AI in hotels and HoReCa: what works, from booking to guest service

HoReCa stands for hotels, restaurants and catering. The pressure in this sector is twofold: guests expect replies at any hour, while labour costs and staff turnover are rising. In peak season, the front desk checks guests in, answers phones and replies to emails all at the same time.

The table below shows the areas where AI makes sense, separating what the machine does from what stays with people. This distinction matters, because in hospitality a badly automated conversation does more harm than no automation at all.

A voice assistant, also called a voicebot, answers the phone, understands a question asked in natural language and replies by voice. In a hotel, it checks availability and rates in the hotel system, takes simple bookings and answers questions about parking, breakfast or pets.

The key is the connection to the hotel management system, known as the PMS. Popular systems, such as Oracle OPERA (formerly Fidelio), Mews, Protel or KWHotel, provide an API, i.e. a link through which other programs read and write data. Before choosing an assistant provider, ask whether it has a working connection to your system, not just a plan to build one.

The law also sets conditions. From 2 August 2026, the assistant must tell the caller that they are talking to AI. If calls are recorded or transcribed, the guest must hear this at the start of the call, together with information on the purpose of the data processing.

Before booking a room, many guests ask about details: is there parking, what time is breakfast, can they bring a dog, what are the cancellation terms. A chatbot on the website, in Messenger or in WhatsApp answers such questions straight away, including at night.

A well-designed chatbot does not stop at the answer. It checks availability and guides the guest to a booking on the hotel’s own website, for which the property pays no commission to a booking portal. That is why it is worth measuring not only the number of conversations but also the number of direct bookings that resulted from them.

In a restaurant, allergens are particularly important. EU rules require guests to be informed about them for dishes served on the premises as well. A chatbot may only give this information from the current menu prepared by the kitchen, and if in doubt it should refer the guest to the staff.

Take a hypothetical 80-room hotel that receives an average of 60 calls a day. Suppose 40% of calls are repetitive questions the assistant can answer on its own. That is 24 calls a day at an average of 3 minutes each, or 72 minutes.

The hotel operates all year round, so 72 minutes × 365 days = 26,280 minutes, or 438 hours a year. That is less than many owners assume. In a property of this size, front desk time saved rarely justifies a voice assistant on its own.

The numbers change once you add calls missed at peak times and after hours. If the front desk loses several bookings a week because nobody answered the phone, work out what they are worth. That figure, rather than the minutes saved, often decides whether it pays off.

Plug in your own figures from the phone system: number of calls, percentage missed and average call length. Without this data, any claimed saving is guesswork.

Setting room rates according to demand is standard practice in the hotel industry, but in a mid-sized property it is often done by the general manager or front office manager, alongside other duties. AI helps because it takes into account more data than a person can review manually: occupancy in previous years, bookings for future dates, events in town and the rates of nearby properties.

The result is a recommended rate for each day, which a person accepts or changes. We recommend working this way for at least one season and comparing the results with the previous year. Automatic rate changes without oversight are worth switching on only once the recommendations have proved accurate.

Tools for such suggestions usually connect to the hotel system and to the program that distributes rates and availability to booking portals. Without history from at least two seasons, AI cannot tell regular seasonality from a one-off event.

Reviews from booking portals, Google Maps and surveys contain knowledge that nobody at the hotel has time to read systematically. AI groups them by topic, such as noise at night, breakfast, parking or cleanliness, and shows which problems come back every month.

For the general manager, such a report is valuable in itself. Instead of a general “guests complain about breakfast”, they can see that the complaints concern weekends and a lack of free tables after 9:00.

The second function is drafting replies in the property’s tone of voice. For positive and neutral reviews, a draft can be approved with one click. A reply to a negative review should be read and signed by a person, because the guest and other readers will immediately recognise a template.

KSeF is Poland’s National e-Invoicing System, run by the government. A hotel issues invoices to businesses and travel agencies in it, while invoices to individual guests can stay outside the system. The largest companies became subject to this obligation on 1 February 2026, and all others on 1 April 2026.

Purchase invoices from Polish suppliers, such as a food wholesaler or a laundry, mostly arrive through KSeF as ready-made data. Reading them from PDFs is no longer necessary. AI is useful for checking them: whether the amount matches the order, whether a price has gone up without warning, and which department the cost should be assigned to.

The same data is used to produce a cost report by department, such as kitchen, rooms and conferences. Management can then see the margin of each part of the property, not just the month’s overall result.

The safest approach is to start with one channel and one type of question, with a measure of success set before launch. Examples of such measures are the percentage of conversations completed without the front desk and the number of direct bookings.

The sequence below works well in properties with a hotel system that can connect to other programs. At the end of the list are situations in which it is better to wait.