AI avatars for reception: how they work and what the AI Act requires

What an AI avatar is

An AI avatar is a digital interlocutor with a face and a voice, which listens to a question and answers by drawing on the information the company has decided to give it. It is not a recorded video and it is not an on-screen menu: the answer is produced on the spot, so it can handle questions nobody had anticipated.

Picture a lobby at eight in the evening, when reception is closed. Someone walks in, approaches a vertical screen and asks in English where tomorrow morning's meeting is being held. The avatar answers, shows the floor map and points them to the lift. That is its job: to welcome, orient, inform and answer, at the hours and in the languages a person cannot be there.

How it works, in brief

Behind an avatar that looks simple, four things work together. A microphone that listens. A system that understands the question and builds the answer using the company's information. A synthetic voice chosen at design stage. A face animated in real time. The best answers are not spoken only: a useful avatar also shows a map, a data sheet, a video or a route, because voice alone saturates quickly.

The detail that decides everything is response time. In our experience, when the wait exceeds a few seconds the conversation breaks and the person walks away, especially if they are standing in a transit space. That is why we design speed as much as content.

The value lies in what it knows

The most common mistake is to think the value lies in the face. The face is the part people see; what makes an avatar useful or embarrassing is what it knows and what it has been taught not to say. A well built knowledge base contains the verified information the company wants to convey, the limits of its scope, the behaviour to adopt when there is no answer (admitting it does not know and pointing to a person) and a simple way to stay current. An avatar still describing last season's exhibition a month after opening does not have an artificial intelligence problem, it has a maintenance problem.

Where it adds value

Corporate reception

In the lobby it welcomes visitors and suppliers, explains who does what, manages waiting time and introduces the company to those seeing it for the first time, at hours and in languages reception cannot cover. It integrates with the other languages of interactive communication already present in the space.

Museums and exhibition routes

It is the always available guide for the questions that arise in front of an object and that no panel anticipates. In a corporate museum, where the audience changes with every delegation, it becomes an international welcome tool.

Hotels, showrooms and trade shows

It handles the repetitive information that occupies the front desk, guides product choice when staff are busy, absorbs peaks of attendance at the stand and leaves the conversations that matter to the sales team. We cover this in the context of shops, hotels and showrooms.

The rule to know: the AI Act

There is a part that is almost always discovered late, and it changes the project. The European regulation on artificial intelligence, Regulation (EU) 2024/1689 known as the AI Act, contains obligations that apply directly to avatars talking with the public.

From 2 August 2026, anyone interacting with an avatar must be able to know they are talking to an artificial intelligence system, clearly and from the first interaction, except where it is obvious from the context. It is not a notice to be added afterwards: it is a design requirement of the interface. Since 2 February 2025, moreover, it is prohibited to use systems that infer people's emotions from biometric data in the workplace and at school, except for medical or safety reasons. If someone proposes an avatar that reads employees' mood, the question to ask is in what context it would be used. And data protection law applies as soon as someone's voice is processed, even if it is not stored: an avatar that keeps nothing is a simpler project, not a project outside the rules.

The practical consequence is reassuring: declaring that the avatar is an artificial intelligence is not a concession, it is the correct way to present it. In our experience people respond better when they know what they are talking to, because they calibrate expectations and stop looking for the trick.

How we build it to measure

An off-the-shelf avatar answers like a search engine. A designed avatar answers the way the best person in that company would, and the difference is built through five choices.

We start from real questions: we collect those that reach reception, front office and floor staff today, and define what the avatar must know, what it must decline and when it must hand over to a person. Then we choose the form, from a realistic face to a stylised figure, consistent with the identity of the place, and the voice in the languages that actually matter. We design the visual feedback, meaning what appears on screen while the avatar speaks. We check noise, position and light in the real space, because a crowded lobby and a museum hall are two different worlds. And we leave the company a simple way to update the content, with an owner and a procedure, so the avatar grows with the space instead of ageing.

The same knowledge base can feed several channels: the avatar in the lobby, the assistant on the website, the kiosk in the showroom. It is the way to have a single consistent voice wherever the public meets the company.

The limits, stated upfront

A language model without constraints tends to answer anyway, even when it does not know: this is why we govern it with rules and a closed knowledge base, not with hope. In noisy environments speech recognition degrades, and a written alternative is needed. A highly realistic but imperfect face can feel more unsettling than a deliberately stylised figure, so the aesthetic choice is made deliberately. A voice-only interface excludes those who cannot hear and those who do not speak the language: subtitles and on-screen text are not an extra. And the avatar does not replace a person where the value lies in the relationship: it takes repetition away from people, it does not replace them.

Where to start

If you are considering an avatar for a space, a useful starting point is understanding what questions your staff receive today, in which languages and at what times of day. Write to us describing the space, the type of audience and the most frequent requests: we will tell you what can be automated with a good margin of reliability, what is better left to people and which on-site checks are needed before defining the project.