Reading and classifying enquiries
A language model understands an email with travel dates and wishes immediately. 40 enquiries a week become 40 prepared offers. The gain is in response time.
Knowledge
AI in the hotel:
what it can do and where it has limits.
Many hospitality chains already use AI. Often, though, a shared direction, clear responsibilities and clean connections are missing. That is where isolated tools, unnecessary risks and frustration in the team arise.1
1 The global h2c study shows: 78 percent of the hospitality chains surveyed use AI, but only 7 percent have a company-wide AI strategy. 189 people from 171 hospitality chains worldwide were surveyed. Source: h2c, AI & Automation in Hospitality, 2025. Read the study
Where AI in the hotel
really helps today.
In six places, very clearly. What they have in common: plenty of repetition, clear rules, a result you can check.
Texts in the guest's language
Confirmations, reminders and replies in six languages, in the tone of the property. Once a cost item, now a setting.
Finding patterns in numbers
Occupancy curves, cancellation rates, source markets. AI finds anomalies in data that is already there, before they show up in the monthly report.
Triggering recurring workflows
Room status, task allocation, official reporting. Where someone works through a list today, the system can make a start and let people decide.
Making knowledge in the property available
House rules, packages, agreements with regular guests. New employees find answers without interrupting someone who is with a guest.
Checking instead of searching
An arrival check that reports before the start of the shift what is missing: registration forms, room allocation, bed details for families. A small thing, every day.
Where AI in the hotel
still fails today.
You rarely hear these six points from vendors. Yet they decide whether a project holds up.
Bad data gets bad faster
AI does not turn untidy data into clean data. It only spreads it faster. Inconsistent room categories and duplicate guest profiles become serial errors instead of being found.
The last step stays manual
A system can write an offer. Whether this regular guest gets a discount this year is still decided by a person. Anyone who forces automation where discretion is needed loses guests.
Confidently wrong
Language models answer fluently even when the basis is thin. Without a visible data snapshot and confidence score, nobody notices in daily operations. Ask every vendor how they solve this.
Costs that grow with you
Every model request costs money. Without a quota, usage display and hard cap, a software line item becomes an open-ended invoice. The most common angry letter after the first quarter.
Acceptance in the team
Anyone who feels replaced will not use the tool. Anyone who sees that the annoying twenty minutes in the morning disappear will. That is an onboarding topic, not a technical one.
Dependence on the vendor
If automation rules exist only in the vendor's system and cannot be exported, switching becomes expensive. Ask about export formats and deadlines before signing the contract.
What AI-native
really means.
The term appears on almost every vendor page. Yet it has a meaning you can check.
AI-native means: AI is the way the software is operated. It is not an add-on next to the actual product. Data model, permissions, log and interface are built so that a system prepares and the human decides.
In the trade, the counterpart is called bolted on: existing software to which a chat window was added afterwards. There, the chat talks to the user, but not to the system.
The simplest check
Imagine all AI functions removed. What is left?
If the product stays as it was before, the AI was bolted on. If operation shifts everywhere to where menus and clicks used to be, it is AI-native. This check costs nothing and shows more than any sales presentation.
Seven signs
that AI is only bolted on.
Take this list to your next vendor meeting. You do not need IT knowledge for it.
- The AI lives in a drawer. There is a button, a window opens, the assistant sits inside. Everything else you operate as before.
- The assistant knows only one area. It can talk about reservations, but not about housekeeping. It sits on an integration instead of the shared dataset.
- It talks, but it does nothing. It answers questions and tells you where to click. It changes nothing.
- No data snapshot, no confidence score. You see the answer, but not what it is based on and how certain it is. AI-native software shows both, because it needs them for its own checks anyway.
- Approvals cannot be configured. A step either runs automatically or not at all. The system does not know the state in between.
- AI usage does not appear on the invoice. No quota, no usage display, no cap. Usually nobody is responsible for the model costs.
- The price depends solely on the number of users. A price per workstation assumes that only people operate the software. For a system that works independently, that no longer applies.
What the law
requires of you.
The EU AI Act distributes obligations according to role. For businesses, what counts are a labelled guest chat, trained employees and traceable use.
Labelling obligation
If a guest communicates with a chatbot, a digital concierge or an AI-supported reply, they must be able to recognise this. The transparency obligations have applied in full since 2 August 2026. Breaches are subject to sanctions.
AI literacy in the team
Businesses that use AI must ensure sufficient AI literacy. A short, documented briefing is therefore part of onboarding.
You are responsible
For the processing of guest data you are the controller, not the vendor. You need a data processing agreement and, in the case of high risk, a data protection impact assessment.
2 European Commission, EU AI Act: transparency obligations under Article 50 have applied since 2 August 2026. The obligation to ensure sufficient AI literacy under Article 4 has applied since 2 February 2025. Which obligations apply to your business depends on the system and role. European Commission guidance
This page is not a substitute for legal advice. It names the obligations that are most often overlooked in practice.
Where felixUnity
stands on this list.
The same seven questions, applied to us. You are welcome to check each of them in the personal appointment.
| The question | Our answer |
|---|---|
| Does the AI live in a drawer? | felix is the operation, not a window beside it. You write or speak in normal sentences, and it happens in the system. The classic views remain alongside. |
| Does it know only one area? | Reservation, offer, housekeeping, official reporting and analytics sit on one dataset. Plus over 400 connections from 46 years of felix. |
| Does it only talk? | It creates offers, assigns tasks, generates lists and notifications. What it may execute is up to you. |
| Does it show data snapshot and confidence? | Yes. Every check shows the time, data snapshot and areas included; enquiries carry a confidence score. |
| Can approvals be configured? | Yes, per workflow and per role. You can move the boundary in both directions at any time. |
| Is AI usage visible? | Monthly quota, usage visible before triggering, hard cap as default setting, warning at 70, 90 and 100 percent. |
| And the guest data? | Storage and processing exclusively on European servers. Model training takes place without your guest data. |
Where we do not yet convince
felixUnity is starting together with selected businesses. We do not yet have reliable impact metrics from live operation. We will publish them as soon as they exist, not before.
All test questions in the comparisonFrequently asked questions
about AI in the hotel.
Does AI replace staff at reception?
In practice it shifts the work rather than replacing it. What disappears is the retyping, looking up and forwarding. What remains is the part guests come back for.
Do I need clean data before I start?
Not in the whole property, but in the starting area. That is why a project sensibly begins with a single workflow. There, the data can be put in order in a manageable time.
Do I have to tell guests that an AI is replying?
Yes. Since 2 August 2026, the transparency obligations of the EU AI Act have applied in full, also to chatbots and AI-supported communication. The notice must be recognisable to the guest.
Is our guest data used for training?
Not at felixUnity. Storage and processing take place exclusively on European servers, model training without your guest data. Ask every vendor to give you this in writing.
What does AI usage cost per month?
What matters is the structure: a monthly quota, visible usage and a hard cap as the default setting. Without these three things, planning is impossible.
Test us
on your own workflow.
Bring a real enquiry from your property. We will go through it together, and you ask the seven questions from this page.