I do want to caveat this by saying that at such an early stage, the future is a wide distribution on a bell-curve. LLMs and AI could have any number of impacts on travel distribution but I wanted to lay out where things stand and hypothesize on the impact of AI on travel distribution.
This post was spurred by comments that
made on the All-In Podcast/Angel Summit last week around Sabre and travel that you can see here:
Chamath agrees with Palo Alto CEO Nikesh Arora that there’s $300b of travel public market cap that is due to be disrupted and that Sabre ($1.3B EV) is likely to be a winner as it controls the data as opposed to the middle-men being Expedia and Booking.com (I think this is directionally right but completely wrong on Sabre).
On the opposite end of the spectrum, Skift estimates that AI could drive a $8b to $28b positive impact within travel whether it comes out of travel planning, inspiration, operational efficiencies, or customer service chatbots.
While LLMs and AI such as ChatGPT have become commonplace and top of mind, the last decade of travel distribution has been a knife fight. You’ve had American and Sabre suing each other, Lufthansa charging incremental fees for GDS bookings, and a general battle of who collects what fees when and where between the traveler and the service provider.
The most important thing for travel distribution, is how to provide travelers with exactly what they want at the lowest price, and the advancement of LLMs means travel distribution is ripe for disruption and the most likely outcome is that all the middle-men are losers fighting over a smaller and smaller pie.
I want to set the stage with how big global travel demand is. IATA estimates that global airline revenue is $800B+, while IBIS estimates global hotel demand is $1.5T for a total of $2.3T annually in 2023 (lets exclude cruise, car rental, experiences and dog walking for the sake of simplicity).
The more important question is how is travel booked and where. Generally 30-40% of bookings are direct (consumer goes to United.com or Hilton.com) vs 60-70% is through a GDS (whether it’s an OTA or TMC like Amex Travel or Concur)
Generally I divide travel info a few different layers:
- Service Providers: Airlines, Hotels, Car Rentals, Cruise, STR owners, those who actually run the physical assets and deliver the real value to the consumer.
- Discovery: Expedia, Booking, Airbnb, American Express Travel, Concur, the players who “help” consumers discover where, when, and how to book their travel.
- Pipes: Sabre, Travelport, Amadeus, Duffel, those who help facilitate the connection between the service provider and the discovery engines.
- Metasearch: Tripadvisor, Kayak, Trivago, those who help consumers take advantage of price discrepancy between the various discovery engines and service providers.
From a distribution standpoint for the consumer it generally looks like this:
This is tremendously simplified and often Metasearch has direct connects to the service provider, but I wanted to flip things on it’s head a bit, usually charts like these ignore the travel consumer. I wanted to do a simple chart showing the consumer perspective of “I want to book travel” and how that flows. Let’s assume I want to fly on United and book a Hyatt hotel. I have three options:
- Go directly to United.com and Hyatt.com and book my choices.
- Go to an OTA/TMC and search for travel, book my hotel+flight through Booking.com or Amex Travel or Concur.
- Look at Trivago or Kayak and find the cheapest (sometimes sketchiest) option, book them either through an OTA or the service provider who sends the metasearch engine referral revenue.
From a consumer perspective, Sabre, Travelport, Amadeus (the GDS’), never really show up at all. I don’t even know they exist because they’re pipes. GDS only really exist because of historical technological failures to adapt.
Back in 1985, the airline industry centralized around using EDIFACT through Global Distribution Systems (GDS). It was revolutionary at the time, but even today the majority of travel bookings are executed using this system:
EDIFACT example for FRA-JFK-MIA Ticket
It’s a complete technological dinosaur and has continued to exist because generally the hotel and airline industries have never had enough FCF to invest in their IT systems over the last 40 years. The downside of this system is that it’s closed and it only enables the selling of tickets as opposed to ancillaries, miles, upgrades, etc.
IATA has focused on using NDC which is an XML based standard that allows all the same information as EDIFACT plus things like selling seat upgrades, lounge access, baggage sales, and even individual seat selection. Despite NDC having been around for over a decade, even Sabre admits that low single digits of airline bookings are done using NDC:
“NDC as a percentage of intermediary airline distribution, is low single digits overall for the industry as well as for Sabre. As we look forward to 2025, we certainly do expect that NDC will gain additional traction and we have NDC maturing as an expectation within our '25 target profile”
So the simple reality is that airline, hotel, and general travel distribution is mostly stuck with 1980s distribution technology ripe for disruption.
The reality is that the architecture of the web with consumers going to Google, typing in “best hotel in Sardinia”, finding a blog with a referral link to Expedia, making a booking, then showing up at the property with an Expedia confirmation print-out is going to change dramatically.
My assessment is that natural language chatbots will win out for travel discovery. Whether it’s going to WhatsApp, a chatbot by Booking, or iMessage, it’s likely that this will be the future of booking travel.
WhatsApp with the Four Seasons Ko Olina
For most of my personal travel today after I book, I use WhatsApp to contact the concierge or property owner to discuss arrival, amenities, activities, and departure. It feels very natural to use conversational chat for this, so why wouldn’t it be the same for discovery and booking? A smart AI/LLM chatbot on WhatApp is perfectly capable given the relevant information of knowing exactly the hotel or flight my family and I are looking for and booking that for us.
More importantly, a personalized chatbot for travel can know what brands I prefer, the room-type, aisle or window, first class or economy, upper floor or lower floor, near the elevator or not, and automagically make all of that happen on the back end. It’s a superior consumer experience vs what we have today.
The winners for travel distribution in the age of AI will clearly be the actual service providers. Hotels and airlines especially have had a contentious relationship with GDS and historical distribution standards for 20+ years as they look to increase margins and own their customers.
