Real estate WhatsApp chatbot: an evaluation checklist

A real estate WhatsApp chatbot is a conversational interface connected to a set of workflow rules and information sources. It may acknowledge inbound enquiries, recognise a property reference, answer approved questions, collect a few preferences, and route the conversation. The word ‘chatbot’ does not tell you whether those actions are accurate, permitted, recoverable, or useful to an agent.

Define the job as a user story before requesting a demo. For example: ‘When a buyer asks about a listed apartment after hours, confirm the exact listing, share the current approved brochure, collect their preferred viewing time, and hand the complete context to the duty agent.’ That story is testable. ‘Engage and convert leads automatically’ is not.

The chatbot should know which material it can use and which fields require human confirmation. Approved sources might include current project brochures, listing data supplied by the team, office hours, viewing procedures, and a defined set of buyer questions. Each source needs an owner, a review date, and a way to remove it when it becomes stale.

Ask vendors to demonstrate what happens when two documents disagree or the answer is absent. The correct behaviour is not a plausible guess. It is a transparent statement that the detail needs checking, followed by a human task. Test price, availability, incentives, handover dates, fees, financing, and legal questions because these are precisely where a fluent answer can create avoidable risk.

Property conversations carry specific context: listing, project, unit type, area, price range, purpose, timeline, and sometimes several properties in one thread. The chatbot should retain that context as the conversation changes. If a buyer moves from the original listing to alternatives, the original interest should remain visible rather than being overwritten.

The opening should be easy to answer on a phone. Confirm what the buyer asked about, provide a relevant option, and ask one focused question. Do not send a menu with ten buttons when a plain-language response would work. Do not ask for details that already came from the source form or portal message.

A chatbot should not score a person’s worth. It can collect observable information that changes what the team should do: buying, renting, or investing; preferred location; property type; timing; approximate range; and whether a viewing is requested. Missing information should remain unknown, not become a negative signal.

The bot should explain why it asks a sensitive or effortful question. ‘A range will help me avoid sending options that do not fit’ is more respectful than demanding an exact budget without context. When the buyer has already asked for a viewing, the workflow should move toward scheduling or agent confirmation instead of continuing an unnecessary questionnaire.

Unavailable listings are a normal part of portal lead operations and a revealing chatbot test. The system should not keep a stale listing alive to hold the conversation. It should state what the team has verified, explain when availability still needs checking, and offer alternatives only when there is a real basis for comparison.

An alternative should identify the trade-off: same building but a different layout, similar budget in a nearby community, comparable handover timing, or another property type. The buyer should be able to decline alternatives and wait for an agent. Honest limitation is a stronger experience than an endless stream of superficially similar listings.

Human handoff is not an exception hidden behind ‘contact support’. It is part of the core design. Trigger it for viewing requests, negotiation, uncertain or conflicting facts, advice, complaints, repeated misunderstanding, requests for a named agent, and any direct agent reply. WhatsApp’s business policy also expects clear escalation paths for automated experiences.

The handoff package should contain the source, listing, buyer’s exact messages, answers collected, material sent, detected language, current owner, and unresolved question. Tell the buyer that a person is taking over and avoid making them change channel unless necessary. Once the agent replies, automation should step aside until the team deliberately returns control.

The buyer should know which business is messaging and why. The workflow needs a record of how the conversation began, what kind of follow-up the buyer expects, and any request to pause or stop. Review the current WhatsApp Business Messaging Policy and applicable law rather than relying on a vendor’s generic compliance badge.

Test simple phrases such as ‘stop’, ‘not interested’, ‘message me next month’, and ‘speak to an agent’. The system should distinguish an opt-out from a timing preference and make both visible. It should never convert silence into permission for a new campaign.

Ask people outside the implementation team to complete realistic tasks on a phone. Can they request a brochure, correct a misunderstood area, ask about two listings, reach a person, and stop messages? Count unnecessary turns. Review whether the bot sounds clear and accountable rather than judging it by how human it appears.

Include messy inputs: voice notes, shorthand, mixed languages, partial property names, forwarded listings, and changes of mind. You do not need every format automated on day one, but the workflow should fail visibly and safely. A buyer should never be trapped because the system could not parse their message.

Agents should test the work queue, not only the chat. Can they see new, waiting, assigned, and escalated conversations? Can they understand why a lead was prioritised? Can they correct a property match, take control, and leave a next action without using a separate spreadsheet? If the chatbot saves buyer time but creates hidden administrative work, it has not solved the operating problem.

Measure first useful response, correct property context, owner coverage, viewing requests handled, handoff completeness, opt-out compliance, and repeated unanswered questions. Review a sample of conversations every week. Do not accept a vendor’s conversion claim as your forecast; establish a baseline and measure your own controlled rollout.