What Hotel CIOs Actually Learned at Destination AI Forum — and What Your Tech Stack Should Do This Quarter

I spent the morning of October 2 reading the Skift recap of Destination AI Forum on a hotel veranda in Coral Gables with a flat white going cold beside my laptop. The general manager had just walked me through the property’s “AI roadmap” — a printed deck, six initiatives, three vendors none of his team had actually deployed yet. He wanted my read. My read, after a week of reporting around the New York event and the post-show conversations operators have been having on group chats and calls, is this: stop reading brand decks for direction. The chain CIOs onstage at the Forum on October 1 told everyone, plainly, that they don’t have the answers yet. They said “early days.” They said “experiments.” They said it about call-center AI, about generative search, about robotics, about the works.

That admission is the most operator-useful thing that came out of the conference, and most independents and small groups are misreading it. The reflex is to assume the chains know something they’re not sharing — that “early days” is a tactical understatement, and the right move is to wait for Marriott or Hilton to publish a playbook. The opposite is true. “Early days” is permission. It is permission to ship narrow, measurable AI on your own timeline, against your own P&L, without waiting for a brand-led integration that may never come on a schedule that suits your asset. My contrarian thesis for Q4 2025 is therefore unfashionably specific: the right operator move this quarter is a phone-AI pilot plus a revenue-management refresh, in that order, and not — under any circumstance — an LLM-everywhere strategy.

I’ll walk you through what was actually said in New York, what the chain CIOs are betting on under the hood, what the Mews acquisition of DataChat signals about where the PMS layer is going, and exactly what I would put in front of an ownership group between now and December 15. If you’re reading this in a corner office on Brickell or above a 38-key boutique in the Cotswolds, the playbook is the same. The capital intensity is different. The discipline is identical.

What “Early Days” Actually Means When Six CIOs Say It On Stage

Six chain CIOs and senior tech leaders sat in front of a room full of operators on October 1 and used variants of the same phrase. “Early days.” “Experiments.” “We’re learning.” The Skift recap captured the tone faithfully — there were no triumphalist deployment numbers, no “we automated 60% of inbound.” There was a lot of careful framing about test environments, about call-center pilots, about generative search tools that are interesting but not yet revenue-attributable.

Read at face value, that’s striking. These are the technology executives at Marriott, Hilton, IHG, Wyndham, BWH, and adjacent groups — the people whose budgets dwarf any single independent. They have access to vendors at terms the rest of us don’t get. They have data lakes the rest of us would kill for. And they are publicly telling a room full of competitors that, twelve months into the post-ChatGPT era of hotel tech, they do not yet know which AI bets will compound and which will be quiet write-offs.

Three things follow.

First, the gap between “chain has it” and “chain has it working” is wider than the press releases suggest. When a brand announces an AI deployment, you are typically reading about a pilot in a small subset of properties, on a subset of use cases, often with a human in the loop. The forthcoming chain-level case studies that operators keep waiting for — including a forthcoming piece on Marriott’s AI deployment that goes deep on what’s actually live across the estate — will likely confirm what the CIOs said on stage. The marketing language is running ahead of the operational reality. That is not a scandal. It is the normal cadence of enterprise software. It is, however, a permission slip for everyone smaller.

Second, the chains are not coordinating. There is no shared playbook. The CIO of one major brand told a small group after the panel that his team’s call-center AI roadmap is “completely orthogonal” to what a peer brand is doing, and that the two teams are not in regular dialogue. Operators who assume the chains have agreed on a direction and will pull the franchise estate toward it are operating on an outdated mental model. The brands are running parallel experiments and the franchisees, in many cases, are the test population whether they signed up to be or not.

Third — and this is the operator-actionable bit — the experimentation gap means that for the next two-to-four quarters, an independent or small group can move faster on a narrow problem than a brand can move on the same problem at scale. That is a genuine arbitrage. It is not infinite. It closes when the chains’ tooling matures and franchisees inherit it. But it is open today.

The Phone Is Still the Highest-Leverage Surface in 2025

Of everything that came up at the Forum, the use case the CIOs spoke about with the most quiet conviction was call-center and front-desk phone AI. Not chatbots on the website. Not concierge agents in the room. Phone. The unloved, voicemail-haunted phone.

