Three enquiries on the same flat, inside an hour.
A real estate AI chatbot that answers all three enquiries at once.
OpusDone answers every enquiry on a listing the moment it lands, using only the listing notes you approved, and puts a viewing in the diary. Three people ask about the same two-bedroom within an hour, and whoever replies first gets the viewing. It takes a phone number, scores the lead on published weights, and writes it to a Google Sheet in your own Drive within a minute. You read the whole thing back when you are out of the car.
- Every enquiry answered the same evening
- Viewings booked without a phone call
- One lead per person, not four
Reviewed
Why do property enquiries all arrive at once?
The good flats go in a day
A well-priced two-bed draws a dozen enquiries before lunch. They all ask the same three things, they all ask them while you are showing someone else round, and the one who hears back first is the one who views.
Most of them are not buyers yet
Neighbours checking prices, people six months from a mortgage, tenants who wanted the other listing. Sorting that by hand is an evening a week, and the real buyer is somewhere in the middle of it.
The same person, three times over
She writes about the flat on Girona, then about the one in Poblenou, then on Telegram with a different name. By Friday she is three entries in a spreadsheet and nobody has called her back.
How does OpusDone score a property lead?
Three enquiries on the same two-bedroom
They all arrive while you are showing something else.
- 18:42
Three at once
Three people open the widget about the same second-floor flat inside the hour.
- 18:44
The same honest answers
Each gets exactly what you approved about the flat, with nothing invented about price or square metres.
- 18:51
Two viewings written in
Both go into one calendar with deterministic event ids, so a retry cannot create a duplicate.
- 19:06
The third leaves a number
One reachable contact carries the score past fifty on its own, so the alert fires.
- 19:20
The one who vanished
A visitor with real intent and no way to reach them goes on the unreachable list.
From your listings to a viewing in the diary
Three enquiries land on the same flat inside an hour. This is how the system decides what to say and who to wake.
- 01
It reads the agency site for the basics
The crawl finds your services, fees, FAQ, contact and hours pages by English and Spanish keywords, then reads the meta description, Open Graph tags and JSON-LD for the office address, phone and opening hours. One model pass turns that into structured services, FAQs, policies and four to ten draft knowledge entries. It writes only into empty fields, so the wording your negotiators agreed on is safe from every later crawl. - 02
You mark live only what you are happy to be quoted on
Drafts, imports awaiting approval, archived and rejected entries never reach the assistant. That is the lever you will use most: when a flat goes under offer you archive the entry, and it is out of reach in about 15 seconds. A quality gate discards menu text and cookie boilerplate, anything under 80 or over 4,000 characters and near-duplicates at 85% overlap. - 03
You set viewing hours, Saturdays and notice
Working hours, and closed days set by clicking one day or dragging a range across the month calendar. Minimum notice defaults to 120 minutes, so nobody books a viewing you cannot physically reach; slot length is 30 minutes, granularity 15, buffers zero and horizon 60 days, all editable. Closures are mirrored into Google as an all-day event carrying your agency’s name, and deleting that event does not reopen the day. - 04
You connect one calendar for the office
Google Calendar over OAuth with no API key, or Cal.com with one. It reads every calendar you mark as busy and merges them, so a negotiator’s own commitments remove the slot rather than colliding with it. It books the agency, not a named negotiator, and its scopes are narrow: read and create events, never delete a calendar. - 05
You test it as three different buyers
The sandbox runs your real instructions and your real tools and stops before writing anything — no viewing booked, no lead scored, no alert sent. Try the serious buyer, the time-waster and the one who wants a price the vendor has not agreed to, and see where each conversation ends. It is free and never gated. - 06
One script tag, and every enquiry lands in one place
The widget is one line on the site your portal traffic already arrives at, and Telegram is the second live channel; both enter the same function, so the door does not change the answer. Hours and closed days are checked before a slot is offered and again before it is written, availability is re-read at the moment of writing so two buyers cannot take the same viewing, and it offers at most six options with no more than two from one day. It replies in the language the enquirer wrote in, with the Article 50 AI disclosure added once per conversation.
What if a flat went under offer this morning?
Then you archive or edit that entry and it stops being mentioned within a minute of you approving the change. The assistant only ever speaks from knowledge marked live — drafts, imports awaiting approval and archived entries are invisible to it — and it will never invent availability, a price or a completion timeline to fill a gap.
What it does with a burst of interest.
The questions it takes off your desk
Five it will answer today, from knowledge you approved: “Is the two-bed on Carrer Girona still on the market?” · “What are the community fees each month?” · “Does the price include the parking space?” · “Can I view on Saturday morning?” · “Do you work with buyers who are not resident in Spain?”.
One person, one lead, scored by plain arithmetic
Opened chat +10, three or more messages +10, name +10, email +40, phone +40, booked a viewing +40. Alerts fire at 50 because one reachable contact is enough. The same person across the widget and Telegram stays one lead, so you stop calling the same buyer twice and never calling the other one.
Hot 75, warm 50, cool 30. Filter, sort and export to CSV.
A weekly digest that shows what the evenings were worth
A weekly email and a monthly report show after-hours bookings, hours saved, the share your assistant handled on its own and the month-on-month change. It is the number you use when deciding whether the portal spend is the thing that needs fixing.
What will OpusDone never do on your listings?
An agent who overpromises loses the instruction. So does a chatbot.
It will not invent availability or a price
If a listing is not in your live knowledge, it says it does not have that and gets you. It cannot offer to check and come back later, because it cannot check anything.
It will not advise on mortgages or tax
No financial guidance of any kind. Questions about lending, taxes or what someone can afford go to a person, with the details already captured.
It will not answer for you on the portals
The live channels are the website widget — one script tag, no iframe — and Telegram. Portal inboxes stay yours. What it covers is your own site and Telegram, which is where the serious ones end up anyway.
It will not open the floor plan you send it
No document or image understanding. An attachment it cannot read skips the model entirely: it says so and fetches a person instead of guessing at square metres.
Real estate chatbot questions, answered
Can it book viewings with a specific agent?
What if the buyer writes in French or German?
A buyer asks to speak to me directly at nine in the evening.
Three enquiries on one flat in an hour. Which one do I ring first?
How do listing details get in without a week of typing?
Somebody clearly serious refuses to leave a number. Is that lost?
My listings change daily. Won’t it keep offering a flat that went under offer?
More for property teams
Every enquiry answered while it is still warm
Website widget and Telegram, one lead per person across both. Filter, search, sort and export to CSV, and every lead is also written to a Google Sheet in your Drive or posted to any https endpoint within a minute.
Request access