Innovation

Every hotel knows the questions. Is the parking on site? Is the room on the street side quiet at night? Can you fit a crib? Is breakfast really included at that rate? For years, a guest who wanted to know asked the hotel, by email, by phone, or through the booking site's message button, before she booked. Someone at reception answered. And a good answer was often the moment the booking was won: yes, the garage is under the building, and the quiet rooms are at the back, ask for one. On Booking.com , that question can now go to an AI instead. Booking.com announced in October 2024 that travelers could ask specific questions about a property and get an answer straight away, generated from the listing, guest reviews and photos. The answer comes back without anyone at the hotel writing it. Booking.com 's AI has answered since 2024 The feature sits in the Booking.com app. When Booking.com announced it, it was live for travelers in the US, the UK, Australia, New Zealand and Singapore. The company said then that it would come to property pages in the near term, and to local-language versions in Spain, Italy, Germany, France, Poland and the Netherlands in the months after. Booking.com described how it works in one line: the AI pulls the relevant information from the property listing, traveler reviews and photos, and presents a short answer. In the same announcement, Booking.com said it was testing a second tool: review summaries, which boil all of a property's reviews down to a few lines, and could later be tailored to what a traveler cares about, such as parking or wheelchair access. It said the summaries would roll out after more testing. Three sources feed the answers. The hotel controls two of them, the listing and its photos. The reviews belong to the guests. The answer comes from your listing and other people's reviews Your listing is what you put into the Extranet: the facilities, the room types, the policies, the description. The reviews are what guests wrote, last month and two years ago. The photos are whatever is on the page. The AI reads all three and writes one answer. Ask whether a hotel is quiet, and it can only use what the listing says and what reviewers mentioned. If three guests last year complained about a nightclub that closed in the spring, that can still be in the answer. If the listing never says the garage is under the building, the AI can't say it either. The same goes for the crib, the connecting rooms, the late check-in after midnight: if it isn't a field you filled in, or a line in your description, the AI has to find it in someone's review or leave it out. Google works the same way. Its developer documentation describes AI summaries of places, written from user reviews alone and labeled "Summarized with Gemini" in the apps that show them. The only fix the documentation describes is

I recently wrote about AI helping a hotel anticipate what a guest needs, and I used a small example. Imagine the system predicts that a guest will want more towels before they ask. Clean, simple, a little bit of magic. Then someone left a comment I haven't been able to stop thinking about, and I want to build on it here because it reaches far beyond hotels. She said, more or less, that predicting the towel is the easy part. In a lot of independent hotels, there is nobody available to actually bring it. The property cannot get its basic systems to talk to each other without manual workarounds. Front desk turnover is high. Housekeeping is so stretched that no one has a spare minute for predictive anything. Her point landed because it was true. My tidy example made a hard operational reality sound simple, and it is not simple at all. So I want to sit with his challenge honestly, because the answer to it is the most useful thing I can say about AI right now, and it applies to almost any team running on a skeleton crew, in any industry. The question most AI conversations skip Most of the AI conversation assumes the problem is a lack of intelligence. If we could just predict better, recommend better, personalize better, everything would improve. The towel comment exposes the flaw in that assumption. Plenty of teams are not short on insight. They are short on hands, on hours, and on attention. For them, a smarter prediction is not a gift. It is one more thing flashing on a screen that nobody has the capacity to act on. I see this everywhere once I started looking, and not only in hotels. The understaffed clinic. The restaurant running three people short on a Friday. The small shop where the owner does payroll at midnight. The nonprofit where everyone wears five hats. In all of them, adding another dashboard, another alert, another clever system makes things worse if it adds work without removing any. Intelligence laid on top of exhaustion just becomes noise. So the question is not how do we make this team smarter. The question is what is quietly stealing their capacity, and can we give some of it back. For an overstretched team, AI should subtract before it adds Here is the shift I would ask any stretched operation to make before spending a dollar on AI. Start with subtraction. Where is your team entering the same information twice? Which workaround eats twenty minutes of every shift? What task exists only because two systems refuse to talk to each other? These are not glamorous problems, and they are rarely what the AI pitch is about, and they are exactly where the capacity went. A front desk clerk re-keying a reservation from the channel manager into the PMS is not short on intelligence. That person is short on time, and the time is being eaten by a gap no one