Report Overview
Recent breakthroughs in Generative AI have had an electrifying effect on the tech world in recent weeks. Many have compared these first artificial intelligence tools, which create new content, such as images or text from simple prompts, to the early days of smartphones or the internet. With the potential for such wholesale disruption, what does generative AI mean for the travel industry? And how do we separate the hype around this new tech from the reality of what it will look like in use by travel companies.
Skift Research’s latest report will attempt to tackle those questions by diving into Generative AI’s impact on the travel industry. This report will focus specifically on large language models and their use cases within travel business.
We examine seven different cases for how AI will be used across travel, including its implications for chatbots, operational efficiency, and search. Across most of these scenarios, we see multi-billion dollar markets being created. Skift Research estimates that altogether generative AI poses a $28 billion-plus opportunity for the travel industry.
What You'll Learn From This Report
- How Generative AI is being implemented in the travel sector today
- Market estimates for the impact of generative AI in Travel
- Four baseline areas where AI will have an immediate impact on travel
- Three opportunities where AI stands to drive disruption in the long-term
Executives Interviewed
- Rathi Murthy, Expedia Group’s Chief Technology Officer
- Shane O’Flaherty, Global Director of Travel, Transportation, & Hospitality at Microsoft
- Luca Zambello, the CEO of Jurny, a short-term rental tech platform
- Richard Valtr, the Founder of Mews, a hotel property management system
- Erik Tengen, the Co-Founder and CEO of Oaky, a hotel upselling software company
Executive Summary
Generative AI and large language models (LLMs) stand to have a significant impact on the travel industry. We see at least four significant use cases for generative AI and LLMs in travel.
- Customer Service Chatbots: LLMs can significantly improve the chatbot experience and make it more useful for customers. We estimate this could be a $1.9B market.
- Developer Efficiencies: Generative AIs can help programmers write better code, faster. This leads to faster development cycles and more new tech. We estimate that this could present $350M in annual savings.
- Reputation Management: AIs can help evaluate customer sentiment and allow travel businesses to respond to online reviews, boosting their online reputation. We estimate that this could be a $1.3B market.
- Performance Advertising: One of the key strengths of generative AI is the ability to search for, summarize, and present information in an accessible way. This stands to change the travel planning searches are made and with that will command a significant share of performance advertising dollars. We estimate this could be a $5B market.
Our baseline is for ~$8.4 billion in AI value created in the travel industry. But there will also be additional, and potentially even more impactful, ways that AI will transform the travel industry but we have less visibility into how these second order impacts will play out.
For instance, AI stands to disrupt not just travel planning searches – and with that performance advertising – but it may also upend searches for travel inspiration as well. With that would come changes in how all sales & marketing throughout travel is done. Or take the case of operational efficiency. These need not be limited to only developer tools, and AIs may set the stage for broad-based productivity gains across millions of travel industry jobs. Another scenario we envision is that AI assistants may change how complex technology stacks are integrated.
These additional categories where we are ‘dreaming big’ about the impact of AI on travel are harder to estimate with any true precision. But we believe they could potentially represent a $20 billion opportunity. Paired with our baseline market estimates, we see artificial intelligence driving a $28 billion impact within travel.
An Introduction to Generative AI
This section serves as an overview of artificial intelligence and generative AI. Readers already familiar with general AI concepts can skip ahead to the following sections.
It was the summer of 1956 when a young mathematics professor at Dartmouth decided to hold a modest workshop to discuss the potential for computers to develop as thinking machines. He got ten other colleagues to attend for eight weeks of discussion and research into this new field of study which they dubbed “artificial intelligence.”
These eleven scientists inadvertently launched the modern field of AI and would go on to be pioneers in the space. Yet by the mid-1970s interest in general computer intelligence waned as computer science shifted to focus on more specialized use cases. But the last few decades have seen a major resurgence in artificial intelligence driven by new statistical techniques and breakthroughs in computing power.
Today machine learning (ML), an AI technique, is being regularly used at major corporations across the world. These existing tools provide a powerful means of understanding and categorizing new types of data and very large datasets. These AI tools underpin modern personalization engines that we regularly experience when, for instance, Netflix suggests a show it thinks we will enjoy next.
Despite its wide usage, machine learning still remains opaque to the average consumer, it is the realm of specialized engineers. ML will often be trained on a specially prepared dataset and can only generate insights on inputs that conform to the preset parameters of those training sets.
But in recent months a new form of AI has taken the main stage. Called generative AI, this is a type of artificial intelligence system that is designed to create or generate new content, such as images, text, or music, that is similar to the type of content it has been trained on. Unlike other types of AI, which are focused on classification or prediction, generative AI is focused on creating something new. For example, Dall-E and other similar programs allow users to generate pictures from simple text prompts, no artistic skills required.
A common use for of generative AI is in large language models (LLMs). A large language model is a type of AI model designed to process and generate human language. They can perform a variety of tasks such as language translation, natural language understanding, and text generation. The two most prominent LLMs are OpenAI’s ChatGPT and Google’s Bard.
Putting the technicalities aside, three things stand out that make generative AI revolutionary to the average consumer. First is the user interface. People are suddenly empowered to create via naturalistic text interaction with an AI, rather than being intermediated by a programming language (or even a team of programmers). The second is just how realistic the results are. We are used to computers creating ‘dumb’ outputs, but LLMs can speak back to us in natural language and hold persistent conversations. Thirdly is that generative AI performs well at creative tasks. We are used to automation impacting ‘blue collar’ jobs but now ‘white collar’ jobs are firmly in the cross-hairs. Programmers, artists, lawyers, writers, and more are seeing AI move into their fields.
