Guest columnist Tom Bacon argues that machine learning could address some of the current limitations in airline pricing

Could it be that pricing opportunities, despite sophisticated airline RM systems, are still not being exploited fully by most airlines?

That is the view of an airline revenue management specialist in the Middle East, who argues that many airlines have downplayed fare rules as a vehicle for price discrimination. Instead, they favour increased reliance on forecasting and inventory controls.

So, rather than using rules - Saturday night stays, for example – as price ‘fences’ to facilitate lower fares for more price sensitive leisure passengers, airlines have come to rely more heavily on inventory controls; airlines limit the number of seats they sell to lowest fare passengers based on forecasts of higher fare demand. To his point, with the growth of low cost carriers, most airlines no longer file low fares tied to Saturday night stays. In many markets, airline fares are said to be ‘fenceless’.

Pricing opportunities are still not exploited fully by most airlines

Besides Saturday night stays, other requirements for lower fares might be a round-trip versus one-way tickets, longer length of stays, certain points of origin, advanced purchase requirements, purchase during sale periods, purchase on the airline website, and so on. Another type of fare rule that used to be exploited more frequently is tied to customer profiles, for example, student fares.

The purpose of such rules, of course, is to ensure that only the most price sensitive travellers qualify. The objective is to drive less price sensitive business passengers to higher fares. Airlines still have pricing departments, which perform the associated segmentation analysis: the pricers perform extensive analysis of stimulation versus dilution. Questions that are asked include:

  • How much of a price discount is warranted?
  • How many more leisure passengers will be attracted? 
  • How many business passengers will buy-down?
  • How will competitors respond? 

Then, after the analysis is complete, the pricer manually files the fares, along with the rules, in specified formats (‘footnotes’) in the industry’s pricing database. What this means is that the fares will be available across travel industry search sites.

This process has a number of limitations:

  • Often the rules are very high level (broad rules across a broad definition of ‘markets’)
  • As such, the analysis, too, is high level with macro estimates of stimulation and dilution
  • After the analysis is complete and the fares filed, they are not kept up to date (the rules and fares are fixed as filed; there is no automated update)
  • Pricers need to monitor the associated base fares and change the whole structure when the base fares change or when competitive pricing warrants changes
  • Filing is cumbersome and arcane. Some airlines outsource the filing to both manage the volume and to tap into necessary specialised expertise.

Given these limitations, it is natural that airlines have moved away from investing heavily in pricing analysis and filing special fares tied to rules. After all, inventory management is, in fact, highly dynamic, as well as being market and flight-specific. With inventory management, there are very few seats available for lower fare passengers when demand for higher fares is projected to be high – and inventory levels by flight are automatically adjusted every night.

Enter machine learning

There is, however, a potential new approach that addresses each of the current limitations with fare rules. This being, to apply machine learning in order to identify opportunities for fare discounts or premiums and to implement the fare adjustments on a real-time basis with business rules instead of fixed fare filings. This is one version of the new ‘dynamic pricing’, as promoted by Amadeus and select other airline revenue management suppliers.

Using machine learning, airlines could identify highly granular opportunities that promise more revenue

A new machine learning approach, for example, could sort through a database of searches and bookings after a macro fare increase. What traveller characteristics, for example, were associated with changes in conversion rates? The system could identify those search characteristics - or combination of features - that are most sensitive to the fare increase on an origin & destination or flight-specific basis.

  • Machine learning (ML) may find that the most price sensitive customers are associated with an unforeseen combination of factors that are not used by airline pricing departments today. Certainly, it could factor in traditional elements more discretely by flight and by market (group size, length of stay, days advanced purchase) but it could also incorporate ancillary purchase propensities or temperatures at the origin and destination and other less commonly used customer or market characteristics.

Of course, the system could be constantly evaluating conversion rates based on changes in fares across markets. At the same time, it could constantly seek robust opportunities for use of granular business rules to drive small fare premiums or discounts for highly targeted customer segments, based on the search criteria. The system, in fact, should also constantly test small price changes for certain types of searches to identify new opportunities or changing rules and recommended fares.

Similarly, the system could evaluate price sensitivity versus competitive fares across customer segments. Using machine learning, airlines could identify highly granular opportunities that promise more revenue – opportunities that are more robust than what is typical for a more macro fare rule. In fact, as a supplement to inventory controls, it would tend to avoid implementation of fare premiums that result in less overall demand in any subset.

This type of dynamic pricing would complement the sophisticated demand forecasting and inventory optimization systems currently used by most airlines. At each fare level, it would identify small premiums or discounts for sub-sets of passengers that would drive incremental revenue above that which would be available via inventory controls (or existing, more macro fare rules) alone. New dynamic pricing, based on machine-learning, could introduce a new form of fare rules to the industry. Airlines would use dynamic pricing via machine-learning along with their modern sophisticated inventory management systems to drive even more revenue.

[Note: This form of fare rules represents a huge paradigm shift for the industry and for travel distribution systems.  Implementation would require changes in pricing and inventory management at airlines, but also major changes in travel distribution systems.]

Tom Bacon has been in the business 25 years, as an airline veteran and now industry consultant in revenue optimisation. He leads audit teams for airline commercial activities including revenue management, scheduling and fleet planning. Questions? Email Tom or visit his website

EyeforTravel San Francisco 2018

April 2018, San Francisco

Related Reads

comments powered by Disqus