The Surprising Ways Generative AI Will/Will Not Transform Airline Distribution
Generative AI is set to become the next major platform-level disruption in airline distribution. It will be used so pervasively within the infrastructure of each stakeholder of the airline distribution chain that it will cause business processes and their supporting technologies to be completely rethought and reoptimized for the new world. There is no “small” way to describe its future impact. It is impossible to cover all its expected impacts in a single piece, so we have narrowed ours to one surprising transformation and another widely anticipated transformation that we believe will not occur to the surprise of some. Fortunately, we have a lot of airline distribution history on which we can rely to provide our prognostication.
The Key Lesson from Other Platform-Level Disruptions in Airline Distribution
Platform-level disruptions have occurred every 10-15 years in airline distribution, and they often come with grand predictions of the demise of existing players in the value chain. What occurs is often very different. At the advent of the Internet, many hailed the consumer’s newfound ability to interact directly with airlines as the end of the travel agent. Whereas the proportion of direct bookings did indeed tick upward somewhat but more so due to the coincidental proliferation of the low cost carrier business model than the Internet itself. The Internet’s primary impact was the shift of transaction volume from call centres to websites within each direct and third party channel. Airlines and travel agents alike became more automated.
At the advent of the smartphone and other devices driven by mobile operating systems, many believed the days of the personal computer, clunky websites, and cluttered screens, and all the companies which depended upon them were numbered. There were certainly some interesting new app-first travel companies borne from the era like Airbnb, Uber, Hopper, and WeChat. However, just like the Internet, mobile environments also presented established brands with new ways to innovate. We now see established companies like United Airlines using app-first methods to communicate and serve customers in new, innovative ways throughout their travel journey. Airlines and travel agents have become more mobile-friendly, but the general structure and mix of airline distribution channels remained the same.
The key lesson from the last two platform-level disruptions in travel is they simultaneously provide an opportunity for new market entry and significant innovations from existing players. We do not normally see complete market upheavals with broad swaths of established stakeholders disintermediated. Likewise, we do not see mass disintermediation occurring because of generative AI.
Surprise #1: Generative AI WILL NOT Result in Mass Disintermediation
ChatGPT Operator recently captured the world’s attention with its ability to take on many of the tedious aspects of travel search with a conversational interface and relatively few keystrokes by the user. It is a great proof point of the promise of generative AI from the perspective of the end user. Gone are the many minutes or hours of inefficient web browsing, replaced instead with a wait time of only a few seconds after what may eventually be a deeper and more expert search. So impressed by their experience with the pilot product, some began to echo familiar refrains about how ChatGPT and other generative AI providers could become dominant competitors in travel nearly instantaneously.
However, becoming a seller of air fares requires a lot more than a user experience which amalgamates data it discovers from multiple sources behind the scenes. Any company which wishes to sell air must either be an airline, or a third party authorised by the airline to offer its fares (e.g., a metasearch engine) and/or sell its products and services (e.g., a travel agent). Metasearch engines are often obligated to airlines to provide fair and accurate displays, in many cases with all bookings referred to the airline’s website or app for completion.
Travel agents have many more obligations to airlines, ARC, and IATA as they are involved in the transactions as representatives of the airline and therefore handle more money and data. Any new third party entrant in the airline distribution value chain would most likely ascribe to one of these two models, and both require quite a lot of hard, travel domain-specific work to be done on the supplier side before becoming viable stakeholders in the air distribution value chain.
Naysayers may argue that OpenAI or any other purveyor of generative AI ought to be free to ingest and use whatever airline information they can freely discover on the Internet to power their new pro-consumer experiences. An airline will argue that it has seen this movie before. The first metasearch engines made similar arguments, only to be challenged in court and later brought under commercial agreements with their participating airlines. We already see similar developments in the generative AI space, with agreements between AI providers and content sources like Reddit to have their proprietary information included in searches.
Our prediction is that we will indeed see some new entrants into the air distribution value chain with slick generative AI interfaces, but we will see innovation by existing players to a much greater degree. Generative AI will become another form factor or means of interacting with the airlines and travel agents which already serve the traveling public.
Surprise #2: Airlines Will (Eventually) Use Generative AI to Respond to Generative AI Queries
To say the least, airlines have a lot of rules! Whether they be imposed by governments or the airlines themselves, these rules often create a set of prerequisites which must be satisfied before the next step of the process can be accomplished. For example, most airline systems need to know the trip’s origin, destination, travel dates, and number of passengers before they can offer a fare. Likewise, the airline needs to know the identity of the traveller before a purchase can be processed. The rules and prerequisites only mushroom in servicing scenarios.
Because on one hand the rules and prerequisites are many and, on the other, humans are hard to train, control, and sometimes trust, the industry has needed user interfaces and underlying APIs which rigidly enforce step-by-step workflows for humans to follow. The workflows have served as curbs to keep airlines safe from humans and humans safe from themselves.
Though rules and prerequisites may still exist after generative AI becomes more prevalent, conversations between generative AI systems used by airlines and third party channels will not need to be as prescriptive about step-by-step workflows. Yes, we said “between generative AI systems” purposefully. An airline is unlikely to be satisfied to have its outdated systems designed for human interaction manipulated by a super intelligent system that represents only the traveller’s interests. The airlines will want their own super intelligent system to serve as its interlocutor, just as a buyer uses a buyer’s agent and a seller uses as seller’s agent in some real estate transactions.
Two generative AI systems interacting with one another will not need always adhere to the same series of predefined steps. Rather, they will each have transactional outcomes they would like to optimise – e.g., the traveller must keep to a budget and the airline must try to maximise its share of the customer’s wallet without undercutting its bid price – and certain constraints like rules, prerequisites, and inventory controls within which they must operate. They may each learn some best practices along the way by trying to ask and answer questions differently each time. A new type of machine learning will be needed to optimise not just the offers themselves but the various ways in which the offers were presented, and “conversations” unfolded!
The airline’s generative AI systems themselves will take on the role and orchestration responsibilities of the “human era” APIs they replace. They will likely need to have direct access to a variety of modular systems and microservices within the airline’s infrastructure to accomplish tasks efficiently. They will not necessarily need middleware APIs which aggregate these services together or enforce rigid workflows.
The use of generative AI systems by airlines to respond to requests received from third party generative AI systems will make the overall experience for customers easier and more robust. Two generative AI systems working together will be able to match optimal product and service options for both the customer and the airline, as well as more quickly sort out issues without having to usher the customer through a maze of prerequisites, rules, and workflows.
Conclusion
The advent of generative AI will cause the airline distribution industry to leave some fundamental concepts behind while preserving and adapting others for a new use. It will be fascinating to watch how generative AI relieves the markets of some constraints and rigid workflows which existed mainly to control human behaviour. The industry will be able to accomplish better results more quickly for customers and airlines alike.
However, we should not be too quick to assume that the advent of a new platform technology will necessarily cause the existing airline distribution value chain to turn over completely. Expect instead that the existing players with mastery of the end-to-end processes of selling air, strong research and development budgets, and a clear path to a return on investment to instead be the primary sources of generative AI innovation.