This leap forward in technology has had an electric effect and sparked the imagination of many, including in the travel industry. ChatGPT reached one million users within just 5 days. For context, Airbnb took two and a half years to reach the same benchmark.
It seems like we are approaching a new ‘internet’ or ‘smartphone’ moment. Generative AI is, like these earlier innovations, a generalist technology that will have far reaching effects across the entire economy. Early smartphones initially made it easier to call, text, and listen to music. These first devices were incremental improvements over existing technology but of course the true impact would be so far reaching as to change the very fabric of daily life across the world.
It took 20 years for 50% of American households to connect to the internet, but smartphone adoption crossed the same threshold in six years. How much faster will generative AI become mainstream? And when it does that, what will the 2nd and 3rd order impacts be?
The Impact of Generative AI on Travel
Who would’ve guessed that a new phone would upend centuries old travel businesses, but yet that’s exactly what happened. Travel companies which did not adapt to the mobile era, or to the desktop internet era before that, did not survive.
This report will analyze the impact of generative AI on the travel industry, and large language models specifically. And while there are many powerful AI tools out there for big data analysis, personalized recommendations, image generation, and more, they will not be the focus of this research. Large language models like ChatGPT have the potential to drive a wave of disruption and displacement in the travel industry similar to that of mobile phones, and yet the sector seems unprepared for it. An Accenture study on AI maturity found that travel was among the least advanced industries for AI. Travel ranked in the bottom five out of the 17 sectors studied.
Similarly, an analysis from Morgan Stanley showed a surge of interest in hiring for AI positions. Nearly 9% of recent job postings in the tech sector were for AI roles. Yet the discretionary sector, where many travel companies sit, saw the lowest scale of AI hiring.
We see at least four significant use cases for generative AI and large language models in travel. These are well-defined markets with medium-to-high visibility into the role that LLMs could play. This allows us to create a dollar estimate for each.
- Customer Service Chatbots: LLMs can significantly improve the chatbot experience and make it more useful for customers. We estimate this could be a $1.9B market.
- Developer Efficiencies: Generative AIs can help programmers write better code, faster. This leads to faster development cycles and more new tech. We estimate that this could present $350M in annual savings.
- Reputation Management: AIs can help evaluate customer sentiment and allow travel businesses to respond to online reviews, boosting their online reputation. We estimate that this could be a $1.3B market.
- Performance Advertising: One of the key strengths of generative AI is the ability to search for, summarize, and present information in an accessible way. This stands to change the travel planning searches are made and with that will command a significant share of performance advertising dollars. We estimate this could be a $5B market.
As far as speed to market, operational efficiency in the form of ‘co-pilot’ tools are already here and will be where the first impact of AI is felt in travel. Customer support and reputation use cases can be developed relatively easily as well and we expect these tools will be the second wave of AI in travel. Changes to consumer search behavior, and with that, the impact of AI on the travel marketing funnel will probably take the longest time to play out.
Our baseline is for ~$8.4 billion in AI created value in the travel industry. But there will also be additional, and potentially even more impactful, ways that AI will transform the travel industry but we have less visibility into how these second order impacts will play out.
For instance, AI stands to disrupt not just travel planning searches – and performance advertising – but it may also upend searches for travel inspiration as well. With that would come changes in how all sales & marketing throughout travel is done. Another case is operational efficiency. These need not be limited to only developer tools and AIs may set the stage for broad-based productivity gains across millions of travel industry jobs. Another scenario we envision is that AI assistants may change how complex technology stacks are integrated.
These additional categories where we are ‘dreaming big’ about the impact of AI on travel are harder to estimate with any true precision. But we believe they could potentially represent a $20 billion opportunity. Paired with our baseline market estimates we see artificial intelligence driving a $28 billion impact within travel.
Baseline AI Impacts on Travel: What Gets Disrupted First?
As already discussed, we see several core areas where AI can have a large impact on travel within a short-to-medium term time frame. These four categories, chatbots, reputation and review, operational efficiencies, and search, make up our baseline estimate for $8.4 billion+ of generative AI value creation in Travel. We discuss each of these in its own section below.
Customer Service Chatbots
Chatbots are fertile ground for AI-powered large language models. Afterall, this type of conversation plays squarely into the strengths of a LLM while at the same time ‘old’ chatbots are in regular use across the travel landscape. Combining the two should be relatively straightforward and we expect it to be one of the first places that this technology will be rolled out in a guest-facing manner.
Luca Zambello, the CEO of Jurny, a short-term rental tech platform, has been among the first to go live with a ChatGPT-powered customer service bot. He sees huge potential for LLM chatbots in travel to revolutionize customer service. “Our AI can already handle full on conversations with guests,” Zambello says, “basically you have an infinitely scalable 24/7 guest support team.”
Jurny reports 85%+ accuracy for its AI, meaning that just 15% of responses needed to be rewritten by a human supervisor. Most striking to Zambello is the AI’s rate of improvement, up from 50% accuracy two months ago. “We didn’t expect to be this good”, he admits, “I think a lot of people are still underestimating the power of what’s happening… This is here, this is now, and the rate of improvement is mind blowing.”Many travelers are likely already familiar with guest messaging tools and chatbots which have been rolled out from retail to banks and more. Yet Skift Research’s Hotel Tech Benchmark: Guest-Facing Technology 2022 finds that chatbot adoption in accommodations has been slow. We estimate that just 5% of hotels use a guest messaging or chatbot application.
Adoption in other parts of travel, like airlines and online travel agencies, is higher but still falls well short of ubiquitous. This will likely change as the value proposition of these tools comes into focus.
Looking outside of hotels and short-term rentals, Expedia Group recently launched a GPT powered chatbot on its mobile app. Chinese giant Trip.com also launched a generative AI chatbot in March. Both bots are relatively basic so far, but deserve credit for their speed to market. They represent just the opening shots in a coming wave of new travel chatbots.
