Italy's top property site now uses a team of AI voice helpers

Immobiliare.it, Italy's largest property website, started with one AI voice assistant on its home listings. It now runs a connected network of AI agents that each handle a different job. More than 70% of sellers chose to turn on the first assistant. The story comes from the blog of ElevenLabs, the company that supplies the technology.

The first assistant

Immobiliare.it lists more than 1.2 million homes in Italy. It draws over 50 million visits a month to its website and app, and it serves more than 31,000 real estate agents and developers. The company put an AI voice agent on each listing. It answers day or night, in the right language, and it answers buyer questions. It also collects contact details and writes a short summary, so a human agent can see how serious a lead is.

Why voice worked

Buyers were more comfortable sharing personal details by voice than by text. That produced better leads, according to the company. The article does not give exact figures for how much better. Sound also mattered. The team wanted native-sounding voices with local accents, because in southern European markets a natural voice can decide whether a caller stays on the line.

From one agent to a network

The first agent became the base for more. Each new agent took on a different part of the marketplace. The article's list of those roles did not come through in the material, so they are not named here.

The company also uses subagents. These are smaller agents that each do one narrow task. They work behind the scenes, so the user still feels like they are in one simple conversation. Chief Product Officer Paolo Sabatinelli said much of the innovation lies in deciding how agents work together and keeping their behavior predictable.

Safety and oversight

Immobiliare.it also pursued AIUC-1, an emerging certification standard for the security, safety and reliability of conversational AI agents. The article says the built-in controls in ElevenAgents, the ElevenLabs product, covered the majority of the standard. It does not say whether the certification has been granted.

What it means for a business

The lesson is the order of steps. One assistant with a clear job came first. Its transcripts showed what happened in real calls, and the team improved from there. Sabatinelli said building an agent is simple but improving it continuously is harder. Note that the source is the vendor's own blog, so the results are self-reported.

• Start with one task, such as answering questions and capturing leads.
• Offer the assistant in the customer's own language.
• Read real conversations to see what to fix.
• Ask what safety controls come built in.