Airline Industry: How Legacy Systems Fall Short


In my last post, I touched on the various reasons airlines are increasingly facing irregular operations impacting their business, and consequently their profit. We talked about the fact that the COVID-19 virus was a primary catalyst and ongoing cause of a lot of this irregularity and disruption, and how further changes to fare rules exacerbated the issue and legacy RM System’s ability to accurately predict demand. However, as COVID rates recede, the need to account for these events (and their further use as historical data going forward) remains.

How Airline Systems Fall Short

Airlines were one of the first industries to adopt revenue management technologies, beginning with inventory control/yield management systems in the early 1990s. These systems were cutting edge for their time and revolutionized the industry and its approach to availability and pricing. While these applications have been updated, modified, and replaced over time, many of the legacy systems supporting the commercial decisions of airlines today are under-powered and limited in their analytical flexibility compared with more advanced AI and machine learning models today. Four key factors limit the effectiveness of revenue systems in airlines.

Post-Covid Forecasting is Difficult

Forecasting is a difficult exercise in the best of circumstances, but the volatility and industry-level change of the last two years has been unprecedented in the airline industry. Changes in business and leisure demand patterns have fundamentally broken some decades-old assumptions in the industry, and legacy RM systems are unable to adjust to the change.

Dependable business travel, which has long subsidized cheap leisure fares, has shifted dramatically, and a profitable passenger mix is more difficult to maintain. These changes in demand, combined with the relaxing of rate restrictions and change fees mentioned in the first part of this article, have resulted in a less dependable base of demand upon which to build a forecast. Legacy systems are struggling to build an initial forecast based on this data, and are incapable of adjusting quickly and effectively when things do change without significant manual intervention and close monitoring.

Manual Intervention is Taking Too Long

While most legacy RM systems offer some form of manual intervention to override/correct issues with system outputs, many of these are highly manual and ineffective. Overrides typically must be managed at the day level, resulting in a great deal of manual work to override the effects of – say – a two-year global pandemic.

This level of detailed management by exception is not adequate to handle the thousands of overrides required with such a significant and unexpected event. Worse yet, many of these dated systems are without AI or Machine Learning capabilities, and these overrides may be lost, or diluted by the ongoing functioning of the system that is incapable of learning from user intervention. This is a challenge with most systems built prior to the last 5-10 years, when advances in AI and Machine Learning have revolutionized the areas of predictive intelligence on which RM systems rely.

Legacy Systems are Built on Decades-Old Models

While advanced for their time, most legacy RM systems are too old to have taken advantage of the latest in machine learning and cloud computing. These systems do an effective job of solving a big optimization problem, leveraging large amounts of historical data via their algorithms, but they’re typically resource intensive, lack flexibility and scalability, and are not modularized – making enhancements a significant and costly undertaking. Moreover most of the forecasting/predictive models are purely statistical. They learn slowly and can take years – even several years – to adapt to a new norm. They are not based on true Machine Learning and cross-validation techniques that incorporate a variety of models, automatically adjust to changing conditions, and can be tuned to adapt quickly (days or a few weeks) to changes in the network, schedule and in passenger behavior.

These systems are built on not only old modeling and technological infrastructure, but also based on outdated business assumptions and largely irrelevant historical data. These simplistic views of demand patterns and business vs. leisure segmentation are hard-coded into the backend of these systems, making them a poor match for the market as it operates today and unable to take advantage of the improved prediction offered by AI/ML-based capabilities.

In the next post, we’ll dig into what airlines can do to make better commercial decisions even in the volatile and unpredictable markets we currently face, using Artificial Intelligence and Machine Learning.

Contact us to learn more.