An AI receptionist for property enquiries: qualify interest without inventing appointments
An AI receptionist can help an agency collect missing information, prepare replies and propose a viewing. It should not manufacture property availability or treat a proposed slot as a confirmed appointment. The buying decision is whether one controlled workflow improves service at a defensible cost—not whether a bot can hold an impressive demonstration conversation.
Choose a small operational boundary
Begin with one office, one inbound channel and one type of enquiry. Name the person responsible for listing facts and the person who approves replies. Map what happens outside office hours and when a customer asks to stop. Do not start with a promise to replace the whole reception desk: urgent complaints, conflicting inventory and complex commercial questions need a reachable human owner.
The initial dataset should contain only material the agency is authorised to supply. A redacted historical sample can test workflow logic without opening a live inbox. Include unavailable listings, duplicated enquiries, incomplete budgets, conflicting dates and requests in another language. A convenient collection of straightforward examples will overstate readiness; keep a separate held-out evaluation set.
Build the funnel before counting success
| Stage | Evidence required | What does not count |
|---|---|---|
| Decision | Keep every stage distinct | No appointment or revenue guarantee |
| Meaningful response | Relevant reply delivered through the authorised channel | A generated draft |
| Qualified enquiry | Agency-approved criteria and missing facts checked | An AI confidence score |
| Confirmed viewing | Customer agreement and actual calendar reservation | A suggested time |
| Attended viewing | Attendance confirmed by the agency | A calendar entry |
Record the baseline before introducing the assistant: received enquiries, response time, staff handling minutes and confirmed attendance. Cost per attended viewing divides attributable operating expense by verified attended viewings. If the denominator is zero, the measure is undefined, not zero. Compare equivalent channels and periods and account for seasonality before attributing a change to AI.
The controls belong in the scope
- Listing facts need a current source, an update owner and an expiry rule; missing facts trigger a question or human review.
- Calendar writes require an approved slot and confirmation; repeated events need idempotency rather than a second booking.
- Opt-outs, complaints and sensitive financial or legal questions follow explicit escalation rules, not an improvised answer.
- The assistant must not select or reject people using protected characteristics or infer eligibility from sensitive information.
- Access, retention, deletion and incident procedures must be agreed for the actual data path before live use.
Acceptance and ongoing support
Start in shadow mode: prepare replies without sending or reserving anything. Review factual correctness, qualification completeness, escalation and correction time. Agree what constitutes a blocking error before testing. A successful shadow test authorises a separate decision about live access, not automatic activation. LEMAIT's public demonstration is synthetic and does not prove that a deployed agency already achieves these results.
A production support scope should name the inventory owner, monitoring coverage, failed-message queue, calendar reconciliation and change procedure when listings or policies change. Out-of-hours availability and response commitments must be purchased explicitly; a general maintenance statement is not a 24/7 service-level agreement. Tool and messaging charges belong in total operating cost.
A useful first brief includes monthly enquiry volume, current systems, the single office/channel to test, redacted examples and a reviewer. LEMAIT offers a bounded pilot from EUR 960 after the scope and inputs are agreed. If a structured web form or deterministic reminder is sufficient, that is a valid recommendation. This is an implementation framework, not Italian legal advice or a promise of appointments.
Prepared with AI assistance and checked against the linked public sources by LEMAIT. Scenarios are illustrations, not customer results. No ranking, ROI or legal-compliance guarantee.