Marriott's Members Day Is the Direct-Loyalty Playbook. Here's How to Copy It.

I was halfway through a flat white in a Bethesda hotel lobby this morning when a Bonvoy Ambassador two tables over told her husband, in the most casual voice imaginable, that she had just used a “chat thing inside the app” to swap an upgrade for late checkout. He nodded the way husbands nod when they are not really listening. She didn’t repeat herself. She didn’t need to. The detail I want every hotel operator reading this to sit with is that the swap happened in roughly forty seconds, with no phone call, no front-desk negotiation, and no escalation to a property GM. That is the entire thesis of Marriott’s Members Day in one anecdote — and the thesis I want to argue today is uncomfortable, so I’ll lead with it.

Hotel operators have a 12-month window — call it through next September — before a generation of AI booking agents disintermediates the direct channel in a way the OTAs never managed. Marriott’s Sept. 23 event is the template, not the press release. Read it as a playbook and copy it. Read it as a marketing stunt and you’ll spend 2027 paying twenty-five-point commissions to an agent your guest doesn’t even know they’re using.

The press version of today’s news is tidy enough. Marriott flew seventy-five of its highest-value Bonvoy customers to its Bethesda headquarters. CEO Anthony Capuano gave remarks. Peggy Roe, the chief customer officer who owns loyalty, walked members through a prototype AI chatbot designed to live inside the Bonvoy mobile app. The company emphasized — repeatedly, and in language that anyone who has briefed a legal team will recognize — that the chatbot is in test, not in production, and that “significant testing and attention to privacy” is the prerequisite for any wider rollout. Six properties are already in beta for the company’s next-generation reservation, property-management, and loyalty platforms. CIO Naveen Manga has told the trade press that the company has prioritized ten high-impact generative-AI use cases for 2025, and that the cloud-native rebuild is the foundation that makes any of them deployable at scale.

That is the press version. The operator version is more interesting, and a little more frightening. Let me walk you through what I think Marriott just did, why I think they did it now, and — most importantly — what the rest of the industry should rip off before the agent layer eats the front door.

The Event Is the Strategy

Start with the optics, because the optics are the strategy. Marriott did not unveil this chatbot at a conference. They did not seed it to a handful of friendly reporters under embargo, the way most hotel groups would, and they did not put it behind a partner-only NDA at a vendor summit in Vegas. They flew seventy-five members — not analysts, not journalists, not influencers, members — to headquarters, sat them in a room with Capuano and Roe, and let them poke the prototype. The press release came second.

That sequencing matters. Marriott is, in effect, building the AI product the same way the best consumer-software companies build features: in front of real users, with real loyalty stakes, and with the visible endorsement of the CEO. The implicit message to a Titanium or Ambassador-tier member is: you helped shape this; you saw it before the Skift desk did; you are not a focus group, you are a co-author. The implicit message to the rest of the loyalty pyramid is: if you want to be in the room when the next one happens, climb.

If you run loyalty at a top-thirty chain — and I know several of you read this column — that is the part to internalize before the part about chatbots. The product is being built in public, with the highest-value customer cohort acting as both QA and word-of-mouth distribution. Every member who left Bethesda today is, by mid-October, going to be the loudest evangelist in their network. None of them needed a press kit.

The Twelve-Month Window

Now the thesis. I keep saying twelve months, and I should defend the number, because if I’m wrong about the window, the rest of this column reads as panic instead of urgency.

The AI booking agent that disintermediates the hotel direct channel is not theoretical. The current generation of consumer agents — the ones that live inside the major model providers’ assistants, the ones that live inside Apple Intelligence, the ones being prototyped inside the big card networks’ concierge products — already book restaurants, already buy concert tickets, and are weeks, not quarters, away from competent hotel booking. The reason they haven’t already eaten the category is mundane: hotel inventory is fragmented across roughly a dozen reservation platforms, rate parity enforcement is uneven, and the call to a property’s API is just hard enough that the agents currently fall back to OTAs. The OTAs, who have spent fifteen years making themselves the path of least resistance for any third-party piece of software, are about to be the path of least resistance for the agents too.

Twelve months is the window because that is, roughly, how long it takes a major consumer-AI provider to ship a stable agent-to-merchant protocol, get the OTAs to certify against it, and switch on a default booking experience inside a billion-user surface. If the chains do not have their own agent-readable surface — a chat, an API, a structured loyalty handshake — by the time that switch flips, then their direct channel becomes a thing that humans use on purpose, and the agent layer routes around them. That is the disintermediation scenario, and it is not a 2028 story. It is a Q4 2026 story at the latest.

Marriott appears to believe this. The shape of today’s event — chatbot inside the app, top members as testers, CEO present, CIO quoting ten prioritized use cases — is the shape of a company that has done the math on the window and concluded that the front door has to be defended now, with a Marriott-branded conversational surface, rather than later, when the only conversational surface available to a guest is the one inside their phone’s operating system. The bet is not that members will love the chatbot. The bet is that members will use the chatbot enough times that, when a generic agent offers to book them a hotel six months from now, the muscle memory routes them back through Bonvoy.

