Six Hotel CIOs Admit AI Is Still 'Early Days' at Destination AI Forum

I spent yesterday at the Destination AI Forum in Washington watching six of the most powerful technology leaders in American hospitality say, in different words, the same sentence: we don’t really know yet.

That is the headline. Not the new tools. Not the partnerships. The fact that the chief technology and information officers of IHG, Wyndham, Choice, Hilton, Hyatt and Marriott — companies that together touch something like a third of all branded hotel rooms in the United States — described their generative-AI programs as “experiments” and used the phrase “early days” without flinching. Nobody quantified savings. Nobody put a number on labor displaced, conversion lifted, or call-center minutes shaved. Skift’s coverage of the day captured the tone: this is a generation of executives talking carefully on a stage that, eighteen months ago, would have been full of swagger.

I want to argue that this is the most useful thing that happened in hotel technology this quarter. Public, coordinated honesty from the top of the segment is permission for everyone else — independents, boutique groups, the regional management companies I write about most weeks — to stop pretending they are behind. They are not behind. Nobody is ahead. The race the trade press has been narrating since ChatGPT shipped has not actually started.

The shape of the admission

What the CIOs said, sorted into buckets:

The first bucket is search and discovery. Several chains have wired generative models into their booking surfaces — natural-language search, itinerary suggestions, the now-standard “find me a hotel near the convention with a pool and late check-in” prompt. The honest framing was that these are pilots running alongside the existing booking funnel, not replacing it. Conversion data exists internally; none of it was shared in the room.

The second bucket is the call center. Every major chain is running some flavor of agent-assist — transcription, suggested replies, post-call summarization — and at least two are testing customer-facing voice agents for tier-one queries. Again: no quantified deflection rate, no published average-handle-time delta. The phrase that kept recurring was “we’re learning what we’re comfortable with.”

The third bucket is the back of house — scheduling, demand forecasting, revenue management copilots, housekeeping route optimization. This is where the most actual money is, and it is also where the chains were least specific. One CIO said, almost in passing, that the most useful AI deployment in their portfolio so far was a forecasting model that nobody on stage would call AI five years ago. That is the tell.

The fourth bucket is what was conspicuously missing: humanoid robots. Skift made a point of this, and it matched what I saw. The robot stories driving headlines out of Japan, Korea and the Gulf are not crossing into US deployments at any scale. The CIOs were polite about it but unanimous: labor law, guest expectation, ROI math and the simple reality that an American hotel lobby is not a controlled environment make the business case fragile. The future of hotel AI in the US, on the evidence of this stage, is software that nobody photographs.

Why “early days” is the story

Here is the contrarian read. For two years the dominant narrative in hospitality technology has been that the chains are racing ahead and the independents are about to be left behind. Vendor decks have leaned on this. Conference keynotes have leaned on it harder. The implicit threat to every general manager I talk to has been: if you don’t move now, Marriott’s bot will eat your direct bookings.

Six chain CIOs publicly admitting “early days” is permission for everyone else to stop running on that fear. The infrastructure isn’t built. The integrations aren’t stable. The ROI cases aren’t proven inside the chains, never mind exportable to a 40-room boutique. The honest move for an independent operator in October 2025 is not to buy the AI suite a vendor is pitching this week. It is to pick one painful workflow — group-RFP response, post-stay survey triage, the call center after 10 p.m. — and run one careful experiment with one tool, the way the chains are.

I’d flag the chains’ caution against my own forthcoming May piece on Marriott’s deployment cadence — the pattern I’m watching is whether the language tightens up between now and then, or whether “experiment” is still the word in eight months.

What I’d watch next

Three things from the floor that didn’t make the stage but seem load-bearing.

First, the procurement conversation is shifting from “buy the platform” to “buy the workflow.” The chains are increasingly comfortable wiring multiple model providers behind a single internal layer and swapping them out. That has implications for any vendor selling a single-model promise — and for any operator being asked to commit to one.

Second, the data side is finally catching up to the model side. Several CIOs referenced internal projects to consolidate the guest record across PMS, CRS, loyalty and CRM before any meaningful guest-facing AI can ship. This is the unsexy work, and it is where the money is being spent. If you’re an operator wondering why the visible AI features feel thin, this is part of the answer: the plumbing is being rebuilt underneath, and the chains are not going to wire a generative agent into a guest-facing channel until they trust the data it draws from.

Third, the labor framing has matured. Two years ago the AI conversation at hospitality conferences was uncomfortable because nobody wanted to say “replace staff” out loud. Yesterday the framing was almost entirely about freeing the staff you can’t hire — overnight desk coverage, multilingual support, the after-hours phone — rather than thinning the staff you have. That reframing matters. It changes the political economy of every pilot the chains run.

The takeaway for operators

Read the candor as a gift. The companies with the largest technology budgets in the segment are telling you, on the record, that they have not solved this. They are experimenting. They are measuring. They are not deploying humanoid robots. They are not replacing their front desks. They are wiring careful pilots into specific workflows and watching the numbers.

The right posture for an independent operator reading this, sitting on a smaller budget and a tighter team, is not to feel behind. It is to feel invited. Pick one workflow. Run one pilot. Measure honestly. When the chains start quoting numbers from the stage instead of adjectives, that is when the race actually begins.

Mark this as the quarter the swagger ended.

— Naomi covers hotel F&B and operator tech for TableTransfers. Tips: [email protected].