There is a reason. The economics of the front-desk phone at an independent or small-group hotel are brutal. A 120-key property might run 200-400 inbound calls per day across a 24-hour cycle. Of those, a meaningful fraction are repetitive: reservation modifications, directions, Wi-Fi questions, late check-in confirmations, restaurant hours, pet policies, parking. Each call eats roughly two to four minutes of agent time. Multiply that out and you are spending six to twelve labor-hours per day on calls that, if handled by a competent voice agent, could be deflected at marginal cost.

The chain CIOs at the Forum confirmed that this is where their most credible internal numbers live. The Skift writeup of the panel cited call-center deployments as the category drawing the most serious experimentation. That tracks with what I’ve been hearing from vendors all summer. The voice models that came out of the major labs over the spring and summer are now good enough — meaningfully good, not press-release good — to handle the long tail of repetitive inbound at a 90%+ first-call resolution rate on a tightly scoped intent set. The technology is no longer the bottleneck. The bottleneck is integration: connecting the voice agent to your PMS, your reservation engine, your CRM, and your incident-management workflow.

For an independent operator in Q4 2025, a phone-AI pilot is the single highest-leverage AI investment available. Here is what I would scope, against a 60-day window, with a 120-300 key property in mind.

Scope the intents narrowly. Pick the five inbound call types that, in aggregate, represent at least 50% of your repetitive volume. For most properties, that’s: reservation modification, arrival ETA, basic property questions (Wi-Fi, breakfast hours, parking), restaurant reservations, and outbound transfer-to-human for anything outside the list. Do not, in week one, try to handle group inquiries, billing disputes, or guest complaints. Those are human problems and they will remain human problems for several more quarters.

Pick a vendor that integrates with your existing telephony. This is not the moment to rip out your PBX. Look for a vendor that overlays on your current setup, intercepts calls before they hit the queue, handles the narrow intent set, and warm-transfers everything else. The good vendors in this category will let you ship a pilot on a single property in six to eight weeks. The ones that quote you a six-month integration are selling you a different product.

Measure three numbers. First-call resolution rate on the in-scope intents. Average handle time across all inbound (including the transfers). Net revenue impact from upsell prompts on inbound reservation calls. The third one is where the math gets interesting — even a 4% lift in upsell attach on inbound reservations, at a 120-key property running 90 inbound reservation calls per day, is real money over a quarter.

Pre-commit to the exit criteria. Before you sign anything, write down the numbers at which you scale to a second property and the numbers at which you walk away. Independents who skip this step end up running zombie pilots for nine months because nobody wants to be the one to call it.

Why Revenue Management Is The Other Q4 Move — and Why It’s Not Optional

The second leg of the Q4 stack move is less glamorous than phone AI and more important. The chain CIOs at the Forum touched on it obliquely, but the through-line is unmistakable: Marriott, Hilton, IHG, Wyndham, and BWH are all, on different timelines, shifting their pricing posture from “maximize occupancy and ADR” toward “maximize total profit per available room across the full stay.” That sounds like a slogan. It is not. It is a structural change in how rate decisions are made.

The traditional revenue-management discipline — and the one most independents still operate under, even if their RMS is technically modern — treats room revenue as the dependent variable. You pick a rate, you forecast demand, you sell the room. F&B, parking, spa, ancillary — those are second-order concerns, handled by separate departments with separate P&Ls, often with no formal feedback loop into the rate decision. The chains are moving away from that. They are building (or, more accurately, buying) revenue-management tooling that prices the room with explicit reference to the expected total spend of the guest who will book at that rate. A $50 rate cut that brings in a guest who spends $400 across two nights on F&B and spa is, in this framing, a margin win. A $50 rate increase that suppresses bookings from high-ancillary-spend segments is a margin loss, even if RevPAR ticks up.

This is the right way to price a hotel in 2025. It is also operationally difficult, because it requires clean data flowing between your PMS, your POS, your spa booking system, and your RMS, and most properties don’t have that. The chains will solve this with money and time. Independents have to solve it with a focused refresh and discipline.

What does “revenue management refresh” actually mean for a Q4 operator? Three concrete moves.

One: get your data plumbing audited. Before you change your RMS or your pricing posture, you need to know what data is actually flowing between systems. Most properties I look at have at least one broken pipe — a POS that’s not pushing guest-level spend data back to the PMS, a spa system that lives in a complete silo, an OTA channel manager that’s overwriting segment codes. A two-week audit by a competent integrator will surface these. Until they’re fixed, anything you do at the RMS layer is operating on incomplete data.

