces, listed features), qualification questions and proposals of free time slots.
• The AI records and forwards: the lead's preferences, questions that are not in the property record, viewing requests, requests to speak with the broker.
• The broker decides: price negotiation, discounts, payment terms, financing simulation, credit analysis, documentation, contracts, comparison with other properties and anything about what the owner will accept.
• Never automate: requests for personal documents over chat, statements about the legal status of the property, guarantees of appreciation or of financing approval.

When a question falls outside this list, the correct AI response is to say it will pass the question to the broker and deliver a summary, not to improvise. This has to be written into the agent's instructions, as we explain in how to write instructions for an AI support agent.

How to answer portal leads quickly and with context

A portal lead arrives with little information and a lot of competition. Whoever is slow often loses the conversation to another agency that replied first. An automatic reply helps, but only if it is specific: it names the right property, stays short and opens a conversation instead of dumping generic text.

The basic flow has these steps:

• Capture the lead from its source: portals such as ZAP Imóveis, Viva Real and OLX, the website form, a social media ad. Depending on the portal and the plan, this arrives by email, CRM integration or webhook. Confirm with the vendor what is available before designing the flow.
• Identify the property of interest and pull its data from the internal record. If the listing code does not match, the AI does not guess: it asks or forwards.
• Check for duplicates. The same person may appear on two portals, or may already be a client of a broker. In that case, the conversation goes to whoever already handles them.
• Send the first message through the channel the lead authorized, usually WhatsApp, naming the property and asking whether they want more details or want to book a viewing.
• Create or update the CRM record with source, property, time and a summary of the conversation.

One important caution: property status changes. If it was sold, rented or taken off the market, the AI needs to know that from the agency's source of truth. If the system does not respond or the status is uncertain, the message should say the broker will confirm availability, not state that the property is free.

For the channel, read how to integrate AI into your company's WhatsApp, especially the part about the official API, so you do not depend on solutions that put your number at risk.

Which qualification questions are worth asking?

Qualifying is not interrogating. The goal is to give the broker enough that the first call or viewing is useful. If the lead already supplied something in the portal form, the AI does not ask again.

Questions that usually earn their place:

• Purpose: to live in, to invest or to rent.
• Area: whether the neighborhood in the listing is essential or nearby neighborhoods are acceptable.
• Property profile: number of bedrooms, parking spaces, whether a property that needs renovation is acceptable.
• Timeline: whether they plan to move soon or are just researching.
• General way of buying: whether they plan to finance or pay another way, only as context.
• Best time for contact or a viewing.

What the AI should not ask for: income, CPF (the Brazilian taxpayer ID), documents, statements or bank details. That information comes later, with the broker, in a suitable channel and at a suitable moment. Collecting only the minimum necessary is also a principle of LGPD (Brazil's data protection law), and we come back to it below.

Another point: the answers should become structured CRM fields (purpose, area, timeline) in addition to the text summary. That way the manager can filter leads and the broker can prioritize without reading the whole conversation. If your team still updates the CRM by hand, this article on updating the CRM with AI shows how to separate drafts from confirmed records.

How to book viewings without double-booking

Booking a viewing seems simple until a conflict appears: two leads at the same time, a broker at another viewing, the key with the owner, a building reception that requires the visitor's name. The rule is that the broker's calendar is the source of truth, and the AI only checks free slots and proposes options.

A practical step by step:

• Define which properties the AI may book viewings for: for example, only properties with a key available and a responsible broker assigned.
• Connect the broker's calendar (the one they already use) with permission limited to checking availability and creating the viewing event.
• Propose a few time options, not the whole calendar.
• Confirm name, property and time with the lead, and ask only for the data the building or condominium requires, when applicable.
• Create the event, notify the broker with the qualification summary and send the confirmation to the lead.
• Send a reminder before the viewing and offer rescheduling. If the lead asks to change, the AI proposes new times; if they ask for something outside the standard, it forwards to the broker.
• After the viewing, the broker records the outcome, and only then is post-viewing follow-up released.

The technical side of this kind of flow is detailed in automatic appointment scheduling on WhatsApp with AI. For a real estate agency, the point of attention is the handoff: if the lead asks to negotiate before viewing, the AI does not hold the conversation. It passes the contact to the broker with the context.

How to follow up without becoming spam

A good share of sales and rentals happens after the first contact, and that is where service often fails for lack of time. AI is good at reminding and resuming, as long as there are clear rules.

