<img alt="" src="https://secure.ruth8badb.com/159247.png" style="display:none;">

The-Overbooking-Debate--A-Step-Backwards-for-Airline-Revenue-Management-Blog.png

There have been two major elements of airline revenue management from when the concept was initially introduced in the 1970’s: inventory controls and overbooking. Together, these are calculated to drive over 5% more revenue and are now considered a fundamental practice for airline revenue management. Both remain long-standing examples of airlines’ leadership in the application of big data analytics to drive improved profitability, and both rely on a combination of forecast algorithms and optimization models.

With the recent focus on overbooking in the wake of the United Airlines incident, many airlines are naturally reviewing their overbooking policies. Delta Airlines and United Airlines have both announced that they will increase the maximum compensation to passengers that give up seats on overbooked flights to almost $10,000. Southwest Airlines, however, had a different reaction – they announced that they would phase out overbooking completely. Southwest Airline’s is a stunning response that is both unnecessary and misguided. But in another sense, the reaction is quite predictable. The decision clearly reflects two challenges of any big data analytics application.

1. No Big Data Analytics Application Can Be Siloed 

Although overbooking practices depend on statistical analysis of no-show rates by flight, by date, by fare type – potentially millions of data points – in the end, success depends upon procedures at the airport on the day of departure. The statistical breakdown will always lead to a small chance of more passengers showing up than there are seats. Thus, airlines need a corresponding airport plan for managing over-sales (overbooking is not just big data optimization process at headquarters).

Delta Airlines understands this very well and is now considered a leader in how it manages over-sales at the airport. In fact, they report among the lowest denied boarding rates in the industry – 99% of over-sales end up being passengers who voluntarily, sometimes enthusiastically, choose another flight in exchange for a flight coupon. Southwest Airlines has among the highest rates of denied boarding, implying less well-developed airport procedures for handling over-sales.

2. Big Data Analytics Must Be Communicated Efficiently Across the Airline

The second challenge for Big Data Analytics is communication of the strategy across, and up, the organization. Overbooking, and other such model-based optimization routines, must be understood and embraced even if they are often the result of complex statistical models based on millions of calculations.

In the case of overbooking, no matter how sophisticated airport procedures are, over-sales is extra work for airport agents who are already tested by late flights, quick turns, missed connections, irritable passengers, disruptive weather, and so on. Unless they see a clear benefit, if given a choice, they would certainly vote to eliminate overbooking. On the other hand, if they believed that overbooking adds 2% revenue – equivalent to 20-30% of their profit sharing check – perhaps they would be more open.  Training airport staff in the benefit of overbooking is a standard responsibility of an airline’s revenue management.

Similarly, the benefits and costs of overbooking can be tracked on a regular basis. Higher over-sales on a flight, or in a market, need to be monitored and reflected in overbooking calculations going forward. More denied boardings at a specific airport, too, may reflect a need for revised airport training or operations.

As part of a corporate statement, Southwest Airline said: “As we have dramatically improved our forecasting tools and techniques, …, we no longer have a need to overbook as part of the revenue management inventory process.”

This is a misstatement of airline forecasting. Airlines can and should forecast no-shows on a flight basis, reflecting different behaviors for flights in different markets and on different dates. But no-shows are the result of a myriad of factors, including traffic jams, oversleeping, missed connections, weather, early or late business meetings, and so on, many of which do not show up until the day of the flight. So, instead of trying to forecast one number, airlines forecast a probability distribution of outcomes – the actual no-show behavior for an individual flight two months out is, of course, unknowable. The fact that there is, on average, 1 no show, is irrelevant in overbooking; the models calculate the probability of 1, 2, 3 or more no-shows by flight and recommend overbooking based on the probability distribution and the cost/benefit of each incremental passenger booked. 

For Southwest Airlines to imply that they have replaced the distribution with an accurate point forecast doesn’t match actual customer no-show behavior. And, of course, if they did have a more accurate forecast they could overbook more confidently – not eliminate it. Potentially, the airline has adopted a simplified story around overbooking – that everyone can understand -- that is now leading them to opt out of overbooking.

Conclusion

Overbooking is an early application of big data analytics yet airlines, like other industries, are finding more and more applications for such forecast/optimization models, in marketing, customer service, and operations. Southwest Airline’s announcement reminds us that airlines may wish to proceed somewhat cautiously in implementation of these models – by which each model needs to have a corresponding operational plan and given the benefit/cost trade-off inherent in these strategies, there needs to be clear understanding and buy-in across and up the organization.


Interested in learning more about Mercator’s revenue management solution for airlines?

Learn more