That is a defensive bet dressed as an offensive one. It is also, in my read, the correct bet.

What’s Actually Inside the Chatbot

I have not been hands-on with the prototype, and I want to be honest about that, because in this column I try very hard not to launder a press demo into a product review. What I can do is read the public surface area and the language Marriott used.

The prototype lives inside the Bonvoy mobile app, which is the right place for it. It is not a website widget, it is not a phone tree replacement, and — critically — it is not yet a booking agent. The descriptions today emphasized stay management: questions about reservations, requests for amenity swaps, the kind of high-frequency, low-stakes interactions that today funnel through Bonvoy customer care or, in a worst case, a property’s front desk at 11 p.m. on a Tuesday. That is a deliberate choice. You do not introduce conversational AI to the highest-value cohort of your loyalty program by letting it touch revenue on day one. You introduce it on tasks where the failure mode is mild and the convenience gain is enormous, and you let the cohort tell you what they want it to do next.

Mark how Roe and her team framed privacy in the room. “Significant testing and attention to privacy” is the phrase that made the Skift write-up, and it is exactly the phrase a chief customer officer says when she has been through a board-level briefing about what guest data can and cannot enter a model. The chatbot, in other words, is being built behind the kind of data perimeter that most hotel groups have not yet built and most certainly haven’t yet had to defend. That is also a strategic asset. A guest who has been burned by a generic agent leaking preferences to a third party will, eventually, prefer the chain whose chatbot has the longest, dullest legal disclaimer.

The Cloud Migration Nobody Wants to Talk About

You cannot deploy a chatbot of any consequence on top of a forty-year-old CRS. This is the part of the Marriott story the trade press tends to underplay, because it is unglamorous, but it is the part that operators most need to copy.

The Hotel Technology News piece on Marriott’s strategy is, in essence, a status update on a multi-year platform replatform: cloud-native reservation system, cloud-native property-management system, cloud-native loyalty stack, unified guest data underneath all of it. Six properties are in beta on the next-generation versions. The chatbot is not a feature that got bolted onto the existing stack — it is the first user-visible artifact of a stack that has been quietly rebuilt to support exactly this class of feature.

I cannot stress enough how much this matters for the rest of the industry. If you are a regional chain, or a soft-brand collection, or an independent group of forty properties, and you are reading the Marriott news and thinking we should also have a chatbot — you should not skip to the chatbot. You will burn $4 million on a vendor that wraps GPT-class API calls around a brittle integration to your PMS, and the moment a guest asks something the integration cannot answer, the chatbot will hallucinate or apologize or both, and your highest-value guests will quietly stop using it. The chatbot is the visible tip; the platform underneath is the part that determines whether the tip works.

If your CRS is on-prem, your PMS is on-prem, and your loyalty data lives in a SQL database that is queried by a vendor who charges per-call, you do not have a chatbot problem. You have a platform problem. The chatbot is the deadline.

Ten Use Cases, Prioritized

CIO Naveen Manga’s quote to CIO Dive — that Marriott has prioritized ten high-impact generative-AI use cases for 2025 — is the operator-actionable line in the entire week’s news cycle. Note what he said and what he did not say. He did not say “AI strategy.” He said “ten prioritized use cases.” Those are completely different statements.

The first signals a company that is going to spend two years on slide decks and pilot programs that never escape pilot. The second signals a company that has, in the language of any product organization that has shipped software at scale, ruthlessly cut the list to ten things, picked owners, set deadlines, and accepted that anything not on the list does not get budget this year. If you are an operator with a CIO or a CTO, the question you should be asking on Monday morning is not what’s our AI strategy. It is what are our ten use cases for the next twelve months, who owns each one, and what gets cut.

I would expect Marriott’s ten to include, in some order: a guest-facing conversational layer (the chatbot we saw today); an agent-facing structured booking interface so that, when third-party agents come knocking, Marriott can answer them in their own protocol; a revenue-management overlay that uses unified guest data to price non-room ancillaries; a property-operations copilot for general managers and department heads; a knowledge layer for customer-care agents that compresses a fifteen-minute call to four minutes; a fraud-and-trust layer that, in the agent era, becomes existential because every booking has to be checked for adversarial agent behavior; a marketing-personalization layer; an in-stay upsell layer; a workforce-scheduling layer; and at least one F&B-specific use case that I will be writing about separately once I find the right operator who will go on the record.

That is my list, not theirs. I would bet money on at least seven of those ten being on the real list.

Why F&B Operators Should Be Reading This Twice

I spend most of this column on hotel food and beverage, and you may be wondering why I’m spending three thousand words on a loyalty chatbot. The answer is that the loyalty chatbot is the front door, and F&B is the room that the front door opens onto.