“It’s easy to imagine integrating generative AI into our AI-powered virtual agent,” Rathi Murthy, Expedia Group’s chief technology officer told Skift. “It already helps travelers self-service complex queries like requesting a refund or changing a flight, but it is still pretty linear in its conversational capabilities, so it’s an area where generative AI could help us deliver a richer, more conversational experience for our travelers, and also help our B2B partners focus their time on what they do best.”
There is a strong cost motive that will lead many companies to adopt AI chatbots. Zambello points out that, “if you look at the biggest payrolls of [short-term rental] management companies it’s actually guest support.” Customer service is a major cost center in nearly every other travel business as well.
Smart chatbots don’t replace an experienced customer service representative but perhaps it can handle a lot of the relatively simple requests that many first line agents receive. At some companies that might mean a leaner labor force, but for others it is about re-allocating the time load on its agents. Smart chatbots will free up valuable time that allows human operators to handle a higher volume of complex calls.
The airlines learned this lesson the hard way after many experienced catastrophic meltdowns in customer service over the past several years. Airline call center wait times reached hold times of up 12 hours during summer 2021, according to The Wall Street Journal.
It’s not just the added expense of growing headcount but the difficult process of training new staff. Southwest Airlines hired more than 1,000 customer service representatives last year and Delta Air Lines says that more than half of its reservation agents were hired in the last two years, according to the same WSJ reporting. But scaling up call centers takes time. These green agents are simply not as effective as experienced representatives and so even following a hiring spree, customer service woes will continue to build until those hires become seasoned.
Chatbots have no learning curve and can scale to handle any volume of requests with little to no training time. Plus, they can speak in most languages, a massive value-add for the inherently global travel industry. The caveat is that chatbots need to be sufficiently ‘smart’ and to interact in a similar manner to a real human agent. No one wants to be banging their head against a wall talking to a robot with preprogrammed, irrelevant, prompts.
ChatGPT and similar language models seem poised to cross this human-like threshold and can improvise responses rather than sticking to a script. This flexibility unlocks the ability for AI chatbots to serve in front-line customer service roles without upsetting customers. We think this natural language capability creates a multi-billion dollar market opportunity.
While there is great potential for AI-powered customer service, it will not be a panacea. With apologies to Ron White, you can’t fix stupid travel policies. Bad company policies will frustrate consumers no matter how smart your AI gets. A particularly infuriating customer service this author recently experienced trying to get an airline to play nicely with an online booking platform served as a reminder that all too often travel companies employ intentionally confusing and opaque selling tactics. The greatest large language model chatbot in the world is still not useful if it is enforcing a policy that is fundamentally frustrating to a customer.
An outgrowth of customer unfriendly policies is that guests often call customer service looking for exceptions to rules, hoping a human will let them slide. Chatbots likely won’t have authority to execute on those exceptions to policies and so don’t solve this issue. Plus, let’s face it, right or wrong customers are often calling just to complain and vent. They just want to be heard as much as anything related to solving their issue. A chatbot simply can’t offer the empathy that a traveler wants to hear in a stressful moment.
These issues ensure that the phones will keep ringing at travel companies for years to come. But alongside those call centers, the coming of AI powered chatbots seems imminent and will no doubt create substantial value for travel and tech providers.
Skift Research estimates that the total addressable market for hotel guest messaging apps and chatbots is $1.2 billion. We expect that AIs will readily take 80%+ of that market, implying a $930 million market for LLM chatbots in hotels.
Our original analysis of hotel chatbots was published before the public launch of ChatGPT. This makes the estimate conservative as the addressable market is now likely larger given the use case for chatbots has expanded since.
We don’t have tech benchmarks for other sectors across travel but putting our hotel numbers in context may help. Our addressable market estimate for AI chatbots is one-tenth of a percent of the estimated $838 billion of hotel room revenue in 2023. If we extrapolate that ratio across other major travel sectors, it implies multi-million dollar opportunities for each. Altogether, Skift Research estimates a $1.8+ billion addressable market for AI powered chatbots in the travel industry.
Reputation and Review Management Software
Another guest-facing area where AI has the ability to make rapid inroads is the world of reputation and review management services. Guest feedback is a key performance indicator across the travel sector and online reviews can have substantial revenue impact. Studies show that 95% of travelers read reviews before booking a hotel and 85% of consumers trust reviews as much as personal recommendations.
Given the importance of online reviews an ecosystem of tech vendors has sprung up to make it easier for travel companies to track, manage, and respond to guest feedback. Good software in this category prompts happy guests to leave positive reviews. But it also helps manage the fallout from negative posts. Skift Research believes that ~45% of hotels use some form of reputation and review management software today.
Travel companies are very interested in getting guest feedback but the process can be difficult. There are many different places where a user can post a review – Tripadvisor, Google Reviews, Yelp, online travel booking sites, Reddit, Instagram, and more. It’s too much for one person to handle, but searching a large corpus of online text reviews is something that an AI is well suited to do.
The next stage after finding a review is to understand what it’s saying and the sentiment expressed. Here too, it’s easy to imagine how new AI will be an improvement over the existing tools. With new advancements in technology, AI tools can deliver real time sentiment reports to travel businesses summarizing what guests most enjoy or what they find the biggest pain points. This, for example, could clue a hotel general manager into what his staff is doing well, or highlight areas of improvement.
The final step of online reputation is to respond to reviews, especially negative ones. Harvard Business Review’s analysis on Tripadvisor reviews shows that hotels that responded to guest reviews received 12% more reviews and their ratings increased by 0.12 stars. The generative nature of large language models makes them ideally suited to this task.