Two: introduce ancillary-spend forecasting into the rate decision, even crudely. You do not need a $300K total-revenue-management platform to start doing this. You can start with segment-level ancillary spend averages from your last twelve months of stayed-guest data, applied as a modifier against your published rates. A leisure guest segment that historically spends $180/night on F&B and spa should be valued differently at the rate decision than a corporate negotiated rate with zero historical ancillary attach. Hardcoding even a rough version of this into your weekly pricing meeting will get you 60% of the way to where the chains are heading.

Three: align your commercial team around total profit, not RevPAR. This is the hard part because it requires changing how people are compensated and how performance is reported up to ownership. If your director of sales is comped on rate and your director of F&B is comped on covers, they will fight each other on group business that depresses ADR but lifts banquet revenue. Get the comp structures aligned with the new pricing posture before you change the pricing posture.

I want to be honest about what I’m seeing in the market. Most independents are not doing this. Most independents are still running pricing meetings that look exactly like they did in 2019, with a slightly newer RMS interface and the same KPIs. The chains will, over the next four to eight quarters, build a meaningful pricing advantage at the margin. The independents that close that gap will do so by being earlier and more disciplined, not by buying more software.

The Mews-DataChat Question, Honestly Handled

I want to flag a source-confidence issue here rather than paper over it. The Mews acquisition of DataChat — a natural-language analytics layer that sits on top of the PMS data — has been circulating in operator conversations as an October event. I have not been able to find a primary, dated October source for the announcement that I can hand you. The anchoring for the acquisition timing currently lives in Mews’s own forthcoming end-of-year recap, which itself is not yet published as of this writing. Treat the specifics as anticipated rather than confirmed.

What I can talk about with more confidence is the strategic logic, because the logic is independent of the exact deal date. The PMS vendors are in an arms race to add a usable natural-language analytics layer to their products. The hotel general managers I talk to do not want another dashboard. They want to be able to ask, in plain English, questions like “show me the segment mix on Tuesday nights in October versus the same period last year, broken out by booking channel, with average ancillary spend per stay.” The technology to do that competently has existed for about eighteen months. The integration of that technology into the PMS layer, where it can actually answer those questions against your real data, is the move every serious PMS vendor is making this year.

For a Q4 operator, the relevant question is not “should I switch to Mews because of DataChat.” The relevant question is: does my current PMS have a credible plan to deliver a natural-language analytics layer in 2026, and if not, when does the migration math start to favor switching? Most operators are at least one PMS migration cycle away from needing to make that call. But if you are running a PMS that is clearly not going to ship this capability, you should be quietly modeling the switching cost now, because by 2027 the gap will be operationally visible and the smaller groups will be moving.

This connects, in a sideways way, to a thread I’ll pick up in an upcoming May vendor comparison between SevenRooms and Tablecheck — the same dynamic is playing out one layer down, in the F&B-side guest-data stack, and the strategic question for operators is the same one: which vendor is going to ship a usable AI layer first, and on what timeline.

What I Would Not Do This Quarter

I want to be direct about the things I’d push back on if an operator brought them to me in October.

Do not commission an “AI strategy.” I have read four of these in the last six weeks. They are all the same document. They identify between fourteen and twenty-two potential AI use cases across the property, score them on a two-axis matrix, and recommend a “phased rollout” that conveniently begins with the use cases the consultant happens to have a relationship with. The output is invariably worse than just picking two narrow pilots and shipping them.

Do not deploy a guest-facing chatbot on your website without a clear job. Guest-facing chat is the most-pitched AI use case in hospitality right now and one of the lowest-ROI ones for independents. The traffic on most independent hotel websites is too low for chat to deflect meaningful labor, and the guest who’s ready to book usually wants to book — not to chat with a bot about pillow options. If you must do guest-facing chat, scope it to a specific job (e.g., recovering abandoned bookings on the IBE) and measure it ruthlessly.

Do not buy a “concierge AI” for the rooms. The in-room voice-assistant category is having a moment in vendor demos and a deeply mediocre moment in the field. The guest research is unambiguous: in-room voice has very low utilization at most price points and the labor savings are not there. There may be a luxury edge case where the brand storytelling justifies the spend. For everyone else, this is a distraction.

