Navan merged 2.6 million properties into one catalogue. Your corporate rate now has to survive being matched to the right room.
Navan says its platform carries 2.6 million properties drawn from Sabre and Amadeus, from Expedia, Booking.com and MakeMyTrip, and from direct connections with hotel brands. The same room arrived through several of those feeds with conflicting descriptions, and travellers were reaching properties to find rooms that looked nothing like the listing. Navan's fix is an AI mapper that merges them into one verified entry, and room rate options rose 70%.

Navan described the problem and its fix in its own announcement of 12 March 2026, filed through its investor relations channel now that the company is listed. The corporate booking platform carries 2.6 million properties, supplied from Sabre and Amadeus on the global distribution system side, from Expedia, Booking.com and MakeMyTrip on the agency side, and from direct connections with brands including Wyndham, Travelodge UK and Premier Inn. Because the same room reaches the platform through several of those routes with different descriptions, travellers could not tell whether two listings were the same room, and some arrived at a property to find a room that looked, in the company's words, nothing like the photos or description. Navan built what it calls a room intelligence mapper, an AI tool it describes as a universal translator turning multiple messy descriptions into one clear verified listing. Dane Molter, the company's senior vice president of travel marketplace, said travellers should not have to play detective to work out whether two room listings are the same, and that using large language models as a translator had removed the duplicates cluttering search results and improved how rates get matched to rooms. The release carries one number, in its subheading: room rate options up 70%.
This piece is for independent hoteliers carrying corporate business through an agency or a global distribution system. The commentary circulating says an AI is rewriting your marketing copy and stripping your differentiators before a corporate traveller reads it. The published account says something narrower and more consequential: an AI is deciding which of your several listings is you.
The layer is a deduplication problem, not a copywriting one
A hotel loading a corporate rate does it once, into one system, against one room type. What arrives at the corporate booking tool is that room several times over, because the property also sits in a merchant feed, an agency feed and possibly a brand direct connect, each carrying its own room naming, its own photo set and its own description of the same physical bed.
The mapper exists because a traveller facing four versions of a king deluxe cannot tell which one carries the negotiated rate. Merging them is plainly good for the traveller and for the property that wants its corporate rate found. The risk is not that the merge loses your adjectives. It is that the merge is a judgement, made by a model, about which of several descriptions is authoritative, and about whether two rooms with different names are the same room.
Get that judgement wrong and a guest books the merged listing believing they have a room the property maps differently. Navan's own account of the pre-existing problem, a traveller arriving to a room that matched nothing they had read, is the failure mode this is meant to solve and the one a bad match reproduces.
Booking.com is a supplier inside the managed corporate channel
The supply list carries a fact the trade has not absorbed. Booking.com sits inside the corporate booking tool as a named content source, alongside the global distribution system feed.
The managed corporate channel has been treated for two decades as the one place where a hotel's negotiated terms hold and the merchant intermediaries do not reach. That distinction is gone at the content layer. A corporate traveller looking at a property may be looking at an agency rate, a merchant rate or the negotiated rate, and the room they are comparing across those options has been reconciled by the platform rather than by the hotel.
For an operator, the practical question is no longer only what rate you loaded and against which room type. It is whether the room you mapped it to is recognisable enough across every feed you appear in that a mapper joins them correctly rather than splitting your corporate rate off into an orphaned listing.
The ranking is a setting, and in one case the content supplier controls it
The other correction to the circulating account concerns ordering, and the documentation holds a detail the commentary has missed entirely.
SAP's own training material for Concur Travel describes a Hotel Sort Default field that lets an administrator choose the opening sort from Preference, Price, Rating, Distance, Policy or Custom, with Custom combining up to three criteria. Preference orders by the company's own designations, from most preferred to non-preferred. That much is a procurement decision made inside your corporate account, and it is knowable by asking the travel manager rather than inferred from a search result.
Then there is the provider-controlled sort. SAP documents a feature under which a hotel content supplier can dictate the order of hotel search results and rates, agreed between SAP Concur, the customer and the content provider. A traveller can change the sort during booking, and selecting Default restores the supplier's ordering.
Read that carefully. There is a configuration in which the order your property appears in, on your own corporate account's screen, is set by a content supplier under a three-way agreement to which the hotel is not a party. Nobody publishes which customers run it or which providers hold it.
The agent-to-agent argument is a forecast, and it should be labelled as one
The second half of the commentary holds that artificial-intelligence agents will soon negotiate directly between a corporate platform and a hotel's systems, cutting out the global distribution system and the online travel agency along with their fees.
No major property management system vendor has publicly launched agent-to-agent booking as a standard feature. What exists is adjacent and real: Navan's assistant already contacts hotels ahead of arrival to confirm payment details and hold rooms for delayed travellers, which is a machine talking to a property about a reservation rather than negotiating one. Amadeus, meanwhile, closed its self-service developer interfaces this year, which points away from open agent access rather than toward it.
Amadeus disputes the reading of its own move, and the objection is on the record. Its published statement responding to the February coverage says its Enterprise portal and Enterprise interfaces remain fully available, that its Ventures and Partners programmes let startups build on the same technology, and that it maintains an open-source programme. What it does not say anywhere is that the self-service portal is closing. The confirmation of that sits on Amadeus's own developer site, which now states the self-service portal was decommissioned on 17 July 2026, and on its developer code repositories, archived on the same date.
The economics are attractive enough that it will be tested. That is a reasonable expectation and it is our reading, not a measured fact, and a piece that presents it as imminent is selling a roadmap nobody has published.
What the seller can put on screen beside you is mostly not commissionable
Amadeus publishes the composition of its own hotel platform, and the split is the clearest available picture of what a corporate rate is sitting next to.
AMADEUS HOTELS, CONTENT AVAILABLE TO TRAVEL SELLERS
├── Commissionable content: more than 120,000 hotels worldwide
├── Amadeus Value Hotels: 740,000 properties at net rates
├── Aggregator partners: 11 or more, for business travel alone
└── Plus each seller's own private agreements
Source: Amadeus, hotel content for travel sellers, amadeus.com
Set the two disclosed property counts against each other and the commissionable tier is roughly one in eight of the two combined: 120,000 divided by 860,000 is 14%. That is not a share of everything a seller can book, because the aggregator content and private agreements are not counted in either figure and Amadeus does not publish a total. It is enough to establish the shape. The channel a hotel negotiated a commission into is the smaller of the two tiers Amadeus names, and the larger one is priced by somebody else at a net rate.
What the disclosed record cannot settle
Nothing published says how often the mapper gets a match wrong, and neither Navan nor Concur reports a match accuracy rate. The 70% increase in room rate options is the company's figure, produced to describe the success of its own product, with no stated baseline or period.
Operators are responding on the one variable they hold. Some are auditing their own room naming across every feed they appear in, checking that the room type carrying the corporate rate uses consistent naming, bed configuration and photography wherever it is published, on the reasoning that a mapper can only match what looks like a match. Others are going at it from the client side, asking their largest corporate accounts what the default sort is set to and whether the negotiated rate displays, which produces an answer about one account rather than a general rule but is the only route that produces an answer at all.
The test worth running is cheap and nobody has published it. Ask a corporate client to run a search on your city and send the screen. Compare the room names, the photographs and the rate against what you loaded. That is the screen your biggest account books you from, and most hoteliers have never seen it, in the same way most have never seen where their net rate travels after they hand it over or what the merchant shift did to the timing of their money.

Guneet Lamba does content and SEO at PriceLabs, where she writes about dynamic pricing, revenue management, and how operators actually run their portfolios. Her work appears across the PriceLabs blog and Rental Scale-Up.