Murthy told Skift that travel suppliers often ask Expedia to, “provide them with insights on new opportunities like what they’re doing well, what could be improved, how they can adapt to attract higher value travelers. Last year we introduced the guest experience score to partners that does just this – it uses AI to look for signals we collect from guest reviews, refunds, and relocation data, along with actions and recommendations partners can take to improve their service and increase their visibility on our marketplace.”
Zambello, the CEO of Jurny has also deployed an AI-driven reputation management tool which is capable of suggesting carefully worded responses to online reviews. He explains, “the AI already knows what [the guest liked or was upset about] And then I say, ‘can you generate me an answer based on this specific feedback so you can address this?’ Then each owner can modify [the AI] based on their style but it will know how to address this review specifically for that guest, that market, and other people that are in the future going to read [the response].”
Zambello says that there can be a lot of nuance to these features around how to build response prompts. For instance, rather than offering a generic ‘sorry’ to a bad review, the AI could be prompted to write a response that includes some persuasive language that appeals to how a future potential booker will see the review.
Skift Research estimates that the total addressable market for reputation and review tech in hotels is $1.25 billion. We expect that AIs can take ~50% of that market, implying a $630 million market in hotels.
We don’t have tech benchmarks for other sectors across travel but putting our hotel numbers in context the addressable market estimate for AI reputation and review is 7 basis points out of the estimated $838 billion of hotel room revenue in 2023. If we extrapolate that ratio across other major travel sectors, it implies multi-million dollar opportunities for each. Altogether, Skift Research estimates a $1.3+ billion addressable market for AI powered reputation and review software in the travel industry.
Operational Efficiencies for Developers
Alongside chatbots, the first real-life use cases for AI will be internal employee productivity tools. Brands are risk averse and don’t want to be associated with the fallout from a glitchy AI gone rogue. Starting with rolling out AI at home limits the damage from any accidents.
Standing up an AI is hard work, and it is better to get your practice at-bats done internally. There will be a lot of data-wrangling involved in building use-case specific AIs. Where is your data stored, who is responsible for it, is it tagged properly, and can it be transferred easily across data centers? Are you still running on-premise software? If so, you have a large data migration project ahead of you before you can even think of launching an AI. Building an internal AI tool can serve as a good dry run to understand what your data requirements will be for a future public-facing bot.
And what about privacy and data sharing rules? Before you run the last ten years of customer emails through ChatGPT it might be wise to pause and understand whose data you are holding and what level of privacy you owe to each different piece of info – both from a strictly legal standpoint as well as from the perspective of public perception and your organization’s internal code of ethics.
Building an internal tool makes a lot of privacy concerns easier to tackle, though it is not carte blanche to ignore data sharing considerations entirely. But let’s not think that the only reason to build an internal AI tool is to save yourself a lot of practical headaches. No, there are major productivity gains to be had from using AI as well.
Shane O’Flaherty, global director of travel, transportation, & hospitality at Microsoft has called AI’s operation efficiencies, “the hidden layer within all of this.” He says that backend AI can “enhance and bring you quicker to market around an application that you’re developing.”
Large language models, it turns out, are quite good programmers, and one of the earliest places we are seeing AI-enhanced employee productivity tools show up is in software development. To that end, GitHub, a software programming platform, launched an AI tool called Copilot. Think of Copilot as a super smart autocomplete geared to specific coding languages and tasks. The tool is now available to all enterprises and Github says that as of February 2023 Copilot is writing 46% of code for developers that use it. That’s up from 27% in June 2022. The numbers are pretty astounding: developers worked 55% faster using Copilot and were 75% more fulfilled with their jobs than peers not using AI.
Travel companies in the U.S. employ ~13,000 people in computer-related occupations such as software programmers and web development, according to the Bureau of Labor and Statistics. This is just about 1% of the U.S. travel workforce. That’s low as travel is a labor-intensive industry with the need for a lot of front-line service staff and back-of-house operational people. Plus, many travel companies outsource some or all of their development efforts, further shrinking the share of in-house techs. The impact is largest in the travel agency sector where nearly 5% of the workforce is in engineering (note: this classification includes offline travel agents, pure-play OTAs will have a higher engineering workforce share).
But despite the small relative share of computer engineers within travel, these are some of the highest paid positions in any travel organization. The average developer for a travel business is paid $88,000 a year compared to an overall average salary of $49,000. At nearly twice the prevailing industry wage, development costs can add up quick. We estimate that globally, the travel industry spends $3.5+ billion a year on developer salaries.
Based on the Github numbers given above, a 10% efficiency improvement across web developers seems conservative. Relative to our cost estimates this implies $350 million of savings per year for the industry, just from a part of the industry that accounts for 1% or less of global headcount.
To be clear, we think that most of this cost saving does not come in the form of direct layoffs. In fact, based on the employment data it seems that travel is if anything under-staffed regarding IT and software professionals. Rather these cost savings come more from needing to hire relatively fewer new expansion engineer slots to keep up with the current pace of growth. A business that needed to add ten engineers might be able to get away with hiring only nine, for instance.
There will also be second order impacts for travel tech. As engineers become more efficient it will be easier for travel tech startups to launch with fewer engineers. AI developer tools stand poised to unleash a new wave of startup creativity and disruption for the travel industry. If past startup successes like Airbnb are any indication, these yet to be launched travel startups could drive billions of dollars of new revenue in innovative product categories.
AI Powered Search
Over the medium-to-long term, online search may be the sector most impacted by the rise of AI tools. Generative AI and large language models seem poised to change the research process by restructuring how information is organized and made useful.
