Rescheduling flights and reassigning passengers during a weather disruption. Predicting and preventing mechanical failures. Forecasting fare variations in the months and days ahead of a flight. In the world of air travel, what do these endeavors have in common? All require complex analysis of massive “big data” sets — work that humans, on their own, are often ill-equipped to handle.
Enter artificial intelligence and machine learning. While earlier technologies provided some means of solving complicated problems, AI and machine learning elevate that capability to an exponential degree. When we give machine learning algorithms a set of parameters relevant to a problem — like a weather forecast or fare histories — the algorithms will run through all possible combinations with unprecedented speed to deliver optimal solutions.
Recent advancements in this technology couldn’t come at a better time. The data created and copied annually in the world is doubling in size every 12 months, according to a report by Deloitte, and is expected to reach 44 zettabytes by 2020. Commercial airlines are major contributors to this growing universe of data, and they, along with the millions of travelers they serve each year, promise to be major beneficiaries of machine learning. Let’s take a closer look at how that will happen.
Planning for and Responding to Travel Disruptions
When bad weather or other major events force the delays and cancellations of flights arriving or departing a particular region, it creates a lengthy domino effect that reaches far beyond that region’s airports. Because an aircraft and crew are typically scheduled to serve multiple destinations, individual delays or cancellations at one airport often cause more delays and cancellations at others. In addition, federal rules governing tarmac times, as well as airlines’ own commitment to keeping customers from waiting too long on planes before and after takeoff, can lead carriers to make preemptive cancellations.
Machine learning can help alleviate these problems. For instance, it can help to analyze both real-time data and archived data can yield predictions days or months in advance on which routes will avoid weather delays and when a route besieged by bad weather will become clear again. Such information can allow more informed planning by airlines and fewer cancellations. A model developed by MIT researchers would have opened up 13 percent more routes in a given scenario. Hamsa Balakrishnan, the leader of the MIT research team, is also the chief scientist at Resilient Ops, a Boston-based startup focused on proactive disruption management. The company uses algorithms that take into account variables such as weather forecasts, historical weather data, and historical flight performance. “The true magic comes in when algorithms that learn from history can predict the future based on what’s happening right now,” Balakrishnan tells us. “You can start thinking about what’s likely to happen and deal with it proactively.”
Dynamic Pricing and Personalized Service
For years, airlines have sought to find the best middle ground for ticket fares, setting prices low enough to attract consumers and high enough to protect profit margins. But that sweet spot can vary based on the day, the season, the route in question, and more. Here again, machine learning can provide answers. The San Francisco-based startup FLYR, for instance, has developed models to analyze years’ worth of airfare records and other data to help airlines develop dynamic pricing, determining the optimal fares for each route, day and season. FLYR, a JetBlue Ventures portfolio company, also offers consumer products, using price prediction models to create what could be described as insurance for fare volatility — its FareKeep product allows customers to be reimbursed when a fare drops after booking. “We develop derivative products that allow travelers to gain peace of mind,” says FLYR CEO Jean Tripier.
Travelers can also gain peace of mind when they believe they’re getting the best value and most-personalized service. Machine learning can help ensure that, too, even if an airline knows little about a customer considering whether to initiate a transaction. Algorithms can crunch historical data about travelers with similar profiles and then provide personalized flight recommendations based on that analysis. In the future, airlines and other travel providers may even offer bundles, such as an airline ticket paired with specific in-flight amenities that are ideally suited for specific customers and use cases. Such bundles may also exclude amenities that analysis has shown is of no use to particular travelers, thereby reducing waste.
Customers’ ticket orders and other purchases, such as in-flight movies and duty-free items, can also offer airlines more insight into how to best serve them. Machine learning algorithms can use this data to predict spending patterns for various travelers. That’s exactly what the information technology company Unisys did, using a client airline’s big data to forecast who would be most likely to purchase ticket and other air travel-related products in coming months. Such analysis can help airlines engage with customers and potential customers using offers that are more likely to be of interest and value to them.
Optimizing Equipment Performance: Predictive Maintenance and Supply Management
In the Internet of Things age, sensors are continually providing data indicating the condition and performance of various machines and equipment. Globally, airline fleets are projected to generate 98 billion gigabytes of data per year by 2026, according to management consulting firm Oliver Wyman. What can manufacturers and airlines do with this data? With machine learning — and the help of physicists, engineers and other experts — they can identify patterns in the data that indicate what sort of conditions or performance declines might indicate a breakdown is imminent. Armed with that information, they can engage in predictive maintenance, taking preventative actions that avoid the significant downtime (and flight delays) associated with major repairs. Machine learning can also address another source of inefficiency: unnecessarily frequent maintenance that causes more downtime. While conventional wisdom may suggest it’s best to service parts often to avoid problems, schedules optimized by algorithms will mean that maintenance happens when it’s actually needed.
Machine learning can influence supply management as well. Through machine learning, airlines can perform analysis on the parts they use and find that lifetimes differ from those that manufacturers estimate. Today, manufacturers’ projections are based on data from a variety of customers working under radically different conditions, leading to wider discrepancies from actual usage. Carriers may be able to use parts longer than recommended by manufacturers, reducing costs while also avoiding the risks associated with installing new, less-tested parts. “If you don’t have to scrap something at a certain cycle or hour limit, you can continue to use it,” says airline industry analyst Bob Mann. “Then you don’t need to buy a new one or need so many spares. You start to save real money.”
Today, we’re getting just a small glimpse of what AI and machine learning can do in air travel. As companies large and small develop better algorithms and other technology, weather disruptions will become more manageable, aircraft will remain in the sky longer and spend less time under repair, prices will meet both carriers’ and consumers’ needs, and customers will have better access to the experiences they really want. For all the complex calculations and sophisticated models that make up machine learning, its impact on air travel is easy to explain: It will result in a healthier, more robust industry that helps everyone reach new heights.
