While we are still navigating through the COVID-19 pandemic, it may seem untimely to start thinking about the future.

Yet the challenges from the pandemic caused us to reflect on the very foundation of airline revenue management systems (RMSs) – their use of historical data as a basis for predicting future behavior.

In my blog, Rethinking airline revenue management forecasting in times of change, I explained how the COVID-19 pandemic represents a unique challenge for RMSs. Canceled flights and abrupt changes in customer demand mean that the content of a RMS’s historical database has become obsolete and we can no longer rely on it to predict the future.

To address this challenge, we need a forecasting model that offers adaptiveness and stability at the same time. These properties lie at opposite ends, but I believe we have found a beautiful solution to this problem through what we call ‘Active Forecast Adjustment’.

What I will address is a more profound limitation of the RMS’s historical database – its incompleteness – and how we can address this limitation through the design of a competitor-aware revenue management system (CARMS) that can adapt to changes in the competitive landscape.

A traditional RMS’ view of the world

RMSs today base their future price decisions largely on internal historical data. Because these data inputs are limited to only the historical observations of the airline’s own flights, the RMS simplifies the customer decision process into a single question: does the customer fly with our airline, or not? This does not mean that the RMS ignores the effect of competition – RMSs do account for competition, but implicitly, and without capturing the complexity of the actual competitive situation.

For example, a customer searching for a trip must select among potentially hundreds of itineraries across multiple airlines, each with its own price and product characteristics. If the customer decides to buy with the airline, a traditional RMS only observes the booking, and is aware of neither the distribution channel nor the other itineraries available to the customer. The RMS therefore makes the simplified assumption that demand for that itinerary only depends on the airline’s own price.

So, at the very outset, an issue with traditional RMSs is that they do not directly observe the competition. RMSs, therefore, make an implicit assumption – that competition remains unchanged compared to last year.

This simplification worked well in a stable airline business environment. However, the current business environment – characterized by fast-changing schedules, fare products, and prices further disrupted by the COVID-19 pandemic – makes this simplification inadequate. In this environment, the RMS’s historical data can no longer represent the future.

Towards a competitor-aware revenue management system for airlines

Over the last decade, RMSs have become increasingly sophisticated in how they model customer choice. These ‘choice-based’ RMSs can model the probability that a customer will buy-up to a higher price point, as well as the probability they will buy each branded fare or fare family the airline offers for the same itinerary. However, no choice-based RMS to date has explicitly incorporated in its core scientific model the effect of other airline competition on customer choice, even though competition can have a significant impact on demand. Why is that? The answer is simple – due to a formidable complexity.

Imagine the system requirements for building a competitor-aware revenue management system that considers the prices offered by other airlines. Such a system would require a choice model of how customers make choices amongst the itineraries offered by all airlines in the marketplace, similar to the models used today in dynamic pricing.

However, since RMSs optimize prices by considering future demand, a competitor-aware RMS would also have to predict what prices competitors will charge in the future for each itinerary, departure date, and day of departure. This goes far beyond what is required for a dynamic pricing system, which requires only information about current market prices.

If this was not complicated enough, a competitor-aware RMS would also have to predict how each airline will respond to changes in all the other airlines’ prices. Depending on the market and the context, the airlines may not react or be highly-reactive to competitor price changes — such response behavior would have to be incorporated into the system. And finally, the system would need to collect comprehensive price information from all airlines in the marketplace at the traffic flow level to make these predictions.

Despite these challenges, our research team has figured out how to put together the building blocks of a competitor-aware RMS. As we presented at the Airline Group of the International Federation of Operational Research Societies (AGIFORS) conference in 2019, it is possible to extend the dynamic programming optimization formulation from the traditional RMS to include a customer choice model that includes the demand for our airline given all airlines’ prices. Furthermore, we demonstrated that we can predict future competitor prices with remarkable accuracy. By making assumptions about other airlines’ price-responses – which could vary by market or competitor – we can determine the optimal price from our airline’s point of view.

Based on benchmarking studies, we found that our competitor-aware RMS prototype produced a revenue increase of 2.0% - 4.6% versus a traditional RMS, depending on the degree of competitor responsiveness. These revenue benefits are separate from dynamic pricing and can be traced to:

• Unbiased willingness to pay estimates: Unlike traditional RMSs, which are unaware of the prices and products offered by competitors, our model corrects for the inherent correlations between the airline’s prices and its competitors’ prices.

• Anticipating competitors’ future prices: By estimating the total market demand, the impact of the competitors’ expected prices can be factored in when determining the airline’s optimal prices.

• Anticipating competitors’ price responses: Understanding competitor price responses allows the airline to look ahead and account for the possible reactions that its price will trigger in the marketplace. We’ve found that informing our model about anticipated competitor price response behavior acts as a stabilizer – discouraging price spiraling down.

Even though a competitor-aware RMS provides a forecast of future demand in the presence of competition, there is still additional value to be gained by combining it with dynamic pricing, which capitalizes on a real-time view of competitor prices in each shopping session.

Looking forward to the future of airline revenue management

The COVID-19 pandemic has caused us to reflect on the content of the RMS’s historical database. Does it adequately represent the future? We have seen how the historical database can be incomplete as well as become obsolete. The current business environment has exacerbated both these deficiencies.

However, we believe that with new revenue management techniques, such as a competitor-aware revenue management system (which addresses incompleteness) and Active Forecast Adjustment (which addresses obsolescence), we will be able to overcome these deficiencies and bring new solutions to life that add value to airlines.

References:

• Fiig et al. (2016) Dynamic pricing – The next revolution in RM? JRPM
• Fiig et al. (2018) Dynamic pricing of airline offers, JRPM
• Fiig, Wittman, Trescases, (2019) Towards a Competitor-Aware RMS, AGIFORS, Panama
• Wittman, M.D., 2018. Dynamic pricing mechanisms for the airline industry: theory, heuristics, and implications. Ph.D. Dissertation, MIT.

Thomas Fiig - Director and Chief Scientist, Airlines R&D, Amadeus

www.amadeus.com