We don’t think that a wide-spread restructuring of search will happen quickly. Chatbots and internal apps will be the most immediate use case. However, nearly every trip in the world involves some sort of search function. And so an AI-led disruption of search will have far reaching consequences that touch every piece of the travel ecosystem.
There are fundamentally two types of travel searches based on where the customer sits in the purchase funnel: dreaming/consideration searches and planning/transaction searches.
Transactional searches that take place while planning travel are lower down on the purchase funnel and are likely the first part of the search industry to be impacted by AI. In this search phase, the rubber meets the road. The idea of the trip has been formed and now the traveler needs to start booking flights, accommodations, in-destination transport, and activities. This is the point of highest purchase intent in the travel planning process and is the most valuable advertising moment for OTAs and travel suppliers.
The key pain point for consumers is complexity. For most destinations there will be many different flight options and a huge range of accommodations. What flight to take? Connecting or direct? Is the hotel location any good? What about price? How’s the service? Are there good reviews? Are those reviews trustworthy?
Consumers want choice when it comes to travel decisions. That is a core reason why online travel agencies continue to do so well – consumers value the ability to comparison shop across a wide range of travel products. Paradoxically, while choice will sooth a traveler’s fear of missing out on a great deal, it also creates a huge amount of stress. Some road warriors enjoy the travel planning process but most find it overwhelming. There is just too much information to sort through and too many hidden pitfalls in the process.
Large language models have the potential to shine here. Their strong suit is synthesis and so AIs can do a great job at sifting through the huge range of inventory out there. This brings together the best of both worlds. You know that the AI is capable of processing millions of pieces of data, far more than a human ever could, and so it helps alleviate the FOMO that comes from booking via an agent or closed loyalty program. But on the other hand, it requires little effort on the part of the shopper. The travel searcher is able to get a summarized list of just a few travel choices to pick from while still having confidence that all possible other options were looked over and exhausted.
A critical part of the convenience that ChatGPT and other LLMs bring which earlier technologies lack is the ability to understand natural language requests. While past travel search tools can summarize millions of results for you, it requires users to craft requests built around filters and maps. The strength of generative AI is that it eliminates the need for complex filters and allows for text searches that were simply not possible before today.
For instance the following search would be relatively easy for an LLM but nearly impossible for google search or an OTA: “What are the best choices for a family friendly hotel room in midtown New York for under $200 a night? Walkability to local attractions and cleanliness is key and more important to me than the hotel’s lobby or customer service.”
True, all of that data does exist today. Between a series of searches (hotels in New York), complex filters (Price < $200 dollars), map geofences (limit to midtown), and reviews, a traveler could parse all of this together and come up with a list of properties on Expedia or Google Hotels. But today there is no elegant way to ask all those different criteria in a single prompt and get a single coherent response back.
While that kind of travel search is the dream, and theoretically possible with today’s technology, the real challenge will be data integration. Google was founded nearly twenty years ago but you still can’t book a Southwest Airlines flight via the search engine. And to state the obvious, it’s not like Southwest can’t figure out how to make their booking engine accessible via Google. The airline has made a business decision to limit third-party data integrations to drive first-party bookings. This is a strategic move that has nothing to do with the technological landscape.
Siloed travel data is a structural problem and not a tech issue. It’s not just Southwest. It’s the industry dragging its feet on NDC, it’s the skirmishes between hotels and the OTAs, it’s the scuffles over the GDS and Google search, and it’s the many travel industry suppliers who run their operations via excel, or worse, pen and paper.
In short, it doesn’t matter if an LLM-powered travel search engine is technically possible today – it is! – what matters is if the travel industry will rise to the challenge of embracing these tools and offering customers a new form of searching and shopping.
In theory, the rise of LLMs will benefit travel distributors over travel suppliers. Today online travel agencies and global distribution services hold the most aggregated data about hotel room properties, customer booking behavior, flight and route data, and more. Many underestimate the sheer scale of information required to train an AI. Even the largest hotels or airlines in the world will likely fall short of the critical mass of data required to train a LLM. This leaves distributors, which have an order of magnitude more data than suppliers, best positioned to build a travel LLM.
Murthy, Expedia Group’s CTO, drove home just how critical the right technology and datasets are to building AI tools. “Travel is a complex industry with lots of tech debt, so not many travel companies can integrate with new AI technologies like ChatGPT with the velocity we have,” she told Skift. “Companies need to make sure the technological fundamentals of their platform are sound, and their data quality is high – we have both already. We also have a highly skilled team of technologists, and over 70 petabytes of travel data on booking patterns, traveler behavior and traveler preferences that’s powering our platform. This is a huge competitive advantage for us. We’re able to unlock the data we have by baking AI and machine learning into the platform that’s powering all of our brands like Expedia.com, Vrbo and Hotels.com and our B2B products.”
In fact, we have already seen the first ChatGPT integrations in the travel sector take place via travel distributors. ChatGPT recently announced plugins that allow its AI – which is only trained on data up to 2021 – to access specialized datasets and up-to-the moment information. Expedia.com, the flagship brand of Expedia Group, and Kayak, the metasearch brand owned by Booking Holdings were both launch partners for OpenAI plugins. Opentable, also owned by Booking Holdings, was a launch partner as well. Navan, formerly Tripaction, a leading corporate travel agency has also been among the first to integrate ChatGPT and generative AI tools along with its first-party data.
In a demo that Expedia shared, ChatGPT uses the plugin to provide flight itineraries, offer hotel and STR pricing, and suggest things to do while in destination. All results include links to the Expedia.com platform to complete the booking process. Murthy said that, “collaborating with large language models in this way, enables us to take advantage of the strength of our data, and tech capabilities and keep experimenting with advanced AI in a way that others can’t.”