If a guest is using a Marriott-branded conversational surface to manage their stay, then every F&B decision that surface routes — the dinner reservation in the lobby restaurant, the breakfast room-charge, the rooftop bar suggestion at 6 p.m. on a Thursday — flows through Marriott’s data stack rather than through Resy or OpenTable or a generic agent. That is enormous. F&B at the major chains has been chronically underserved by the loyalty program because the loyalty program has not had a high-frequency conversational touchpoint with the guest during the stay. Once it does, every meal becomes a Bonvoy moment.

Independents and smaller chains should worry. The minute a major chain has a working conversational surface, the F&B inside that chain’s properties enjoys a permission-to-suggest that no neighborhood restaurant has. The chain’s chatbot can say “your usual sparkling water and a 7 p.m. table for two at the lobby restaurant?” with a one-tap confirmation. The chatbot for the independent across the street, if it exists at all, has to win every interaction from cold.

I cover this dynamic more directly in a forthcoming May piece on Marriott’s broader AI deployment timeline, where the F&B implications get more attention than today’s loyalty framing allows. The short version: every F&B operator inside a major chain should be asking their corporate counterpart, this week, how they fit into the ten use cases.

The Playbook — What to Copy

I promised to make this a playbook, so let me get concrete. If you run a chain, a soft-brand, or a large independent group, and you are reading this between flights, here is what I would lift from today’s event, in priority order.

One. Decide, this quarter, whether you are going to defend your direct channel with a conversational surface of your own, or whether you are going to cede the front door to a third-party agent. There is no third option that survives 2026. If you are going to defend, you need a project name, an owner, and a deadline visible on the CEO’s calendar. If you are going to cede, you need to negotiate the agent-protocol terms now, while you still have leverage.

Two. If you are defending, do not start with the chatbot. Start with the data perimeter. Decide what guest data can enter a model, under what consent regime, in what jurisdictions, and with what retention. The reason Marriott can ship a prototype in front of seventy-five top members is not that they have a brilliant AI team — they do, but that’s not the bottleneck. The reason is that they have spent two years building a data perimeter that legal and security will sign off on. That perimeter is the gating dependency for everything else.

Three. Audit your CRS, your PMS, and your loyalty stack against the question can a conversational surface actually take action against this system, on a per-guest basis, in under two seconds. If the answer is no, you have a platform project, not an AI project, and it is more urgent than your AI project. The Marriott six-property beta of next-generation CRS, PMS, and loyalty platforms is the dependency that the chatbot is sitting on top of.

Four. Pick your ten use cases. Resist the temptation to pick eleven. Resist the temptation to pick a moonshot. Pick five guest-facing and five operations-facing, give each one a single owner and a six-month deadline, and accept that anything not on the list does not exist for the next year. The discipline of cutting to ten is, in my experience, the difference between AI organizations that ship and AI organizations that present.

Five. Borrow Marriott’s launch motion. The next time you have a feature that touches your highest-value guests, do not put it on stage at a trade show. Fly seventy-five customers to your headquarters, put your CEO in the room, and let them touch the prototype. The press release will write itself, and the testers will become your distribution channel. The cost of seventy-five flights and two nights of hotel is smaller than a single Skift Forum sponsorship and approximately a thousand times more effective.

Six. Get loud about the privacy posture. The chains that will win the conversational era are not the chains with the most clever AI features. They are the chains whose guests trust them with the data needed to make the AI features good. That trust is built in the small print and in the events like today’s, where a chief customer officer can look a member in the eye and explain what the chatbot does and does not see. If your privacy posture is something your general counsel writes in isolation and ships as a PDF, you are not going to win this round.

Seven. Plan the agent-facing surface in parallel with the guest-facing surface. The conversational layer you build for your own app is the same logical layer — different protocol, same permissions, same data perimeter — as the one that, twelve months from now, you will expose to third-party agents under whatever protocol emerges as standard. The chains that have built one will, when the protocol settles, ship the other in a quarter. The chains that have built neither will be six quarters behind.

A Note on the Members Themselves

One last thing, because I cannot stop thinking about the Ambassador-tier member at the table in the lobby this morning. The disintermediation thesis only holds if guests, by 2027, prefer the convenience of a generic agent over the recognition of a branded loyalty surface. That is not a foregone conclusion. The seventy-five members in the room today are exactly the cohort least likely to defect to a generic agent — they have status, they have history, they have a chief customer officer who knows their name. They are also the cohort whose habits the rest of the loyalty base imitates over the following twenty-four months.

If Marriott can turn that cohort into evangelists for an in-app conversational surface — and the optics today suggest they can — then the muscle memory question I raised at the top of this column starts to resolve in the chain’s favor. The chatbot does not have to be better than a generic agent. It has to be the thing the guest reaches for first, on the surface they already trust, in the moment when the alternative is a phone call to a property. That is a winnable fight.

It is not a fight the chain can win by reading the press release. It is a fight the chain wins by copying the motion: the cloud-native platforms underneath, the ten prioritized use cases, the data perimeter, the high-value-member preview, the CEO in the room. The chatbot was on stage today. The playbook was the stage.

The window is twelve months. The directions are in plain sight. The only remaining variable is how many operators read today’s headline as news and how many read it as instructions.

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