Hotels are losing the retail margin before check-in

The room rate is still largely hotel-controlled. The retail conversation around that room is not.

That distinction matters because upgrades, breakfast, late checkout, and alternatives used to be sold when the hotel could see the guest, judge the inventory, and decide what was worth giving away. Now a Booking.com loyalty tier or an AI booking agent can put a personalized offer in front of the guest before the property sends its first message. The channel that frames the offer gets the acceptance data, learns the guest’s preferences, and increasingly has a claim on the margin.

The first material loss from AI-mediated distribution may not be a room booking or an OTA commission. It is the hotel losing the right to decide which guest gets which offer, at what price, and with what trade-off against operational cost. Multi-property operators should stop treating ancillary sales as a front-desk function. They need an offer ledger by channel, one that shows who made the offer, who set its price, what data was used, who fulfilled it, and how much contribution the hotel retained.

The room rate is controlled, but the offer is not

The central warning in Hotels Set the Room Rate, but Channels Now Decide Who Sees the Upsell is easy to dismiss because hotels can still load rates, manage restrictions, and close inventory. Those are real controls. But a guest does not experience a stay as a BAR rate alone. The guest experiences a proposed bundle: a room, a breakfast decision, a better view, a larger room, a late departure, and sometimes a service promise attached to the trip.

That bundle is where retailing happens.

In his Hospitality Net argument, Markus Busch points to Booking.com’s Genius Levels 2 and 3 accounting for more than half of room nights. The operational question is not whether Genius produces demand. It plainly reaches a large share of it. The question is what the channel can do with that relationship before the hotel speaks to the guest.

A channel with loyalty behavior, past booking history, search behavior, and payment context can decide that one traveler is a breakfast buyer while another is a late-checkout buyer. An AI booking agent with a richer spending history can make a still more specific recommendation. The hotel has the room type and the inventory. The intermediary has the moment of attention and a broader portrait of the buyer.

That changes the old division of labor. The front desk used to hold the final retail conversation because it was the first human point of contact and because the desk could see the available room inventory. Revenue management protected rate. Front office turned an arriving guest into a better stay. If the upgrade did not sell, the hotel still owned the decision to comp it for service recovery, release it for a walk-in, or protect it for a higher-value arrival.

When the offer is accepted before arrival through a channel, that sequence reverses. The hotel becomes the fulfiller of a promise it did not fully design. It may receive an upgrade request at a price that does not reflect that night’s displacement risk. It may need to deliver a breakfast inclusion with labor and food cost attached. It may learn that the guest wanted a late checkout only after the operating plan has already been set.

This is not an argument that every channel offer is bad for the hotel. It is an argument that an offer is a commercial right, not a decorative booking-path feature. If someone else controls it, the hotel should know what it is paying for that control.

Ancillary revenue has moved into the booking path

The old front-desk upsell was imperfect, but its economics were visible. An agent asked an arriving guest whether they wanted breakfast or a better room. The hotel retained the transaction record, knew the incremental revenue, and could generally see the direct fulfillment cost. A missed upsell was disappointing, but it did not become a third party’s permanent behavioral data.

Pre-arrival retailing works differently. The party that presents the offer chooses the framing. It can call breakfast convenient rather than optional. It can establish a reference price for an upgrade. It can describe late checkout as a benefit of loyalty rather than a scarce piece of inventory on a high-turn day. It also observes whether the guest accepted, declined, or ignored the offer.

Those observations matter beyond one booking. Acceptance data becomes the basis for the next recommendation. The channel gets better at deciding who should see which offer, while the hotel sees an already-made reservation and a limited set of guest details. The intermediary’s advantage compounds because each transaction produces another signal.

The earlier trajectory is clear. AI Bookings Just Bypassed the OTA Entirely, in China, at Scale described Fliggy’s 800% year-over-year surge in AI-mediated hotel bookings during Spring Festival 2026 after its booking skill was embedded in on-device AI assistants. The guest could book through voice or text without opening the Fliggy app. In the West, Google’s AI Mode Is Compressing the Hotel Research Session Down to One Exchange described a related shift: discovery, comparison, and purchase are moving toward one conversational exchange.

The commercial implication is bigger than a new booking surface. The old research journey gave the hotel several chances to re-enter the guest’s consideration set. A native agent flow can go from intent to booked room without a hotel website visit, an email inquiry, or a front-desk sales moment. Once that flow can sell the room, it can sell the stay around the room.

The party that owns the pre-arrival conversation does not just sell an extra. It learns what the guest will pay for before the hotel gets a chance to ask.