However, in recent years we have seen many suppliers push back on distributors and work hard to drive first-party bookings. Hotels, for instance, have spent billions of dollars of the last decade on M&A to build scale within their loyalty programs. But while suppliers have grown their property counts or flight routes, they have not monopolized their internal data since even the largest travel brands still rely on third-parties for incremental distribution.
But with the rise of AI, data will be the newest front in the ongoing channel wars. Perhaps this will spark a new wave of pushback where suppliers try to claw back even ‘basic’ data (e.g. route and price data, room rates and availability) from their partners?
Or does this technology finally break down those barriers? Perhaps AI and the powerful guest experience it offers could act as the ultimate carrot to drive the industry to finally embrace data sharing and collaboration.
This push and pull reminds us of the impact of the internet and, later, mobile phones in travel. Early adopters embraced the new tech while just as many dragged their feet. It’s true that the early hype never came to fully pass, but even so these tech cycles were eventually too powerful to ignore. Travel brands that didn’t adapt to the digital or mobile era eventually had to make large investments to catch up or went bust.
Just how big is the market for AI and large language models in travel search? Let’s look at performance advertising via Google search. These are some of the highest intent searches with strong conversation performance, making Google one of the most effective advertising channels in the travel industry.
Although queries to LLMs are structured differently from a traditional Google search text box, they are performing functionally the same role. Looking at the below answer from ChatGPT to our earlier question about budget family hotels in New York, we can plainly see that there are low-hanging opportunities to integrate sponsored ads into the results page.
At the end of the day, ad dollars will follow consumer eyeballs. And if AI is able to deliver smart and relevant answers to common questions about travel planning, then its user base will grow and advertisers will follow.
Skift Research estimates that the global travel industry spent nearly $14 billion on Google performance ads in 2022. That is up from $9 billion in 2021, and 14% above 2019 levels. Of that spend, we estimate that 53% was from online booking sites and that Expedia Group and Booking Holdings together spent nearly $7 billion on Google last year.
Google is just one player and we estimate that the global travel industry spends nearly $25 billion on advertising. It won’t happen immediately, but we think that over the medium-to-long term it is reasonable to assume 20% of these ‘old’ performance advertisements will be converted to some form of new LLM-driven ads.
By using our more conservative figures for Google only this suggests that, at a minimum, the market for AI search could be $2.9 billion a year. Our more aggressive estimates of global advertising imply nearly $5 billion a year in AI advertising spend.
Bear in mind that this does not necessarily mean dollars flow away from Google. If Google’s AI tools are good enough then the revenue will stay within the Google ecosystem but be reallocated from the traditional search engine results page to Bard and other similar tools.
And in fact, since AI is a big data game, Google may even grow its overall advertising share as smaller travel metasearch players are unable to train LLMs that can compete. So while AI may be opening the door to Microsoft Bing and OpenAI as rising threats to Google’s position in travel search, the same trend may help cement Google’s growing lead over travel-specific metasearch platforms.
Dreaming Big: How Could AI Transform Travel in the Long Term?
We have now covered our four baseline buckets where we see direct line of sight to AI market share that is readily measurable. But just like with the smartphone, the first order impacts of generative AI on the travel industry will likely be the smallest.
The second and third order impacts of AI on travel are longer term and harder to measure with any meaningful accuracy. But they likely stand to be a full order of magnitude larger than the baseline impacts. In this following section we consider three areas where AI could have deep and long-lasting impacts on travel. These areas are software integration and personalization, travel inspiration, and broad-based labor efficiencies.
We do try to put numbers around these opportunities and estimate that if we ‘dream big’ about the impact of AI in travel, the opportunity could be $20 billion vs. our $8 billion baseline. However, we caution that these numbers are meant to be illustrative exercises that demonstrate just how much larger the AI opportunity could grow in the long-term rather than firm projections.
Software Integration and Personalization via AI
One area where AI’s might change travel spending in the long term is software integration fees. Today’s travel tech landscape is highly fractured and integrating multiple pieces of tech into a single working platform can be a heavy lift. As a result, many vendors charge integration fees to install their software and to keep it connected to all the different pieces of travel tech in the marketplace.
These integration fees are not chump change either. The center of any hotel technology system is its property management system (PMS). The PMS is the brain of most hotels but needs to integrate into all the different parts of the hotel’s operational and distribution systems to work effectively. With so many connections needed, Skift Research estimates that PMS vendors generate nearly $200 million in integration revenue a year. Other heavily cross functional tools like revenue management systems and customer relationship management tools also levy hefty integration fees.
We estimate that hotels today spend $470 million annually on integration fees, about 6% of the total tech spend they invest in those same categories.
In the long run, automated AI tools have the potential to ease these tech integrations. Microsoft’s O’Flaherty envisions a world where, “I’ll have my own AI assistant, and then I will permission brands into my world that I love… this idea that as I walk into a hotel, my AI assistant has all my content of who I am or what’s my purpose on my journey. Then the AI assistant of the hotel will connect me once I walk in.” In this world where AI assistants are common and exist for both the consumer, the hotel, and their vendors, integrations become quick and seamless.
Growing from a current $471 million market today, we see the full potential addressable market of tech integrations to be $1.35 billion in hotels. We believe it is reasonable to assume that 60% of that can be replaced with or captured by AI tools, implying an $810 million annual value-add.
Using the same methodology as we have done for earlier hotel tech estimates, this represents 0.1% of global room revenue which we then use as a scalar to extrapolate this category globally across all travel sectors. Our numbers imply a $1.6 billion annual value add from AI integrations.
But if we are going to dream truly big, the real opportunity is not from cost savings but from new revenue opportunities that can come from AI-led personalization. O’Flaherty explains that, “the digital experience from a human perspective is here. I think [AI is] going to bring those two together where high-tech and high-touch come together for airlines, and hotels. I think it’s going to be super dramatic.”
