Enterprise AI is becoming a sales channel, not just a feature. Radisson's ChatGPT plugin shows what that takes
Radisson and Accenture built a ChatGPT plugin for hotel discovery that hands bookings to Radisson's own site. The reusable part is the MCP and API layer behind it, not the chat.
Listen to this article · 4 min
AI-generated narration of the full article.
About 1.5 times the conversion of organic search. That is what Radisson Hotel Group reports for travelers who reached its booking flow through ChatGPT in July and August 2026, according to a case study OpenAI published on October 7. The number is early and company-reported, but it points at something larger than one hotel group’s marketing result: generative AI moving from inside the product to in front of it, as a channel where customers find the product and start buying it.
The business problem
Hotel groups have spent two decades fighting to bring bookings back to their own channels instead of paying intermediaries. A new discovery surface where travelers plan trips in conversation reopens that fight. If the assistant recommends hotels without the brand’s live prices and availability, someone else’s data shapes the answer.
What Radisson built
Radisson worked with Accenture Song on a ChatGPT plugin for finding and comparing hotels while planning a trip. Travelers see suggestions on a map, compare prices and amenities inside ChatGPT, and then continue to Radisson’s own site to complete the booking. The checkout does not happen in the chat.
Behind the plugin, Accenture built an MCP server and APIs that also power Radisson’s sponsored advertising in ChatGPT, which Radisson started running in July, and that are meant to support future features. Accenture says the plugin took about six weeks to build. Beyond the conversion figure, the companies report that more than half of recorded checkout and booking events were attributed to ad views through view-through measurement.
What the company owns, and what it rents
The useful way to read the case is to separate the layers by who controls them.
| Who controls it | What to watch | |
|---|---|---|
| Discovery in the assistant | The AI platform | Ranking, display rules, how the model describes the offer |
| MCP server and APIs | The company | Current prices and availability, reuse across plugin and ads |
| Checkout and payment | The company | Terms, consent, loyalty |
| Measurement | Shared | View-through attribution versus holdout tests |
The capability layer is the durable investment. The same MCP server serves the plugin and the ads, and will serve whatever comes next, which is the logic we described in building once and reusing across agents. It is also where the hard work lives: prices and availability in a conversational channel must be as current as on the website, and a plugin that quotes yesterday’s rate creates a complaint, and in some markets a consumer protection problem. That is the same challenge as connecting AI-built apps to live business data.
Keeping the transaction on the company’s side keeps payment, loyalty, terms and the customer record in its own systems. The assistant brings the customer to the door; the company still runs the counter.
The transferable lesson, with caveats
The rented layers deserve skepticism. View-through measurement credits an ad that was seen, not necessarily clicked: a useful signal, weak proof of incremental revenue, so budget shifts should wait for holdout tests. Two summer months with an engaged early audience may not hold as usage broadens. And discovery inside someone else’s platform depends on that platform’s choices: Radisson notes that users once had to type the plugin’s name before a query, until a ChatGPT update removed the need.
For any company with products a customer could discover in a conversation, the order of work follows from the table. Build the capability layer as an API or MCP server you own, keep checkout and consent on your own channel, define measurement before launch, and review what the assistant says about your offer as you would review a reseller. The chat interface is the part that will change most often.