Chatbot or AI agent: what a hotel actually needs
A chatbot answers questions. An AI agent also does the work around the answer: it checks your rules, prepares a booking request, drafts a reply to a review or sends the night report to the manager. For a hotel the difference is practical. A bot that only says “please call reception” does not save anyone time. An agent that collects dates, number of guests and room type, and hands a complete request to reservations, does.
You do not have to choose one forever. Most hotels start with answers to frequent questions and add actions one by one, as trust in the answers grows.
Five hotel tasks that suit an AI agent
1) Pre-booking questions in messengers and on the site: check-in time, transfer, breakfast, pets, parking. The agent answers from your approved knowledge base and passes anything unusual to a person.
2) Booking requests: the agent collects dates, guests and wishes and hands reservations a complete request instead of a chain of ten messages.
3) In-stay concierge questions: how to order a taxi, when the spa is open, where to get a late dinner.
4) Review replies: the agent drafts a reply in your tone, and a manager approves it before it is published.
5) Daily summary for the manager: arrivals, departures, open guest requests and complaints in one message.
Where a person must decide
Leave to people everything where a mistake costs money or trust: confirming a price that differs from the published rate, refunds and compensation, overbooking, complaints about safety or health, requests from guests with special needs, and any conversation where the guest is upset. The agent’s job in these cases is to recognise the situation quickly, collect the facts and pass them to the right person with the full history.
Write these boundaries down before launch. A one-page list of “the agent never does this alone” protects both the guest and the team.
What to prepare before launch
An agent is only as good as the information you give it. Prepare four things: 1) a knowledge base with real answers — room types, rates and what they include, check-in rules, services, addresses and opening hours; 2) the tone of voice with three or four example replies; 3) the handover rule — who receives a request from the agent, in which channel and within what time; 4) a list of forbidden topics and promises.
If the answers live only in the heads of the reception team, start by writing them down. That document is useful even without any AI.
Guest data and the booking system
On day one the agent does not need access to your PMS or booking engine. A safe first step is an agent that answers questions and prepares requests, while a person enters the booking. Connect the booking system later, once you have seen several weeks of correct answers, and start with read-only access to availability.
Give the agent only the data the task needs. Passport details, card numbers and health information should not go into the conversation with the model at all. Tell guests they are talking to an AI assistant and always offer a way to reach a person.
How to measure the result
Choose two or three numbers before the pilot and record their current values: the time to the first reply in messengers, the share of requests answered without a person, the number of booking requests handed to reservations and how many of them became bookings. Compare the same weeks before and after.
An illustrative example, not real data: if reception spends two hours a day on repeated questions and the agent takes over half of them, that is about thirty hours a month returned to the team. Whether this pays off depends on your salaries, season and the cost of the tool, so count with your own figures.
A 30-day pilot plan
Week 1: collect the fifty most frequent guest questions from messengers and email, write approved answers and the list of situations that always go to a person.
Week 2: set up the agent on one channel only and test it internally — the team asks real and awkward questions and marks wrong answers.
Week 3: turn it on for guests during limited hours, with a manager reading every conversation at the end of the day.
Week 4: compare the numbers with the starting values, fix the knowledge base and decide whether to add a second channel or the next task.
FAQ: AI and chatbots for hotels
Will an AI agent replace reception? No. It takes repeated questions and preparation work, so the team has more time for guests who are standing in front of them.
Does a small hotel or a villa need this? If you answer the same questions every day and sometimes reply hours later, yes — start with one channel.
Do I have to change my booking system? No. Start without a connection and add it later where it is practical.
Is there a ready-made design? I sell a hotel AI system blueprint with agent roles, workflows, prompts and an implementation plan. It is a document set for you to configure; implementation is a separate service.
Sources and scope
NIST — AI Risk Management Framework ↗
Voluntary framework for managing AI risks, including human oversight and accountability.