The 30% Distribution Tax: Market Power in Agentic Commerce

How we book travel is about to change fundamentally. We have gone from human travel agents in the pre-internet era to OTAs in the search era to AI agents booking in the years ahead. The question on any hotel owner’s mind is how agent bookings will affect total distribution costs. To lament the dominance of OTAs is one thing, but what if distribution costs rise in the AI era? What if the platform take rate ends up at 30% by 2030? After all, that is the going rate in the most successful digital marketplace ever built.

Distribution has never been free, and its price has climbed with each new intermediary. In the analog era, the commission was about 10%, the standard rate an agent earned on a hotel booking or an airline ticket before the mid-1990s (GAO, 2003). When online travel agencies replaced traditional travel agents, the rate rose to 15%-25%, and a single company, Booking Holdings, held roughly 71% of the European OTA market by 2021 (HOTREC, 2022). The broad pattern is disturbing: the closer an intermediary gets to demand formation, the greater its pricing power.

Each is captured more by sitting closer to the moment of decision, from the human agent who advised, to the platform that ranked, to the software that will soon choose. Agents are likely to win a meaningful share of discovery, especially for transient travelers. A traveler today juggles OTAs, brand sites, metasearch, loyalty schemes, and review platforms to plan a single trip.

A consumer AI agent, like ChatGPT or Claude, knows its user’s preferences, compares the entire journey, and books it end-to-end with almost no usual friction. Whoever owns discovery owns the economics. The agent becomes the point of first contact for the whole trip, and that is where the tax will be collected, one way or another.

Booking a trip carries friction in several forms: finding that an option exists, comparing options, entering your details, paying, and trusting that the room will match its photos. Every intermediary in the history of travel has earned its margin by taking some of that friction off the traveler, and an AI agent is the first that can take care of all of it in one pass. That is what makes it genuinely useful. The unsettled question is who keeps that value: the traveler, the hotel, or the agent provider that now sits where the friction used to be.

To see where the rate could land, let us look at the mobile app economy. Apple’s App Store has taken up to 30% of in-app purchases since it opened in 2008. The fee was held for over a decade for one reason: a developer who wanted to reach an iPhone user had no second door. One storefront, one payment system, one set of rules, and the platform set the price.

Regulators and courts in the United States only began forcing that door open in 2025, after years of litigation and a market that had long since organized itself around 30% (Reuters, 2025). Apple later cut fees for some categories and smaller developers, but 30% remained the headline rate. That is how a gatekeeper behaves. It can afford to be generous at the margins while holding the line where the money actually is. The App Store is not a perfect parallel, as online travel is its own market. The point of the analogy is that whenever a new way of interacting on the internet emerges, the actor who controls the front door tends to exploit the market power conferred by it, much as economic theory predicts (Rochet and Tirole, 2003; Edelman and Wright, 2015).

A consumer AI provider with full booking capability could become the default way people plan and book and would sit at that chokepoint, owning discovery, comparison, and the transaction in one place. A hotel that wanted to be recommended would be back to buying placement, now from a single gatekeeper rather than several competing ones. The agent could set its take rate as Apple does in the App Store, and a hotel would have little leverage to refuse, because refusing would mean being invisible in the one channel where guests now look. The fee can surface as sponsored rankings, preferred-agent access, transaction processing, conversion optimization, or bundled placement. The AI advertising market is only now forming, so the shape the charge takes is unclear, but the temptation to maximize the value extracted is certainly there.

The strategic question is less about whether the agent works than who controls the interface once it does. An intermediary that controls access between two sides of a market, travelers on one side and hotels on the other, can price that access at a level set by its dominance rather than by its costs. The more of the traveler’s daily life the agent already owns, through the device in their hand and the payment behind it, the harder it becomes for a rival to undercut it, and the more the charge starts to behave like a private tax rather than a competitive fee. Whether we get a 10% world or a 30% world depends on how concentrated the agent layer becomes. Concentration is also the default outcome rather than the exception, because the same network effects that make one agent useful enough to adopt make it expensive for a competitor to pull travelers away once they have settled in.