Do not, under any circumstance, buy a robot. The Skift recap of the Forum noted, drily, that the U.S. is not the robotics market. The CIOs onstage were polite about it but the message was clear: the labor math, the regulatory environment, the guest expectations, and the maintenance burden all argue against robotics deployment in U.S. hotels in 2025-2026. The Japanese and Korean markets are a different story. If you are operating in those markets, this paragraph does not apply to you. If you are operating anywhere else, save the capex.

The 90-Day Plan I Would Hand to Ownership

If I’m sitting across from an ownership group in the second week of October with a single page in front of me, here is what’s on it.

Weeks 1-2: Data plumbing audit. Hire an integrator (not a strategy consultant) for a two-week diagnostic of your PMS-POS-RMS-CRM data flows. Output is a one-page diagram of what’s flowing where, with broken pipes flagged. Budget: $15-25K.

Weeks 3-6: Phone-AI vendor selection and contracting. Run a three-vendor bake-off on a narrow intent set. Sign a 90-day pilot contract on a single property. Insist on month-to-month after the pilot.

Weeks 5-8: Revenue-management posture alignment. Get your commercial leadership in a room and rewrite the weekly pricing meeting agenda. Introduce segment-level ancillary spend modifiers into the rate decision. Realign comp structures by the end of Q4.

Weeks 7-12: Phone-AI pilot live. Ship to a single property. Measure FCR, AHT, and upsell lift. Pre-committed exit criteria in the contract.

Week 12: Decision gate. Scale phone AI to a second property or kill the pilot. Re-evaluate revenue-management refresh against measurable margin impact. Do not, at this gate, commission a new “AI strategy.” Repeat the loop with a new narrow pilot.

That is roughly $80-150K of all-in spend across the quarter for a small group, depending on portfolio size and current state. It is meaningfully smaller than what most properties are about to commit to “AI transformation.” And it is, in my read of where the chains actually are versus where they’re claiming to be, the disciplined path.

Mark Interpretation: What the Forum Tells Us About the Next Eighteen Months

Let me pull back and try to mark the larger picture, because the operator playbook above is only useful if the strategic frame holds.

I think the most important signal from the Forum is not what was said about specific use cases. It is the tone. Six senior chain technology leaders, in front of a room full of operators and vendors, declined to claim victory. That is unusual. The default mode in hotel tech for the last twenty-four months has been competitive PR posturing — every chain announcing AI partnerships, every vendor announcing chain customers, the trade press dutifully reproducing it. The shift to “early days” framing on a public stage suggests one of two things, and they have different implications.

The first possibility is that the CIOs are managing expectations downward because the internal results are below what the marketing implied. The deployments are smaller, the ROI is harder to measure, the integration burden is higher than expected. In this reading, the Forum tone is a soft pre-announcement of the gap between hype and operations, and we should expect a noticeable cooling in chain-led AI announcements over the next two-to-three quarters as the focus shifts to making existing pilots actually work. This is, broadly, the reading I’d assign more weight to.

The second possibility is that the CIOs are managing expectations downward because the real deployments are about to step-function up, and they don’t want to be embarrassed by the velocity. This would be the optimistic reading — that the next year sees a meaningful shift from pilot to scale across the chain estate, with measurable guest-experience and labor impact. I would not bet the property on this reading, but I’d assign it nontrivial probability for at least one or two of the brands.

Either way, the operator implication is the same. Move now on narrow, measurable wins. Do not wait for chain playbooks. Do not commit to platform-level AI deals on the assumption that the brand will pull you into the future. The brands are figuring it out in real time, and they have publicly told you so.

The independents and small groups who use the next two quarters to ship two or three disciplined pilots — phone AI, revenue-management refresh, maybe one operational efficiency play in housekeeping or maintenance scheduling — will end 2026 with a meaningful operational advantage over peers who spent the same period commissioning strategy decks. That advantage will close eventually. It is open now. The CIOs onstage in New York told you it was open. They were honest. I’d take them at their word.

The flat white on the veranda in Coral Gables is fully cold by now. The GM is asking me what’s on the page. The page says: phone AI, RM refresh, two pilots, ninety days, measurable exit criteria. The page does not say “AI strategy.” It is a quieter document than the one he brought to me. It will, I think, do more work.

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