AI for hotels: where it touches the business
Most hotels are already using AI somewhere, often without calling it that. A survey of more than 500 hoteliers run by the hotel software company Mews in March 2026 found that more than nine in ten use AI in at least one area of their operations, and on average in 11 of the 19 operational tasks the survey tracked. It is a vendor's survey of its own market, so treat it as a sign of how common AI has become rather than a precise measure.
For a GM, the useful question is which uses earn their cost. The easiest way to answer that is to ignore the technology for a moment and ask what each tool is supposed to change on the P&L: more revenue, less cost, or neither yet.
Sorted that way, almost every AI product sold to hotels falls into three columns. The rest of this guide goes through them in order, then gives a simple way to pick a first project and a list of questions to ask any vendor.
Revenue use cases
Start with pricing. Revenue management systems that set rates automatically from demand, pace and competitor prices are the oldest and most mature AI use in hotels. The evidence for revenue management in general is decent but mostly correlational. A Cornell study of US hotels between 2001 and 2005 found a strong correlation between pricing above the competitive set and effective revenue management, while stating that it did not show causality. We found no independent causal study of RMS impact on RevPAR, so treat any "+X% RevPAR" claim as marketing unless the vendor shows you how it was measured. The practical test is simple: run it against the same months last year and against your competitive set.
Upselling is the second. Tools that offer room upgrades, early check-in, breakfast or parking before arrival, priced and timed automatically. The value is easy to measure because every upsell is a line in your PMS. Check the take rate and the average upsell value after three months, and compare against what your front desk sold before.
The third is being recommended by AI. A growing share of guests ask ChatGPT, Gemini or Google's AI Mode where to stay before they search. If an assistant names your hotel and links to your own site, that is a direct booking you did not pay commission on. If it names a competitor or an OTA, the booking goes elsewhere before you know the guest existed. This is the use case Staylight works on: we measure how often AI assistants name a hotel and show what to fix. It belongs in this column because the outcome is direct revenue, and like the others it needs a baseline to prove anything.
Cost use cases
First, guest messaging: AI assistants that answer guest questions by chat, WhatsApp or email before and during the stay: check-in times, parking, the Wi-Fi password, restaurant hours. Guests are open to it. An Oracle and Skift study published in 2022, covering more than 5,000 consumers, found 77% of travellers were interested in using automated messaging or chatbots for customer service requests. That is a survey of interest from a few years ago, not of satisfaction, but it suggests guests will use a good bot.
The saving is staff time, and staff time only becomes money if you do something with it: answer the phone faster, reduce overtime, or avoid a hire. Before you buy, count how many routine questions your team answers per week. If the answer is a few dozen, the saving may not cover the subscription.
Back-office work is the other half: invoice processing, review summaries, rota planning, drafting replies to reviews and emails, translating guest communications. These are often built into tools you already pay for, or cheap general-purpose assistants. The payback is real but small per task, so the case rests on volume.
A warning for both: an AI that answers guests confidently needs correct facts to work from. If your policies, opening hours and room details are not written down accurately in one place, the bot will be wrong in a polite, fluent voice. Fixing that fact base is worth doing first, and it helps every other AI use as well.
Experimental use cases
Booking agents come first here. AI assistants that search, compare and book a hotel on the guest's behalf are moving from demos into products. Google has started rolling out hotel booking inside AI Mode in the US, in English, with integrated partners. How much volume flows through agents, and on what commercial terms, is not yet known. Our guide to AI agents that book hotels covers what is documented and what a hotel can prepare now.
Voice is next: AI phone agents that answer the reservations line, take bookings and handle common requests. The technology works in demos; what matters for a hotel is how it handles a confused caller in a second language at 11pm, and whether it books correctly into your PMS. Pilot it on overflow calls before it touches your main line.
Then personalisation, using AI to tailor offers and emails to each guest's history. This depends on having clean guest data with consent, which many independents do not yet have. It is often a CRM project before it is an AI project.
What to do with this column: keep your facts structured and your data clean so you can move when one of these becomes proven, and spend no more than you would on any other experiment.
Picking the first project: a scoring table
Score each candidate from 1 to 3 on five questions, then add up. The highest total is your first project. The scores are judgement calls, which is fine; the point is to make the judgement explicit and comparable.
Question 1 (weak) 3 (strong) Is there a baseline you can measure against? No data today Clear monthly figure Does it change revenue or cost directly? Indirect Directly on the P&L How soon would it pay back? Over a year Within a quarter Do you already have the data it needs? Needs new data Uses existing data Can you run it without new staff or skills? Needs a specialist Runs inside a team
Pricing and upselling tend to score well for hotels with a decent PMS and steady demand. Guest messaging scores well for hotels with high volumes of routine questions. Being recommended by AI scores well on baseline and payback for hotels with a high OTA share, since every direct booking it wins saves a commission. Experimental projects usually score low on baseline and payback, which is the honest reason to keep them small.
Questions to ask any vendor
What will you measure, and what is our baseline today? If a vendor cannot tell you the starting figure, you will not be able to tell whether the tool worked.
Where does the uplift figure in your sales deck come from? Ask for the comparison period, whether it controlled for seasonality and market demand, and whether anyone independent checked it. Most uplift claims compare a good year with a bad one.
What data do you need from us, where is it stored, and what happens to it if we leave? Guest data in the EU comes with GDPR obligations that stay with you, not the vendor.
What does it do when it is wrong? Ask how the system handles an answer it is not sure of, a price it should not set, or a guest who wants a human.
How long until we can tell whether it is working, and what happens if it is not? A contract you can leave after a fair trial period is worth more than a discount on a long one.