How much money AI puts into a hotel: the real 2026 numbers

In my first review of hotel AI cases with ROI I already set the rule: if no one outside the vendor has verified the number, it's not data, it's a brochure. A year later, the list of cases with real, verifiable figures has grown — not by much, but enough to update it with new data instead of repeating the same three names.

Wyndham, again, with the missing figure

We already knew the Wyndham case from its quarterly results: 250 voice agents, more than 500,000 interactions handled, 25% less average handling time. What was published in May 2026 is the piece that was missing, the one that really matters to whoever runs the business and not just whoever measures operations: the hotels that used Wyndham Connect the most, its AI platform for guest engagement, averaged more than 60,000 dollars in incremental revenue in 2025, and the best-performing one topped 200,000, plus 7% less time on call center calls. Wyndham frames this in the context of more than 450 million dollars invested in technology since 2018 — a long-term investment, not a six-month pilot project.

It's still a figure from the chain itself, so it should be read with that caution. But it is, by far, the most groundable return figure in the whole industry: not an abstract productivity percentage, but a dollar range per hotel per year that any owner can compare against the cost of implementing something similar.

Hilton and Choice: the first operational percentages to come from a panel, not a press release

In June 2026, a panel at the NYU International Hospitality Investment Forum made public what until then had only circulated as intuition: the first concrete operational percentages from two major chains.

Hilton: its global AI messaging handles around 55% of its call center chats and 36% of chats at the individual hotel level. And a figure that says just as much as the previous one: 80% of the recommendations from its AI hotel-lifecycle tool (licensing) were approved by the human reviewers — neither total replacement nor total distrust, a middle ground that is probably the most realistic for any business considering it.

Choice: in its group proposal (RFP) tool, 30% less response time and 250 basis points more conversion. It's a smaller case in scope than Hilton's, but with a very concrete business figure — conversion, not just speed.

Umar Riaz, from EY, took part in that same panel with an idea that's worth clearly separating from the figures above: well-executed agentic AI plans over a 3-5 year horizon could cut costs by between 30% and 50%. It's a projection, not a measured result, and it appears in the same coverage as Hilton's and Choice's real figures — which makes it easy to quote as if it had the same status. It doesn't.

The Spanish case that's still the best with a name attached

Ávoris, in Spain, deploys AI agents built on Azure OpenAI in customer service that automatically resolve more than 70% of its travelers' incidents — they interpret the request and resolve it without human intervention. It's not an informational chatbot: it's an agent that resolves on its own, at level 2-3 of any autonomy ladder. A year later, it's still the Spanish case with the strongest figure in the sector, alongside the 75% of its more than 1,500 daily enquiries that RIU's Claud·IA chatbot was already handling on the spot in 2024, by the chain's own account.

Outside the hotel, the same pattern: customer service is where it already pays

The platforms' 2026 results point in the same direction as Ávoris: where AI already delivers measurable savings is customer service. Airbnb said in its second-quarter 2026 shareholder letter that its customer support cost per booking fell by about 16% compared with the same quarter of 2025, and that almost 45% of issues that start with its AI assistant are now resolved without a human agent. Mind the unit: it is 45% of the issues that go through the assistant, not of all issues.

Club Med offers the version with people in the loop. As the company itself reported at VivaTech (June 2026), its AI agent on WhatsApp, live in 23 markets, handles around 30% of inbound contacts and resolves close to 48% of those on its own. Its chief data and AI officer, Siddhartha Chatterjee, sums it up this way: AI should give time back to people, not replace them. As with Wyndham, these are the companies' own figures, but they confirm the order set out in the last section of this article: start with customer service.

The figure almost no one separates properly: the promise versus the data

Here's the exercise that really tells apart someone who reads this sector well from someone who repeats headlines. EY's 30-50% cost savings figure is a promise over 3-5 years. Wyndham's 60,000-200,000 dollars, Hilton's 55%/36%/80%, Choice's -30%/+250bp and Ávoris's >70% are already-measured data, from this year or the previous one. Both types of figure appearing in the same coverage doesn't make them equivalent.

There's a piece of context that explains why that 30-50% generates so much interest even though it's a projection: labor costs account for between 30% and 40% of a hotel's spend (CoStar, January 2026). It's the denominator behind any agentic savings promise — and also the reason why any conversation about hotel AI ends up, sooner or later, talking about staff.

How to apply this if you run a mid-sized hotel

None of these cases can be replicated as-is at a 40-room hotel — Wyndham builds on 450 million dollars of accumulated investment since 2018. What is replicable is the order: first the smallest, most measurable case (a well-defined customer service task, as Ávoris does), then actually measure the result in money, not just hours saved, and only then consider a second task. Wyndham's figure — dollars per hotel per year, not a productivity percentage — is the format you should demand from any vendor selling you AI: if they can't give it to you in those terms, they probably don't have it yet.