The prospect of guest AIs interacting with travel AIs and driving a fully personalized experience creates new opportunities for brands to surprise and delight and to offer the right upsell at the right moment. Erik Tengen, the Co-Founder and CEO of Oaky, a hotel upselling software company, says that, “as a company we try to use data and knowledge from the hotel about their guests to try and be as proactive as possible to offer the right type of deals at the right times through right channels basically, and that brings us to AI. One component that is key is when do you do the upsell? What content are you actually offering?”
Tengen has found that when hotels target their upsell offers by segmented customer profiles, the more personalized deals sell 80% more than non-segmented deals. That is a huge amount of uplift from what is a fairly simple data analysis with existing tools. What will AI bring then? “It’s going to drive significantly more ancillary revenue for the hotel on the property experience,” says O’Flaherty, ”I think that’s huge because that’s complete white space for the hotels.”
But now for a quick reality check. Plenty of hotels still struggle to wrangle the data needed for even the most basic customer segmentations. So while hyper-personalization is still far off, we can begin to get a sense of the massive revenue uplifts that AI might unlock.
Searches while dreaming of travel
We discussed highly transactional trip-planning searches earlier but here we will move back up the marketing funnel to talk about how AI could impact searches that travelers make at the start of the trip-planning process.
Most trips, especially leisure trips, start with the inspiration and dreaming phase. Here, the traveler is open to most destinations and wants to explore options and firm up ideas for the trip. Friends and family are the most important source of ideas. Also key are visually rich mediums like social media, television, and glossy magazines. We see a world where travelers in the early stages of trip consideration turn to AI as an additional source of inspiration.
ChatGPT is already able to provide some very basic inspiration for travel destinations. In the below screenshot we prompted ChatGPT to suggest five kid-friendly destinations that also offer culture and dining for parents. The app suggested Tokyo, Japan; Barcelona, Spain; Vancouver, Canada; Copenhagen, Denmark; and San Diego, USA. It also gave some recommended activities in each location such as Tokyo Disneyland or the San Diego Zoo. Honestly, this is a pretty good result. Especially for an early-stage technology that is still in beta testing for all intents and purposes.
Skift Research survey work shows that online search is the single most important source that travelers use when planning a trip, used by 46% of all travelers in 2022. As LLMs improve they will increasingly be used in these online search use cases and with that stand to drive a large shift in how search engine optimization and brand building is done for travel companies.
Expedia Group believes that AI can help travelers search for “trusted advice.” CTO Murthy told us that, “the more sophisticated AI gets at recognizing a traveler’s intent and predicting the next best course of action for them, the more complexity and friction it will help take out of the planning, shopping and booking experience for travelers.”
However, we think there are limitations to how far LLMs can expand within this segment. For starters, we know that travelers consider multiple sources when planning a trip. And while search is the most commonly used trip planning tool, that is nearly always complemented by other channels like friends and family.
Social media too, is very important to trip planning, especially in early stages of inspiration. ChatGPT does not have the visual appeal of Instagram or TikTok. It’s never been more true that a picture is worth a thousand words. A photo of beautiful blue water inspires more beach-goers than the most eloquent description that ChatGPT will ever write up. And as the world of social media moves even deeper into video, we think that a well-designed travel reel still has more power to inspire than an AI search.
There is also an issue of trust and authority. Few things in the world of travel carry the same weight as a recommendation from a trusted friend. And an AI endorsement is not going to change that. Travel is a very high consideration purchase and AIs have already built quite the reputation for confidently spouting off nonsense. Travel is expensive and vacation days are limited. Are you really going to pick your next vacation destination solely based on the advice of an AI? We think not.
Creativity is a limiting factor here too. Friends, family, and even good travel agents can make smart and unique recommendations for where to travel next. These individuals can draw on past behavior, personality traits, and unspoken behavior to suggest travel destinations. This is especially important for those seeking distinctive getaways. AIs and LLMs are based on pre-existing data and so tend to skew towards the lowest common denominator. Sure, you could ask Chat GPT what the top beach vacation destinations are, but you will get an answer skewed towards what has already been written up in best of lists.
In fact, we asked ChatGPT this very question and it gave us Maui, Cancun, Phuket, the Maldives, and the Seychelles. With the possible exception of the Seychelles, none of these are out-of-the-box suggestions. Recommending off the beaten path destinations is not just a hipster affection, it makes business sense. High net worth travelers want to visit distinctive destinations and Skift Research finds that 70%+ of Gen Zs and Millenials want to go to a place their friends didn’t think of.
There might still be a place for LLMs to recommend mass travel but we also note that, looking at ChatGPT’s beach getaway suggestions, many of these destinations struggle with overtourism. ChatGPT may not only be leaving money on the table by recommending destinations that don’t appeal to richer travelers and the rising generation, but it may make overtourism worse by concentrating tourists rather than dispersing them.
Despite these limitations, we expect AI will become integrated in various forms into the search process. AI is unlikely to be the sole tool used when dreaming up new travel destinations, but it will be used alongside traditional sources of inspiration to enhance recommendations from friends, travel agents, and social media.
As LLMs and AIs become a larger part of the early-stage search process, we expect these tools will capture a growing share of travel company brand marketing budgets. Currently these “top of funnel” campaigns are designed to build brand awareness rather than drive booking conversions and are concentrated in visual mediums, but we could see them shifting towards AI.
When asked about changes to search engine optimization due to AI, Murthy said that, “we see any advancements that democratize search and make it a more level playing field as a good thing. For us it’s important to meet travelers wherever they are and offer intuitive ways for them to start planning their next trip.”