With any disruptive technology the winners are often those who provide the actual service or reach new customers. American Airlines is likely beyond excited for the idea of LLM bookings as opposed to the knife fight relationship they’ve had with Sabre.
At the top of funnel for discovery it’ll be those AI/LLM companies who can provide the best ease of discovery. Whether it’s ChatGPT, WhatsApp, or iMessage (with Siri), all of the 10 blue links that Google provides will be disintermediated by language models and natural chat. It’s too early to call a winner here, but it’s where I see things going.
To be clear, with ~50% of travel bookings still done through a GDS, it’s difficult for 100% of relevant information to be available to an LLM without GDS partnerships.
My sense is that the service providers (who actually provide the consumer value), will aggressively push towards the adoption of NDC (for airlines) or other direct connects (for hotels). United Airlines a couple months ago finalized an NDC deal with Sabre that allows among other things access to all fares, seat selection, and booking servicing through the GDS. If I was Scott Kirby I would today give API/NDC access to LLMs and see what the market creates.
After all, the end goal for service providers is to maximize exposure, price, and bookings to their product. Let the market dynamics play out and see where consumers prefer to book whether it’s United.com, Expedia, a TMC, or using ChatGPT/Whatapps/Character AI.
In the absence of NDC/Direct Connect I would guess that the OTAs come out as relative winners in this. Certainly margins will see compression from disruption, but there is a real value to what OTAs provide in terms of unique hotel/STR supply and customer service.
If I was running WhatsApp I would let loose an LLM with partnerships from an OTA and any airline/hotel that wants to give me API access and see what happens. It’s the natural language way for consumers to book and cuts through most of the middlemen.
The best case scenario for LLMs (and worst for the incumbent travel intermediaries) is that LLMs advance to the point that they can make and edit any travel booking for you on any website. If the LLM can go to a random bed and breakfast on the Maine coast and navigate their custom html booking engine for you then it’s almost game over for the rest of the industry.
In an earlier post I mentioned how Expedia has a plug-in for ChatGPT which was a decent first push for them into AI. But the reality is that the plug-ins are clunky, not natural, and fairly useless. It’s a band-aid, but for OpenAI, I would expect travel bookings to be an integral part of the business that is cohesive into the product. Expedia can’t win with a plug-in, they have to be a core part of the ecosystem and compete on brand and service-quality.
Additionally despite the potential for WhatsApp to be a winner, image based networks like Instagram are likely to weaken. When discovery is offloaded to LLMs the importance of travel discovery on Instagram (or TikTok) diminishes.
10 Blue Links like the results on Google are structurally challenged. Why would I go to Google and ask for the best hotels and flights to Tuscany when I can ask my personalized AI chatbot?
Metasearch will be completely made irrelevant by LLMs. Why on earth would you go to Google, click on a Trivago link, and then compare pricing for Booking, Expedia, and the hotel itself when all of that information will be available and processed through your personal AI booking assistant?
TripAdvisor is a big question mark. They’ve tried and failed multiple times with becoming a full-fledged OTA, but in the age of AI information is king and TripAdvisor has decades of user review data that is important.
NDC Aggregators like Duffel and TravelNDC were cutting edge startups that aggregated NDC data that is now easily replicated and connected to by an LLM.
Sabre/Travelport/Amadeus as middlemen are completely screwed. They do have SaaS businesses where they run the inventory/revenue management for hotels and airlines, but taking one look at the Sabre revenue breakdown shows how at-risk the business is:
Sabre Q1 2023 Revenue breakdown
If I was running Sabre, Amadeus, or Travelport I would immediately attempt to put all of my inventory into OpenAI or whatever Meta is working at with WhatsApp. Your biggest hope as an operator of a GDS is that there is still room for a middleman aggregator in the ecosystem. If the back-end remains as convoluted as today and solutions to simplify into APIs for every hotel on the globe don’t develop, then GDS theoretically have value.
Given the 50% of bookings that are still using 1980s technology, a GDS could argue that they are the only true holders of the world’s travel demand data and that is justification enough for them to exist. The race is on whether more open infrastructure like NDC disrupts faster than the GDS can get their heads on straight in regards to the changing travel distribution ecosystem.
Despite the concerns over margin compression (does Expedia really deserve 20% of the GBV for a hotel booking?), I do believe there is real value to what the OTAs provide.
Let’s imagine booking a flight and hotel through your personalized WhatsApp concierge. The discovery, booking, and payments are all fairly straightforward. But what happens if you need to change your reservation, or you have a death in the family, covid, or any number of things that you would usually call a customer service line for? Certainly your WhatsApp customer service agent could help, but customization and personalization when it comes to customer service is something that can be better provided by an OTA.
Additionally the unique inventory that OTAs like Booking, Expedia, and Airbnb have on the vacation rental/STR side is extremely valuable. That exists outside of the traditional GDS coverage and is something that the individual brand owns.
My sense is that a decade from now Booking and Expedia will exist, but they will transform from a discovery and booking engine to an aggregator/fin-tech/customer service engine. You’ll book your travel with your WhatsApp or ChatGPT agent and specify you want it through Booking or Expedia (so brand still matters), and if you have any issues your personalized agent will reach out to the agent at EXPE to resolve the issue.
There is still room in an AI/LLM travel world for loyalty, brand, and customer service, but the traditional OTAs will need to evolve to fit the consumer demand.
- Watch Video: Breaking Down the Opportunities and Challenges of Adopting AI in Travel
- Airline Profitability Outlook Strengthens
- Global Hotels & Resorts - Market Size 2005-2029
- Lufthansa Group Shares the Details of its Continuous Pricing Solution
- American Airlines is Pleased with its NDC Rollout
- Sabre Q1 2023 Earnings Transcript
This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.