Minimum follow-up rules:

• Only send outbound messages to people who have permission to be contacted on that channel. On permission and message templates, see WhatsApp outbound messages: templates, opt-in and best practices.
• Define an attempt limit and an interval between attempts. The team decides the numbers, but the limit has to exist and live in the system, not only in someone's head.
• Stop when the lead replies, when they ask to stop receiving messages or when a broker takes over the conversation.
• Tie each message to a concrete reason: a viewing that did not happen, a pending question, a property similar to what they were looking for, a callback the broker promised.
• Always offer a simple way out, such as replying that they are no longer interested.

After a viewing, the message can ask what the lead thought and whether they want to talk to the broker about next steps. If the reply shows any sign of an offer, a price objection or a question about terms, the AI records it and calls the broker. It does not counter-argue or suggest a discount.

Outside business hours the same principle applies: answer the basics, record the request and promise only what the team will deliver. There is more on this in after-hours customer service with AI.

Illustrative scenario: a Saturday night lead

Illustrative scenario, not tied to any real client. A person asks for information about an apartment advertised on a portal, on a Saturday night. The AI agent checks the property record, confirms the property is listed as available and replies with the neighborhood, bedrooms and parking spaces as shown in the listing. It then asks whether the purpose is to live in or to invest, and whether they prefer to visit on a weekday or at the weekend.

The lead says they want to live there, need to move soon, and asks whether the owner would accept a lower price. The AI answers that this conversation is with the responsible broker, records the question and offers two viewing times from the broker's calendar. The lead picks one. The event is created, and on Monday the broker receives a summary with purpose, timeline, chosen time and the price question, highlighted as a point for the broker to handle.

Nothing was promised about a discount, and the broker arrives at the conversation knowing what matters. That is the design: speed and organization up front, commercial decisions behind.

How to start safely: LGPD, risks and pilot

A portal lead provides data to be contacted about a property. Using that data for other purposes, or keeping it longer than necessary, requires care. Define the purpose, collect the minimum, tell people that the service uses automation, offer a way to speak with a person and have a retention rule. The ANPD, Brazil's data protection authority, is the official reference for data protection guidance in the country. If the model provider processes data outside Brazil, see international data transfer under LGPD when using AI providers.

There is also a technical risk: the text the lead writes is untrusted input. Someone may try to instruct the agent to ignore its rules, reveal other leads' data or promise conditions. OWASP's guidance on risks in LLM applications helps in designing minimum permissions: the agent checks calendars and creates events, but it does not change prices, delete records or access data it does not need.

The method we apply in projects like this has four stages:

• Map: how the lead arrives today, who answers, how fast, where contact is lost, and what the starting point is before AI.
• Limit: which data the agent sees, which actions it can take, what always goes to the broker and when it must stop.
• Pilot: one real flow, for example one portal and one team, with conversation logs and a metric agreed before starting, such as time to first reply and viewings booked that actually took place.
• Hand over: document how it works, train the team to review the logs and decide from the data whether to expand, adjust or end it.

To structure risk and review, the NIST AI Risk Management Framework is a useful reference. And for designing an agent with few steps and well-defined tools, Anthropic's guide Building effective agents sets out clear principles. A note on authorship: the official sources describe general practices. How they fit into each agency's routine is a hypothesis to validate in the pilot.

Frequently asked questions

Can the AI tell the lead the price of the property? It can repeat the price shown in the listing and the property record, as long as the source is the agency's own system. It should not negotiate, hint at a discount margin, simulate financing as if it were an offer or comment on what the owner will accept. Any question in that direction goes to the broker.

Do I need to change my CRM to use AI for lead service? Usually not. The starting point is understanding how leads arrive and where they are recorded today, and then integrating the agent with those systems. If your current CRM does not allow integration, the pilot can start with a simpler record, but this should be assessed beforehand, case by case.

Does the lead need to know they are talking to an AI? The prudent recommendation is yes: state clearly that the first contact is automated and that there is an easy way to reach the broker. This reduces misunderstandings and fits the transparency LGPD expects. For decisions that affect the data subject, also see the article on LGPD automated decision-making and the right to review.

How do I know whether the pilot worked? Compare with the starting point you mapped beforehand: time to first reply, how many leads reached a viewing that was booked and held, and the brokers' own rating of the summaries' quality. Also review real conversations to find wrong answers or improper promises. If the numbers improved and errors are rare and fixable, expand gradually. If not, adjust or stop.

Sources

• ANPD: Autoridade Nacional de Proteção de Dados
• NIST AI Risk Management Framework
• OWASP Top 10 for LLM Applications
• Anthropic: Building effective agents