This is the first in a four-part series looking into how airlines manage the application of artificial intelligence in commercial functions.

Artificial intelligence and machine learning are said to be the future for many commercial operations – millions of daily decisions can be taken by machines based on Big Data, replacing often biased or limited human decision making.  

Of course, airline pricing departments have relied on such machine learning for decades, establishing optimized fares by flight by day across broad networks.

As airlines now explore new applications for Machine Learning – even broader applications in pricing or in commercial decision-making – they need to build upon learnings from history.

Many lessons from airline’s almost 50 years of experience with AI/ML

First, exploiting AI/ML is critical to long-term success. The first airlines to adopt sophisticated revenue management models are still the leaders today.  Airlines that were slow to innovate found their competitive positions untenable; many failed.  

"Automated systems need to be designed by commercial decision makers – including a clear understanding of the (quantifiable) objective, the necessary data inputs, and assumptions inherent in the chosen model."

Today, virtually all airlines apply some version of revenue management; large RM suppliers offer a menu of alternatives to meet the needs of airlines of all sizes.  

New innovative applications for AI/ML are already beginning even at some of the smallest carriers today.

Second, the automated systems need to be designed by commercial decision makers – including a clear understanding of the (quantifiable) objective, the necessary data inputs, and assumptions inherent in the chosen model.  

Third, an ongoing process needs to be created to actively manage the models and to fill in its known gaps.

Model Design – The Commercial Role

Data Scientists are probably best equipped to select the appropriate model for a given application.

But they have a lesser role in setting the objective, in deciding on the inputs, and on agreeing to the model assumptions. Those functions must be overseen by the commercial decision-makers.

Let’s start with the objective. The objective of airline pricing has changed over the past fifty years. Initially, airlines were satisfied with using fare rules to segment travelers – lower fares were non-refundable and needed to be purchased 21 days in advance of the flight date.

The principal objective was to use lower fares to fill more seats while retaining the high fare business passenger.  

"The input initially was solely booking history by fare level. Often tapping into the new benefits of AI/ML, inputs now may include competitive fares, searches, and even weather."

Then, airlines adopted inventory controls, still with the objective of maximizing revenue on a flight-specific basis.

Then, as airlines recognized that hub networks require a more coordinated approach across connecting flights, airlines adopted O&D revenue management to maximize revenue across large systems.

The input initially was solely booking history by fare level. Often tapping into the new benefits of AI/ML, inputs now may include competitive fares, searches, and even weather.

Model assumptions, of course, are both critical to identify and sometimes not always obvious.

They are too often buried in the algorithms. Traditionally, for example, RM models assume each “bucket” of demand on each flight can be forecast independent of the others – an assumption most users know is wrong but is generally recognized as a simplifying assumption to make the models more usable.  

Even today, most interventions apply across such supposedly “independent” subsets of demand, effectively retaining whatever errors occur because of that assumption.

Model Management – The Commercial Role

Airlines employ teams of pricing and revenue management analysts to manage the existing systems. There is no notion that the models eliminate analyst headcount or that they can run themselves without intervention.  

Analysts monitor trends in demand and model variances versus actuals, interceding when the model appears too slow to adapt.  

They also adjust demand manually when there are extraneous events – including conferences and holidays that move across the calendar, and, in the case of Covid, country-specific outbreaks.

"Airlines employ teams of pricing and revenue management analysts to manage the existing systems. There is no notion that the models eliminate analyst headcount or that they can run themselves without intervention."

AI/ML is an exciting technological advance for many industries; it has applications in operations, marketing, and customer engagement as well as new opportunities in pricing.  

All airlines need to explore potential uses, especially in customer engagement and pricing -- ultimately, as with RM itself, AI/ML will be found to be essential.  

Also as has been shown by the use of AI/ML in RM over decades now, AI/ML requires the hands-on involvement of commercial leaders in both the design of the systems and in ongoing system management.

Tom Bacon is an airline industry consultant based in Denver, Colorado. With 30 years experience in a variety of travel companies and business situations, he has a track record of dramatically improving profitability through innovative revenue strategies.