A revenue leader should therefore separate room contribution from stay contribution. A reservation sourced through a channel might carry an acceptable room commission and still be economically poor if the channel also controls the most profitable ancillary offers, owns the acceptance history, and leaves the hotel with the service obligation. Conversely, a channel offer that fills breakfast covers on an otherwise soft morning could be worth paying for. The point is to stop assuming the room commission captures the whole cost of distribution.

A direct API is an offer pipe, not a visibility project

The technical discussion around agentic distribution often gets reduced to visibility. Can an AI agent find the property? Can it return rates? Those are necessary questions, but they are not enough.

An agent cannot make a reliable offer from stale or partial information. To recommend a property in a live conversation, it needs availability, rate, room attributes, policies, and a path to complete the booking. To sell a meaningful bundle, it eventually needs a current answer on the things that make the bundle real: eligible room types, breakfast rules, late-checkout availability, and the terms under which those benefits can be honored.

That is why Les Roches’ Derchi: Hotels Without MCP and Direct APIs Face ‘Algorithmic Invisibility’ matters beyond direct booking share. Francesco Derchi’s argument is that hotels need Model Context Protocol connectivity and direct API architecture to respond instantly when an agent asks. If the property cannot respond, it does not appear. If it can return only a room rate while another party can return a complete package, it appears as a commodity room while the other party controls the retail layer.

A direct connection is not simply a distribution expense-reduction project. It is an offer pipe. It determines whether the hotel can put its own inventory logic and product rules into the agent’s answer, or whether an intermediary will interpret the property on its behalf.

The constraint is severe. Only 7% of Hotel PMS Vendors Publish Self-Serve API Docs, New Study Finds reported that only 24 of 343 PMS vendors publish fully self-serve API documentation. That does not prove that the other systems cannot connect. It does show how difficult it is for an operator to assess, test, and build a direct agent-facing connection without entering a gated vendor process.

This bottleneck shifts bargaining power toward parties already connected. An OTA or booking platform can offer immediate reach because it has normalized inventory across many hotels. A single independent property, or a management company running several different PMS instances, may have no practical way to return a live, governed answer without that intermediary. The agent may be willing to connect directly. The hotel’s stack may simply be unable to answer.

Ask the vendor a narrower question than, “Do you support AI?” Ask whether its current production endpoints can return live rate, availability, room attributes, and bookable ancillary eligibility to an external agent. Ask for the documentation, the approval process, the audit trail, and the limits on what an external party can do. A roadmap slide does not protect an offer right.

Most hotels cannot yet outperform the intermediary in their own inbox

There is an uncomfortable objection to the hotel-control argument: many hotels have not earned the right to claim they would convert or retail the demand better themselves.

Only 44% of Multi-Question Hotel Inquiries Get a Complete Answer provides the operational test. Lobby found that only 44% of multi-question inquiries receive a complete answer. It also found that 42% of declined requests receive no alternative booking offer, and nearly 12% of follow-up emails are guests asking again about questions they had already submitted.

That is a direct-channel failure before it is an AI failure. A guest asks whether breakfast is included, whether a larger room is available, whether parking works for their arrival, and whether late checkout is possible. A partial response creates another email. A refusal without an alternative sends the guest back to search. In both cases, the property had the demand in hand and failed to turn it into either a booking or a better booking.

An effective agent will look superior in this environment because it can answer a complete request in one interaction. It can propose a different room type if the first choice is unavailable. It can package breakfast. It can preserve momentum. That is not mysterious personalization. It is disciplined reservation selling performed at the moment the guest is ready to decide.

Before a hotel complains that an agent has taken the conversation, it should inspect its own. Pull twenty recent multi-question reservation inquiries. Check every question against the reply. Then pull twenty declined requests and check whether the team offered a different date, room type, rate, or property. This is not a customer-service audit alone. It reveals whether the direct channel is capable of making and completing offers with the demand it already attracts.

The practical opportunity is not to imitate generic agent language. It is to give reservations and front-office teams a complete view of offerable inventory, clear authority, and a measurement standard based on resolution and contribution. A faster incomplete reply is still an incomplete reply.

Intermediaries can create value, but that value needs a price

The strongest objection deserves to stand on its own. Booking platforms and AI agents can create ancillary revenue the hotel would not otherwise earn. They have loyalty audiences, behavioral data, payment context, and product-design capability that many independent hotels do not possess. For an independent without a meaningful first-party database or a capable direct booking path, a channel’s breakfast or upgrade offer can be genuinely incremental.

