“Buying AI is easy. Getting value from it is not.” So said Bhanu Chopra, founder of the software company RateGain, in Lisbon in September, where he presented the State of Distribution 2026 report together with New York University and the industry association HEDNA.
According to the press release on the report, more than half of hotels are using or procuring generative AI. Yet fewer than one in ten say AI has cut their manual work by more than 30%. And more than 80% of commercial teams (sales, marketing and revenue management) still spend one to two days a week building and analysing reports by hand.
My view: in many hotels, the AI is already in place. What is missing is a redesigned process in which it actually takes work off people’s hands. As long as everyone experiments alone in a chat window, the work stays where it was.
What the numbers say
For its AI Opportunity Study 2026, the Ratingen-based consultancy h2c surveyed 113 hotel chains with more than 8,200 properties between them. The study was sponsored by eight technology providers, including Apaleo and Shiji. Its findings are uncomfortable for the industry all the same: 91% of the chains use AI. 13% report a measurable return on investment. 28% have a company-wide AI strategy led by senior management. The chains rate their own AI knowledge at 3.4 out of 10.
For Germany, there are figures covering all sectors. According to the Institute for Employment Research (IAB), the research institute of the German Federal Employment Agency, just under one in four businesses used generative AI in 2025, up from 5% in 2023. Hospitality sits near the bottom at 13 to 15%. Of the businesses that use AI, 90% work with publicly available models such as ChatGPT. Of all businesses using generative AI, 21% have internal rules for it and 27% offer training.
My reading: AI is in use at many businesses. Only a small share of them have rules, training and a measurable benefit to show for it.
Why the returns fail to materialise
In May and June 2026, McKinsey surveyed 1,719 people worldwide, across industries and levels of seniority. Of these, 37% say AI has contributed to their operating profit. McKinsey counts about 6% as high performers: they attribute 5% or more of their EBIT to AI and report significant value. The report is fairly clear about what sets this group apart. Nearly three quarters of them have fundamentally redesigned their workflows because of AI, compared with about a quarter of everyone else. And they are twice as likely to have defined how they measure impact.
Here is how that might look in a hotel. A front desk agent asks ChatGPT to draft a reply to an annoyed guest. That saves her a few minutes. The process stays the same, though: she looks up the booking in the PMS, copies data back and forth, and asks her manager whether she can offer a credit. The minutes saved are scattered across the day and never show up in any KPI.
Reporting is no different. A chat window does not do away with the Monday when someone pulls together figures from the PMS, the revenue system and Excel. A report that populates itself from those sources does, with the AI only explaining the variances. That requires someone to decide where each figure comes from and who signs it off, and the tool is of little help there.
The case against
You could argue that teams learn mainly by trying things out. Wait for a finished strategy before allowing AI and you will have waited too long. And especially in smaller hotels, ROI is hard to measure when the gains amount to minutes at the front desk.
That is fair. Experimenting is the right thing to do, and I would actively encourage it. The trouble starts when an experiment has no goal and no end date. It then runs on because nobody stops it, and nobody knows whether it has achieved anything. If you don’t decide up front how you will measure success, after three months you will struggle to say more than that it felt good.
Five steps for the next 90 days
1. Start with a process. Pick the process that eats the most hours in your hotel and is easy to measure, such as weekly reports, the reservations inbox or shift handover. Only then decide which tool fits. In a group, head office may pick a different process from the properties.
2. Measure first. How many hours a week does the process take today? How often does work have to be redone? Without that baseline, it will be hard to show any effect, positive or negative, in three months’ time.
3. Write down the new process. Where does the data come from, what does the AI do, who signs off, what happens with exceptions? Only once that fits on one page is it worth choosing a tool.
4. Separate the roles. Head office sets the framework: which tools, which data, which rules. The properties find the use cases, because they know the work. In an independent hotel these may be the same two people, but the separation still helps.
5. Decide after 90 days. Keep it, expand it or stop it. All three outcomes are fine, as long as a decision is made.
What I would not do: buy licences for everyone and hope something happens. Nor would I report the number of users as a success. The fact that lots of people open a tool says nothing about whether anything has improved.
The yardstick
I would watch two things: whether someone at the front desk has more time for the guest, and whether the revenue team can spend Monday on decisions rather than spreadsheets. That is what every AI investment should be measured against.
That is why, at 5213 LABS, we would start an AI project by counting hours. Which model fits in the end is often the lesser question.