In other words, Expedia will be optimizing its sales and marketing to follow the consumers and is relatively agnostic about the search platform used. We suspect that Booking Holdings and many other travel brands feel the same way. And in truth, many travel brands may be eager to see a move away from a Google-dominated marketing funnel. All the more so if they have a data-driven edge over other competitors.
While providing an exact estimate is difficult, if AI captures even a small amount of travel’s brand marketing spend, that could equate to tens of millions of dollars in investments. On top of the performance advertising dollars we discussed in our baseline, there is also potential for AI to drive other forms of sales and marketing. This includes spend on search engine optimization, brand awareness, AI-generated art for marketing campaigns, and direct sales.
We estimate that travel companies spend ~$68 billion a year on sales and marketing. Less the amount spent on Google, that leaves $54 billion on these other activities. Search engines are most in the crosshair of AI and so these other activities should see a lower displacement than Google. If just 5% of non-Google travel sales and marketing activities are replaced by AI-tools, that implies an additional $2.7 billion in annual spend.
Operational Efficiencies beyond Developers
We discussed earlier how productivity improvements from developers across travel could drive $350M+ in annual savings. But focusing on just computer programmers, while an immediate and actionable use-case today, is ultimately thinking too small about the potential for AI-led operational efficiencies in travel.
One Github study found that developer productivity is linked to the ability to have “good days” at work. You know those kinds of days, right? The ones where you are fully engaged, checking items off your to-do list, and hitting a flow state. But those days can be hit or miss, dependent on your mood, the weather, your coworkers, or a long meeting.
One way to think about Github’s Copilot is that these AI tools make it easier for developers to have “good days” of peak productivity more consistently. Imagine what that would mean in travel and hospitality broadly? When a front-line worker has a “good day” it creates long-lasting positive guest experiences. We’ve all also experienced the opposite. The dread of approaching a customer service worker who is clearly having a “bad day.” What would it mean for the travel industry to consistently unlock “good days” across the tens of millions of workers we employ and who interact with guests every day?
Here is where the true potential of operational AI in travel lies. Richard Valtr, Founder of Mews, a hotel property management system reminds us, “that’s basically what technology does. It actually empowers people.” Valtr says that a focus on how many people AI can replace is misguided since most travelers will always expect a human touch in hospitality. Rather, AI tools can make staff more flexible and allow them to be switch-hitters that can seamlessly play multiple roles. Valtr believes that AI tools will give us, “the ability to not have specialized staff… now everyone should be a concierge.”
AI tools can create greater productivity, higher employee satisfaction, and shorten training learning curves. They have the potential to free us from the tyranny of bullshit jobs. The kind of menial work that saps our mental energy and leaves us having a “bad day.” Those bad days are even more costly for travel than in most industries, given the service-led nature of this business.
Imagine what the productivity gains would look like with a few more “good days” for every worker. Even a small improvement, just 1% could have big repercussions. In the U.S., the average revenue earned per travel worker per year is ~$260,000. And so, a 1% improvement doesn’t feel like a lot. That’s just an incremental $2,600, it feels like small stakes. But if AI tools succeed in making every travel worker just 1% better at their job, we estimate the results would be a massive $15 billion in incremental revenue on a global scale.
Conclusion
It’s fascinating to reflect on that final category. The potential of just a slight productivity boost across the entire travel workforce outweighs the value of AI in nearly every other category we considered combined. So just how realistic is the 1% per person productivity gain we envisioned?
Did the computer drive more than a 1% productivity gain per person? Did the internet drive more than a 1% productivity gain per person? Did smartphones drive more than a 1% productivity gain per person? The answer is clearly yes, and generalist AI tech has the potential to do the same. With that comes tens of billions of dollars in potential new revenue streams and costs savings.
But let’s not get ahead of ourselves. This technology won’t be implemented overnight. AI requires massive amounts of data and for that data to be well cleaned, tagged, and easily accessible. That’s a tall order for the travel industry which, let’s face it, still has a large amount of tech debt to work through. It is practically a requirement for AI to be run via cloud computing infrastructure but so many companies still run old on-premise architectures. Data silos are common and few companies have a clearly laid out data collection and privacy policy.
Nor will AI solve the many deep and long-lasting structural problems built into the sector. Long-standing distrust between distributors and suppliers, customer unfriendly policies, and overtourism all call for strategic solutions and cannot be fixed by new technology, no matter how smart the AI gets.
Challenges notwithstanding, AI holds the potential for an exciting new world of smarter, seamless, and personalized travel. There will be a lot of work ahead for the industry but early adapters are already hard at work to unleash the power of this new technology. As those first beta products make headway and, hopefully, showcase impressive results, we think that the promise of AI can serve as a catalyst for a much needed upgrade of travel’s tech infrastructure.
There was a palpable excitement from the experts we interviewed for this report. Almost uniformly they were impressed with the speed at which generative AI has advanced over the last twelve months. Many believed that we are standing at the beginning of a new technological era, comparable to the early days of the internet or smartphones. And although we curb some of our enthusiasm given the many practical challenges ahead in the near-term, we too, see the long-term potential of generative AI.
Travel will always be fundamentally about the physical locations we visit and the very real people we meet along the way. AI won’t change that. But travel is also about as pure a service business as you can get. And here generative AI and other AI tools have the potential to elevate the traveler experience to previously unattainable heights. AI holds the potential to surprise and delight guests throughout their entire customer lifecycle, from search and planning, to the travel experience itself, and through any issues that arise along the trip. In this way AI will complement and elevate the travel industry. And if we can succeed as a sector in using these tools to their full potential, Skift Research’s $28 billion AI estimate will look conservative in hindsight.