A blanket strategy of excluding intermediaries from pre-arrival offers would be self-defeating. It would deny the fact that some demand arrives only because a channel assembled, recommended, and converted it. It would also ask properties with weak systems to compete immediately with platforms built around personalization and transaction design.

But incrementality is not a reason to give unpriced control over every offer. It is a reason to measure the net contribution of each arrangement.

The test should include room commission, discounting, the ancillary price charged, the hotel’s retained ancillary margin, fulfillment cost, service recovery cost, and the data rights attached to the transaction. The data-rights point is not theoretical. The signal on channel control notes that Booking.com guest emails expire seven days after checkout and Expedia email aliases expire after 45 days, while phone numbers can arrive masked. A hotel that cannot retain a usable relationship after fulfillment has paid for more than demand. It has paid to rent access to the guest.

There is also an inventory-cost question. A late checkout sold at a modest price on a low-occupancy Sunday may be highly attractive. The same late checkout promised on a turn-heavy day can complicate housekeeping, delay room readiness, and create recovery costs that exceed the revenue. The intermediary sees acceptance. The hotel bears the operating consequence. The ledger must show both.

This argument would be wrong if channel-controlled offers consistently produced higher net ancillary contribution, no material operating friction, and usable guest-data rights compared with hotel-controlled offers. Some properties may find exactly that. The point is that they should discover it through contribution reporting, not accept it as a feature of distribution.

What cannot be measured will be conceded by default

Hotel organizations have already demonstrated the broader measurement problem. 91% of Hotel Chains Use AI, but Only 28% Have a Company-Wide Strategy and 13% Measure ROI found that 91% of chains use AI, while only 13% measure ROI. That gap is especially dangerous in distribution because the losses do not appear as one obvious new line item.

A revenue report will show occupancy, ADR, RevPAR, channel mix, and room commission. An operations report will show breakfast covers, late-checkout requests, housekeeping pressure, and perhaps upgrade revenue. If those reports do not connect the offer to the channel that originated it, management sees activity but not economics.

The offer ledger closes that gap. It does not need to be a new platform in the first quarter. It can begin as a disciplined reporting structure across a limited set of offers: paid upgrades, breakfast, and late checkout. For each channel, record offer exposure where available, acceptance, price, discount, commission or fee, direct fulfillment cost, recovery cost, retained contribution, and the guest data available to the hotel after checkout.

Then add control fields. Who set the offer price? Who chose the eligible guest? Could the hotel change the terms? Was the offer made before the hotel had a direct interaction? Did the hotel have a comparable direct offer available? These fields make the commercial trade-off visible.

This is where the second-order effect appears. Once an operator can show that one channel creates profitable breakfast demand but underprices late checkout on high-turn dates, it can negotiate specific rights and rules. It no longer has to argue abstractly about control. It can set blackout conditions, price floors, inventory caps, or channel-specific compensation. If it cannot measure the offer, it will concede those choices by default because the intermediary is the only party holding the acceptance data.

Regional operators should install an offer-rights audit this quarter

A regional VP of operations or management-company revenue leader should treat this as a 90-day commercial audit, not an AI transformation program.

Start by inventorying every pre-arrival offer that a guest can see through every meaningful channel. Include OTA booking paths, loyalty programs, metasearch and agent flows, confirmation emails, pre-arrival messages, call-center scripts, and the hotel’s own booking engine. For each offer, identify the seller, the price setter, the inventory owner, the guest-data recipient, and the property team responsible for delivery.

Next, put three offers into a channel-by-channel contribution report: room upgrade, breakfast, and late checkout. These are concrete enough to audit and common enough to expose the problem. Do not report only total ancillary revenue. Report what was offered, what was accepted, what it cost to fulfill, and what margin remained after channel economics.

At the same time, run the direct inquiry test. Audit multi-question inquiries for complete answers and declined requests for alternatives. The property cannot credibly demand more control of the guest conversation while failing to finish the conversations it already owns.

Finally, require PMS and CRS vendors to document their live external endpoints. The question is not whether they have an AI strategy. It is whether a governed outside agent can retrieve a live answer today, what that answer contains, and what the hotel can control in it. Where a direct connection is unavailable, name the dependency clearly. That is an input to distribution strategy and vendor renewal, not an inconvenient technical detail.

The aim is not to eliminate intermediaries. It is to decide deliberately which offer rights are worth paying for. Over the next quarter, watch one signal above all: whether agent and channel bookings begin arriving with ancillary promises that the property cannot price, alter, or trace back to contribution. When that starts happening at scale, the hotel is no longer just buying demand. It is renting out its retail floor before the guest reaches the lobby.