Saturday, 4 p.m. A couple spots the two-bedroom apartment they've been looking for on a listing portal and writes to the agency. Nobody answers until Monday. By Monday morning they've already booked a viewing with another agency, for another apartment. Ever happened to you? Then this article is for you.
Siena, 1338: a city paints its good government
Between 1338 and 1339, in the Sala dei Nove of Siena's Palazzo Pubblico, the painter Ambrogio Lorenzetti frescoed a cycle that still leaves visitors speechless: the Allegory and Effects of Good and Bad Government. On one wall is the well-governed city: builders at work on the rooftops, open shops, merchants, people dancing in the street, tended fields. On the opposite wall, Bad Government: ruined houses, empty streets, abandoned fields. The Nine, who governed the city, met right there, beneath those images.
You may say: Professor, lovely story, but I sell houses, I don't paint palaces. Exactly. That fresco says something simple that also applies to an agency: the value of houses depends on how well everything around them is governed. In your case, around every property there's a system: inquiries, property sheets, viewings, promises made to clients. If it's well governed, homes sell. If it's badly governed, even the loveliest apartment stays on the portal.
And there's a second lesson. Lorenzetti painted both walls, the good and the bad: an honest picture, not a postcard. In a real estate agency, artificial intelligence is useful when it helps govern the work well. And it becomes dangerous when it paints homes more beautiful than the real ones.
Where agencies lose time (and clients)
Before talking about tools, let's look at the work. In an agency with 4–8 agents, the typical problems are nearly always the same:
- Inquiries come through too many channels: portals, website, phone, social media, word of mouth. They end up in a shared inbox and nobody knows who replied to whom.
- The first reply comes late, especially in the evening and on weekends, when people are house hunting.
- Wasted viewings: agents show properties to people who haven't checked their mortgage options or are looking for something else entirely.
- Listings are written one by one, often in the evening, in Italian and perhaps English.
- Saved searches gather dust: a property comes in that's perfect for a client from three months ago, and nobody remembers.
What AI can actually do
1. Reply right away, every time
A system connected to the portals and the website can send, within minutes, a reply with property details, viewing availability and two or three useful questions: timing, budget, need for a mortgage. People must know they're receiving an automated reply: fairness requires it, and for chatbots so does the AI Act. Then the inquiry goes to a flesh-and-blood agent.
2. A single register of contacts
This is good government for an agency: everything in one CRM, each inquiry assigned to an agent, automatic reminders if someone is left without a reply. When a new property comes in, the system flags clients with matching requests. We described a project like this in the typical scenario of a real estate agency.
3. Better listings in half the time
Starting from the property sheet (square meters, floor, energy rating, features), AI can draft the listing in Italian and English. The agent rereads it and adds what only someone who has seen the home can tell: the morning light, the quiet courtyard, the café downstairs.
A non-negotiable rule: the data comes from the property sheet, never from the AI. Floor area, energy rating and features must be the verified ones. A listing with invented details isn't just a mistake: it can become misleading advertising.
4. Photos and virtual staging, but disclosed
Virtual staging helps buyers picture the space. But an edited image must always be flagged as such: fairness to the client demands it, advertising rules require it, and the AI Act sets transparency obligations for generated or manipulated content that could look authentic. A buyer who arrives at a viewing and finds a different home from the photos never comes back.
5. Paperwork in order
Document checklists for a sale, deadline reminders, viewing summaries for the owner: small automations that make sellers feel looked after and free up hours for the agent.
The rules to follow
- Contact privacy. The details of people who write to the agency are personal data: you need a clear privacy notice and specific consent for promotional messages. The GDPR applies.
- No automated promises. Automation doesn't negotiate prices, guarantee availability or make commitments on the agency's behalf.
- No discrimination. Qualifying criteria concern timing, budget and the features people want. Never origin, nationality or other personal characteristics.
What it takes to get started
For an agency with six agents, a project combining CRM, automated replies and listing drafts usually takes 6–8 weeks, including moving existing contacts and training. You can also start with a single piece, usually instant replies to inquiries, and add the rest later. You'll find more ideas on our page for real estate agencies and on our solutions page. To work out which tasks are worth automating, read We don't automate everything. We automate what makes sense.
What to do on Monday morning
- Measure your first-response time. Take the last 20 inquiries from the portals and check how long each waited for a reply. The number will surprise you.
- Count the channels. List where requests come from and where they end up. If the answer is “in different places”, your first project is a single register.
- Write the three questions. With your agents, agree on the three questions that separate a ready buyer from a browser. They'll be the basis of every automated reply.
Want to find out how many inquiries your agency is losing, and how to win them back? Book a free intro call: we start from the way you work, not from software.
The Professor at Penzai