The force pushing in the other direction is competition, and it is the only thing that drives cheaper distribution. The price of a service falls toward its cost of production when several sellers compete for the same buyer. For agents, three things have to be true at once.

1. Travelers have to use more than one agent, or be ready to switch, so that no single agent is anyone’s only route to a booking.

2. Switching has to stay cheap so a traveler can move to a rival without losing their history, preferences, or earned benefits.

3. New agents must be able to enter, which depends on securing hotel supply on the same terms as the incumbent.

Where all three hold, a handful of agents will push the fee down toward the cost of a credit card fee, and an agent asking 30% simply loses the traveler to one asking 5%.

Take any of the three away, and the market tips toward a single winner, free to set its own fee. The technology is the same in both futures; the price is set by the market structure we let form around it. It is inevitable. The 30% that Apple kept was the product of a particular market structure, and that structure is being unwound in court precisely because it was a choice rather than a law of nature. The foundations of agentic commerce are still being built, so its terms remain open. A handful of decisions taken in the next few years will determine whether hospitality inherits the 10% world, perhaps through a credit card fee, or the 30% one.

Hotel owners and brands hold the first lever. The mistake of the search era was to surrender both the inventory and the guest relationship to the OTAs and then complain about the cut they ended up taking, once well established. The discipline this time is to connect to many agents, on terms the hotel can read and reproduce, and to keep the guest record under the property’s control rather than the platform’s. A hotel reachable by every agent has more leverage than one dependent on a single closed pipe.

Technology vendors hold the second lever. Whether agentic booking runs on open, shared standards or on closed, proprietary pipes is being settled now through engineering choices that look mundane but are anything but. A common, public way for any agent to read a hotel’s rates and availability is the difference between a competitive market of agents and a single gatekeeper. The industry has the leverage to insist on that openness while the layer is still forming, and very little leverage to claw it back once it has set.

Governments and industry associations hold the third lever. Europe already treats Booking as a gatekeeper subject to rules on rate parity and self-preferencing. The same posture, applied early to emerging consumer AI providers, would avoid the worst-case scenario. The associations that fought OTA parity clauses for a decade already know this fight. It is the same one, arriving sooner and with higher stakes.

For an individual hotel, the practical work starts before any of that is settled. Embrace direct, real-time connectivity so that any agent can transact without a toll-taker in the middle. Keep the property legible to a machine and treat dependence on a single agent as the risk it plainly is, the same risk a generation of operators learned to regret one channel too late.

The AI travel agent is coming, and for the guest, it will be a relief, one capable helper in place of a dozen open tabs. For hotels, the technology opens new paths to be discovered. Instead of the static, preset search filters of the OTA era, an agent can match a property to the actual intent behind a trip, favoring operators who keep their data clean and up to date. Whether the total distribution fee ends up in single digits or closer to 30% remains unsettled, and it depends on the choices owners, vendors, and other stakeholders make over the next few years. The last time the industry waited to find out, it learned the fee's magnitude only after it had lost the leverage to change it. The cost of acting early is a few uncomfortable conversations about standards and dependence. The cost of acting late is a recurring distribution tax embedded in every future P&L.

References

Edelman, B. and Wright, J. (2015) Price Coherence and Excessive Intermediation. Quarterly Journal of Economics, 130(3), pp. 1283–1328.

GAO (2003) Airline Ticketing: Impact of Changes in the Airline Ticket Distribution Industry. GAO-03-749. Washington, DC: US General Accounting Office. Available at: https://www.gao.gov/products/gao-03-749 (accessed 31 May 2026).

HOTREC (2022) European Hotel Distribution Study 2022. Brussels: HOTREC. Available at: https://www.hotrec.eu/ (accessed 31 May 2026).

Reuters (2025) Apple ordered to stop charging commission on App Store external purchases. Available at: https://www.reuters.com/ (accessed 31 May 2026).

Rochet, J.-C. and Tirole, J. (2003) Platform Competition in Two-Sided Markets. Journal of the European Economic Association, 1(4), pp. 990–1029.