Blog_Header-4
Blog_Banner_v3

How Data Science Helps Ease the Pain of Flight Cancelations

Posted by Chunping Wang on Wed, Sep 03, 2014

OS SignalCentral Airline Cancellations OPT A GP 2014 V1.0We’ve all been there. That place in the airport where you’ve just learned your flight has been canceled. If the weather’s bad, you can understand and roll with it. But if it seems arbitrary and royally screws up your plans — well, that’s another story. As a customer, you wonder, “Why do airlines do this? And how do they decide that my flight is canceled when others are not?” But airlines are asking their own questions, namely: “We have to cancel X number of flights, but which flights should we cancel to minimize the loss of revenue and customer loyalty?” Here, we delve into both sides of the issue, and the answers should provide a little context — and hopefully quell the frustration for everyone.

Nobody Likes Flight Cancelations

Let’s start with the premise that the airlines don’t want to cancel flights. It costs them time and money and disrupts the intricate schedules of crews and planes, all of which have somewhere they need to be — just like you. In addition to making payments on the planes (many of which are leased), airlines spend money repositioning aircraft and crew that are displaced because of a cancelation. And then, of course, there are the passengers. For a large weather delay, airlines may face thousands of stranded travelers, putting additional airport and call center staff on duty to provide assistance. Airlines are also giving up revenue as they cancel passengers’ flights, and the net impact depends heavily on how much of that revenue they can recapture by rebooking customers to other flights within the airline.

Other than direct financial losses, passengers’ anger and frustration may cause potential future losses. After getting stuck overnight at the airport or missing an important meeting or vacation, customers may write off an airline forever. Such potential losses of customers’ loyalty are even more difficult to predict than the actual losses at the time of cancelation.

Often flights need to be canceled for reasons beyond airlines’ control, such as when unfavorable weather conditions or issues pertaining to air traffic control reduce the number of planes that can take off or land each hour. And even for extreme storms, usually the situation becomes worse gradually. Before the situation gets so severe that the airport has to be closed, airlines still need to decide which flights to cancel. Another possible scenario is that a plane for an important route (e.g., a long-haul international route) could not fly due to a mechanical problem. In that case, the airline may need to cancel a different flight and have that airplane take the long international flight. Finally, flight crews that max out on their hours can also force airlines to cancel flights, and it’s often necessary to shift crew around to minimize the fallout.

Making these decisions is not easy, and airline management does not take them lightly. When facing the decision to cancel a flight, senior management has been known to debate the issue for 30-minutes or more in an effort to mitigate the pain for both passengers and themselves. In the end, it’s hard to tell who was right, as the ripple effects can last for years in the hearts of some customers.

Enter Big Data science. With the help of data science, they can now better determine which flights will cost less at the point of cancelation and for the long term. This is done by defining the Flight Value Score, which incorporates Signals — or, in this case, lost revenue forecasts— about the flight to measure the value of the flight in the event of a cancelation. With the Flight Value Score, airline management has a much more straightforward metric to use, enabling them to make the cancelation decision in seconds, not minutes. More important, it helps ensure the least amount of passenger disruptions as possible, which is good news for everyone.  

Flight Value Score to the Rescue

Many factors go into determining the Flight Value Score. Surprisingly, the actual dollar amount lost is not the best metric to use. For starters, attaining the cost in dollars is imprecise because airlines often don’t have all the necessary data (the cost of the plane is usually unknown, for instance, though we can estimate it based on size and model) and because so many elements need to be factored in. To calculate this cost in dollars, airlines must account for the constant cost of the aircraft (lease or payment), aircraft and crew repositioning, passenger rebooking, and passenger assistance, which could expand to multiple departments. Even if we simulate the cancelation cost in great detail for all the flights that could potentially be canceled, the margin of error is still relatively large. Second, they cannot accurately measure the cost of losing customers’ loyalty in dollars over the long term.

The Flight Value Score assigns a score to each flight, based on delay time. We define the value of a flight cancelation as the total passenger delay (in minutes) per canceled flight. We make the assumption that all the passengers will be rebooked onto the subsequent flights with the same destination, and the delay for each flight is the difference between the original arrival time and the assumed rebooked arrival time. For example, if the flight was scheduled to arrive at 10:10am and a passenger is assumed to be rebooked on a subsequent flight arriving at 1:20pm, then the delay for the passenger is 190 minutes. The total minutes of delay for all the passengers booked on the flight reflects the Flight Value Score. Airlines are advised to cancel flights with less total delay time, or a lower Flight Value Score.

While the Flight Value Score is mainly a reflection of delay time, there’s actually a lot more complexity that goes into it. For instance, it should also incorporate missed connections, length and/or purpose (business or leisure) of the trip, passengers that require extra assistance, passengers' membership and engagement history, crew with subsequent duties, and so forth. To keep with the numerical value, we weigh each minute of delay accordingly.

Total delay time is a good measure because it closely correlates with other factors that add to both direct costs and unpredictable potential losses. For instance, a flight with a high delay time typically indicates a larger plane. This fact alone indicates a higher monthly payment on the plane, as well as either a larger number of passengers or an especially difficult rebooking process or both. All of these factors indicate a higher risk of future revenue losses. It’s also a good measure because it aligns the airline’s goals with the customers’ goals: getting customers to their destinations as quickly as possible. By using the Flight Value Score, airlines are able to save the most passengers’ time and protect the most customers’ loyalty.

Conclusion

The Flight Value Score is a powerful tool that allows airlines to make cancelation decisions based on a numerical value instead of assumptions, projections, and guestimates. More important, the Flight Value Score drastically reduces the debate and time required to make these decisions — time that could be spent getting customers onto rebooked flights.

We work with some of the world’s biggest airlines and are constantly expanding the scope of how we use Big Data science to improve everything from personalized marketing to flight demand forecasting. Download our case study to learn more, or contact us at  signalhub@operasolutions.com to learn more.

 

Download Now

 

You can also download the Amadeus report, “At the Big Data Crossroads: turning towards a smarter travel experience.” And you can read about it in "Why the Airline Industry Needs Another Data Revolution" in The New York Times.

Chunping Wang is a senior scientist at Opera Solutions. Based in Shanghai, she is focused on the research efforts of Opera Solutions’ Customer Signal Hub for airlines. Chunping holds a Ph.D. in machine learning from Duke University.


 

Topics: Data Science, Signal Hub Technologies, Marketing