Enquiries don't arrive during office hours. Applicants browse the portals in the evening and at weekends, and the agency that replies first tends to get the viewing. That's the case for an AI assistant on your website, your portal enquiries and your out-of-hours phone line.
The case against is what happens when it's set up badly. An assistant that's allowed to make things up will tell an applicant the flat has parking, offer a viewing on a day the negotiator is off, and quote a rent from a listing that changed last week. Each one is a complaint waiting to happen, and in a regulated business, a wrong statement about a property isn't a small thing.
What should an agency decide before an AI assistant goes live?
Four things: what the assistant is allowed to say and where it gets the facts from, how it tells people it's not a person, when it hands over to someone who is, and what it never does. Decide those, and it's a useful colleague. Skip them, and it's a liability with a friendly tone.
1. Where does it get its facts?
From your systems, and nowhere else. The assistant should answer questions about a property from the listing as it stands today, from the CRM or from a document you've given it. If it doesn't have the answer, it says so and takes a message. It never fills a gap from general knowledge, because general knowledge doesn't know whether the boiler was replaced.
This is the single decision that matters most. An assistant built to answer only from your data can't invent parking. One built to be helpful will.
2. How does it say what it is?
In its first message, every time. "I'm the agency's automated assistant. I can answer questions about this property and book a viewing. If you'd rather speak to a person, say so and I'll pass you on." People don't mind talking to software when they know that's what they're doing. They mind finding out later.
3. When does it hand over?
Before anything that commits the agency. It can answer questions about the property, take details, and offer viewing slots from the diary. It shouldn't negotiate on price, confirm a tenancy, give an opinion on whether an offer will be accepted, or answer anything about a tenant's rights that it isn't reading from a document you've approved. At each of those points it takes a message and tells the person when they'll hear back.
A good test: if a negotiator would want to see the message before replying, the assistant shouldn't reply to it.
4. What does it never do?
Promise. "You'll definitely get a viewing on Saturday." "The landlord will accept pets." "Your application will be approved." The assistant deals in what the listing says and what the diary shows. Anything else is a person's call.
What about the people who don't want to use it?
Keep the phone number and the email address where they were. Some applicants and some landlords will always want a person, and a line that says "or call us on" costs nothing.
What does this look like when it's built properly?
I built something along these lines for Ladies Who Lit, a book club with around 3,000 members. Their bookings and retreat payments were tracked by hand in a spreadsheet, and the community lived in WhatsApp. The member portal I built now sends welcome emails, payment reminders and event updates on its own, and AI watches the flow and pushes the points that need a person to the team rather than deciding for them. The principle is the same as an agency assistant: the software handles what's routine and known, and hands over the moment something needs judgement.
For an agency, the assistant sits on top of the listing data and the diary, answers what it can from those, and sends everything else to a named negotiator with the conversation attached. The negotiator sees the whole exchange and picks it up in the morning. Nobody has been misled, and the applicant has had an answer at nine on a Sunday.
What should an agency do this month?
Write the four decisions down before you buy or build anything. Then test the assistant with the ten questions applicants ask most, and the five they ask that it shouldn't answer.
- Decide the facts it may use: the live listing, the diary, and documents you've approved.
- Write its first message, including how to reach a person.
- List the handover points, and name who picks them up.
- Test it with real enquiries from the last month before it goes live.
I build this for agencies as part of the tools for your clients work, and the team AI rollout sets up the policy and the register around it. The AI governance kit's General edition has the policy and register ready to fill in if you'd rather do that part yourself.
Sources
- 01ICO, guidance on AI and data protection. ico.org.uk
This article is general information, not legal advice. Check the current guidance from your redress scheme and the ICO, and speak to your compliance lead about your agency.
Readiness & governance audit
The audit writes all of this down for your firm.
£395, delivered within 10 days, and credited in full against any pack or subscription booked within 60 days.
Guy Habgood helps independent professional firms use AI safely, take admin off their teams, and build software when nothing off the shelf fits. He's based in Alresford, Hampshire, and works with firms across the UK.
About Guy