«AI for hotels» case studies: how to read them without getting fooled

Anyone running a property and weighing up an AI tool keeps running into the same kind of material: a case study with a big number up top, plus nineteen percent, plus eighteen percent, fourteen thousand hours saved. The question isn't whether AI works in a hotel: in some cases, for some tasks, it does. The question is whether that specific number, in that specific case study, is true.

Three figures that lead nowhere

While researching this article, three claims surfaced that circulate in the industry's content marketing, and they share the same underlying problem: none of them leads to a source you can check.

The first: a hotel on Italy's Lake Como supposedly saw RevPAR, revenue per available room, the industry's standard measure, rise by 19.3% thanks to an AI project. Searching for this figure turns up no property name, no statement, no document with a methodology: just the number, repeated.

The second attributes to chef Massimo Bottura an AI project at his guest house near Modena, Casa Maria Luigia, applied to the guest experience. A real project does exist, but it's a different one: in September 2025 Google wrote on its own blog about a collaboration with the chef to design, with help from Gemini and Imagen, new staff uniforms for Osteria Francescana's thirtieth anniversary, also worn by Casa Maria Luigia's staff. It's a real project, documented by the party that carried it out, about fashion and workwear design, not a system that handles check-in, pricing or guest requests. The difference between the two isn't a detail: it's the difference between exists and doesn't exist.

The third attributes to the design hotel chain citizenM an 18% RevPAR increase from AI. citizenM itself never made this claim: the real material about the company covers something else entirely, a change of revenue management software vendor, a project on the hotels' network infrastructure, an interview with a former executive about internal automation. None of it contains that percentage.

The check that takes a minute

Before being swayed by a case study, three questions are almost always enough.

Is the name verifiable? "A hotel on Lake Como" isn't information you can check. The name of a real company, with a website and an address, is.

Does the number lead to a primary source, or just to other articles repeating it? Searching the exact figure together with the company's name should lead to the document from whoever measured it, not to a chain of blogs citing each other without ever reaching the origin.

Who published it, and what did they stand to gain? This doesn't disqualify a figure on its own: it's the question that tells you how much skepticism to apply, and what else to check before trusting it.

Not a yes or no: a scale

AI-in-hotels case studies you find online don't split into true and false: they sit on a scale, and the three figures above sit at the bottom, the point with no traceable source at all.

One step up are case studies published by whoever sells the technology, but that disclose how they measured the result. Hotel software vendor Mews published an analysis of over six thousand hotels, matching those that had used its automatic pricing mode for at least nine months against similar hotels that hadn't, over an eighteen-month period: it reports a 13% increase in revenue per square meter, not the industry-standard RevPAR but a different metric, and the probability that the result was down to chance. The study also admits its own limits: the effect across its full customer base wasn't statistically solid, and the nine-month threshold was chosen after looking at the data, not set before looking at it. It's still a figure published by the party selling the product, so it needs the skepticism that position calls for. But at least it can be judged, because the method is written down.

The same holds for the best-documented case on hotel chatbots, the Leonardo Hotels chain: the chatbot vendor publishes the raw numbers, 213 hotels, over 280,000 conversations, the assumed average handling time used to calculate the hours saved, and this lets anyone recalculate the final figure themselves. Doing that recalculation shows how much more modest the result is per single hotel than the headline it's sold with: just over an hour a week, not fourteen thousand hours.

At the top of the scale are studies run by parties with nothing to sell tied to the result. The most solid figure available on actual AI adoption in European hotels comes from a Swiss university institute, funded by a public innovation programme, on a sample of nearly 1,500 hotels across six countries: this study, too, discloses its own limits, a volunteer sample, Greece overrepresented among respondents, but it does so because it has no product to sell alongside the result.

What to ask, before signing anything

Next time a vendor shows a case study with an impressive number, three requests matter more than any generic objection: the name of the company in the case study, the time period measured, and whether anyone outside the vendor verified the figure. A vendor with a solid case answers with precise details, even though it remains their own case, not universal proof. A vendor who answers with other equally vague figures is probably repeating the same chain of citations that produced the Lake Como number: a figure heard somewhere, never verified by whoever told it first.

If you want to know whether the case someone showed you holds up, or whether the tool being pitched actually solves something for your property, the digital check-up is free if you then start one of the subscription services. Otherwise get in touch: send me the case study you were shown, and I'll tell you what's missing to trust it.