AI for student residences: it answers the building's WhatsApp and knows when to call a person

What AI does in a residence, and what it does not do

The rule fits on one line: the AI answers information and hands on judgement. Whatever is written in the centre's manual it can answer on its own, at any hour and in the language of whoever asks. Whatever calls for a decision about a person, about money or about life in the building it takes in, puts in order and places in front of whoever has to decide. The table brings it down to the messages of an ordinary day:

What it can answer alone and what it must pass to a person, category by category and with what a good handover looks like, is in the guide what a residence's AI should answer and what it must hand on.

It answers from your centre's manual, not from the internet

An AI assistant understands what it is told even without keywords, but it should only answer with what the centre has given it. In a residence that is the manual: the house rules, dining and reception hours, how the laundry works, what happens with visitors, where parcels are collected. We give it to the AI separately and with priority, with an instruction not to invent a rule or a fact.

When the answer is not in the manual, the right thing is to say it does not know and pass it to the team. An AI that fills the gap with something plausible is worse than none, because residents trust it: that is called a hallucination, and in a centre it is the costliest failure because it comes to light late.

• It answers in the language of the last message: if someone writes in French, in French; if they switch to Spanish, in Spanish.
• The team sees every conversation in the dashboard and can correct a reply, pause the AI or take over the conversation whenever they want.
• If a reply went wrong, the text of the manual is corrected and the next one comes out right. The AI does not learn on its own from what residents tell it.

Repairs: from the resident's photo to the phone of whoever is on shift

Half of an AI's work in a residence is not answering but sorting. A message with a photo of a dripping tap becomes a repair ticket with its category, its urgency, its room and its deadline, and it goes to one person's phone, not to a group.

• The ticket goes to the post that handles that category, according to the org chart and shifts the centre sets up. If it is urgent, to the whole post at once.
• Whoever receives it accepts it with a button and says when they will come. The resident hears when it is accepted, when it is resolved and if it is reopened.
• Whoever owns it gets a reminder at 50 % and at 85 % of the deadline. An urgent ticket that runs out rings again once a day, and after three days management is told.
• Every ticket goes into a spreadsheet with the columns of your own log, and each night management receives the day's summary by email.
• More detail in faults and incidents.

Dining, packed lunches and notices from the team

In a residence with a dining hall, a good share of the messages are about food: not in for dinner tonight, a packed lunch tomorrow, back on Sunday. The AI logs every cancellation and every packed lunch where the kitchen sees it, so the day's count does not depend on who was at reception when the message arrived. The detail is in dining.

In the other direction, the team sends announcements to residents from the same number: a water cut, the exam timetable, an event. They go out exactly as the team writes them. The AI neither rewrites nor translates them, and we say so, so that nobody counts on it.

What always goes to a person

Some messages a machine should not answer even when it knows what to say. For those, the AI acknowledges the message so the resident knows someone will read it, and puts it on the right phone according to the escalation to people the centre has set up:

• Health: a resident who feels unwell, an allergy, a fall. No diagnosis and no advice.
• Building emergencies: fire, water, someone who cannot get out. They go to the whole on-call post, at any hour.
• Conduct: conflicts between residents, noise, complaints about another person. The AI neither mediates nor sanctions.
• Harassment, or anything not to be said on the shared WhatsApp: that is what the confidential channel is for, encrypted and kept apart from the chat, designed for what Spain's Act 3/2022 on university coexistence asks. When it applies and to whom, in the guide to the whistleblowing channel.
• Money, exceptions and complaints: a fee, a refund, a rule someone asks to skip, a formal complaint. Whoever has to decide, decides.

The AI Act and the GDPR: what the rules ask

From 2 August 2026, Article 50(1) of Regulation (EU) 2024/1689 requires that anyone talking to an AI knows it, unless that is obvious. That is AI transparency, and the duty falls first on whoever builds the assistant: Bedel says so from the first message. An assistant that serves residents is not among the high-risk uses in Annex III as long as it does not decide admissions or score anyone. Who answers for what, with the articles, in the guide to the AI Act and a residence's WhatsApp assistant.

On data protection, the residence is the controller and Bedel is its processor, with the processing agreement published. The database is in the European Union, in Ireland. The AI model runs in the United States under standard contractual clauses, and its provider does not train on the centres' data. What the residence has to have in order on its side, from lawful bases to health data, is in the guide to GDPR in student accommodation.

Isn't the WhatsApp group enough?

To tell the whole building about a water cut, the group is enough. For everything else it falls short: it assigns nothing, has no deadlines, shows every resident's number to the rest, and at night whoever is awake answers. The six usual ways of running the channel, compared on the same questions, are in WhatsApp group or assistant.

How it starts

• A session with whoever runs the residence and whoever runs maintenance: who handles what, the fault categories with their deadline, what counts as urgent at two in the morning and who is on call.
• The number: the one residents already have, or a new one, connected to Meta's official platform.
• The manual the AI may use. Whatever is not written down goes to a person, and that shows what is still to be written.
• In the first weeks, the team reads the conversations in the dashboard and corrects the manual wherever it needs it.

Where Bedel fits, and what it does not do

Bedel is that AI layer on the centre's WhatsApp, for student residences, university residences and colegios mayores in Spain. What the channel looks like day to day is on the page about the WhatsApp chatbot for student accommodation. For operators outside Spain, the English-language overview is AI for PBSA and student accommodation.

What it does not do, said just as plainly: it is not a property management system, so it does not allocate rooms, handle contracts or bill. It has no group view: one account is one centre. It does not control access. And it decides none of what the table above leaves in the right-hand column.