As consumers, we encounter applications of AI and machine learning almost every day. Self-driving cars on the road, type-ahead search results, and recommended shows on your favorite streaming service are just a few of the ways artificial intelligence is embedded into our day-to-day lives. And it’s no secret that modern companies employ this technology to build and maintain a competitive edge.
In the back office, many industries seek to adopt AI-based solutions to improve operations, enhance commercial decision-making, and drive better outputs. While the aviation industry initially may have been cautious to adopt, matured AI-powered solutions – specifically deep learning – have unlocked incredible new opportunities for commercial teams who have long relied on tools and techniques of the past.
Overcoming Constraints of Legacy Systems
Most airlines rely on historical flight data to form the basis for forward-looking forecasts, which typically end up as inputs into pricing optimization. While this may have been successful in the past, the volatile nature of the current demand environment largely invalidates the applicability of past years’ data on predicting the future.
To counteract this, many airlines try to rely on historical data from a year they believe is most representative of the current time, often focusing on 2019 to determine 2022 decision-making. There are obvious limitations to this. First, the demand environment among consumers is fundamentally different post-pandemic. Second, airlines are operating vastly different networks with differing routes, capacities, and competitors than they were three years ago. Additionally, there is no responsiveness in a forecast that cannot decompile the inputs that drive demand. In all cases, airlines often struggle to confidently predict the arrival of bookings, which hinders a range of commercial functions.
Deep learning algorithms represent an opportunity to break free from these limitations. While no human can comprehend and understand the individual context of billions of data points, deep learning thrives with vast amounts of data – from bookings and searches to events, promotions, and competitor prices. When trained and deployed effectively, these models empower analysts with predictive and responsive forecasts that guide strategic decisions across commercial functions.
Aligning Cross-Functional Decision Making
Due to the often siloed nature of airlines, data and reporting that support decision-making either come from a disparate team or are internal to the division. In both cases, humans often manipulate raw data from expectation, experience, or intuition. As that information is passed downstream to other functions, teams are only as confident as the resource before them. This “multi-step metrics” approach means trust must exist universally – and often blindly – across teams and functions. When that trust breaks down, metrics are either ignored or manipulated to match expectations, rather than actually supporting decision-making.
This doesn’t detract from the value that human experience brings to the table. Rather, it highlights the value of new AI-powered solutions that enhance decision-making abilities by replacing a need to rely on instinct with one grounded in data and modern science.
Consider a scenario within airlines: revenue management, network planning, and marketing typically rely on their own forecasts to answer commercial decisions. With a single source of truth for forecasts, cross-functional collaboration advances from reconciling expectations to managing decisions in alignment amongst teams. An RM analyst that notes higher-than-usual final forecasts can not only correct their inventory strategy but also empower network planning to consider additional capacity.
The Human Element
Think about how the introduction of computers to airlines replaced much of the manual, repeatable, effort that came from processing reservations and ticketing requests. These “task manager” systems, far outdated by today’s standards, still represent the core principle of many of today’s tools – take repeatable “if-then” tasks, and complete them as fast as possible.
Deep learning solutions flip the paradigm. Instead of telling a computer what process to follow, human experience, captured through data, helps train deep learning models to complete complex tasks, much like onboarding a new hire to the industry or teaching a young student in school. Put another way, rather than relying on computers to repeat mathematical equations, deep learning models are akin to teaching the computer how math itself works, and through training, the model itself will learn the optimal way to solve any equation, regardless of if it was explicitly shown how to do so.
While previous generations of technology improvements justify why airline analysts may be weary of machine learning models replacing their roles, the reality is that deep learning models allow teams to deploy AI as an omniscient analyst, one who individually understands and recalls the contexts of billions of decisions made in past years, something far beyond the scope of human knowability. Analysts are now more than just analysts; they are teachers, providing guidance to systems meant to support even the most complex decision-making.
Lastly, human analysts will always be able to see around corners to act on information the data itself can’t see, such as schedule changes, operational considerations, natural disasters, or breaking news that changes consumer demand and sentiment. Helping convey and “teach” deep learning models the impacts of these events pairs the wisdom of a hivemind AI with real-time human intelligence. This future of AI-equipped commercial teams within airlines is already here, and we suspect it will continue to grow in ubiquity as the airline industry writes its next chapter.
Visit Skift to read more on this topic from Alex Mans, founder and CEO of FLYR Labs.
Join our team
FLYR is a team of industry experts passionate about improving the practice of Revenue Management.View roles
Subscribe for all the latest industry insights and updates from FLYR.