American Airlines revolutionizes airport gating with machine learning

Many travelers have experienced the excitement of their flight arriving early and started gathering their belongings, only to be told that their gate is not available, and they’ll have to wait on the tarmac. Such gating issues can be frustrating for customers and inefficient for airports and airlines. American Airlines, one of the world’s largest, set out to address this very problem using cutting-edge technologies. According to Sumit Batra, managing director of operations planning and performance at American Airlines, the results revolutionized American’s two largest hubs at Dallas-Fort Worth (DFW) and Charlotte (CLT). “We shared a bit of magic this past year in benefits to our team members, our customers, and to the environment,” says Batra.¹

American organizes several hundred departures and arrivals each day at its major hubs. Each flight must be matched to a gate based on aircraft type, runway concentration, taxi times, and other operational metrics. For decades, this gate assignment process typically required four hours of human labor from 10 p.m. to 2 a.m. to drag and drop flights to gates on a legacy system. Yet, the next day, if flights were delayed or otherwise shifted, the entire process could restart. “This was a really labor-intensive process that wasn’t living up to how math, people, and technology should be working together in the 21st century,” says Batra.

Instead of replacing American’s gating technology, which could’ve taken years, Batra’s team designed and delivered within six months a machine learning (ML) engine that could be embedded on top of the legacy system. Data scientists spent time shadowing gate planners to understand their needs, use of technology, and operations lingo. The data scientists then made sure to build in transparency, so that experienced staff could review and verify a report of gate changes whenever the engine was run. Crucially, according to Anne Moroni, vice president of operations planning and performance at American Airlines, “The mathematics of the engine was customized to fit DFW and CLT individually because the inputs varied greatly.”² The result: Gate planning time was cut from four hours to just 2.5 minutes.

Moroni believes American’s approach to building on top of the legacy system also drastically reduced the need for change management—team members quickly bought into using the ML engine. “We showed operations staff the value of repetitive task automation and they were immediately excited about freeing up their time,” she says. With the ML tool addressing daily planning, operations teams focused on higher-order tasks that required a human touch, such as managing real-time surprises and improving the customer experience. In fact, since implementing the new tool, DFW has seen record lows in missed flight connections.

Even with just two airports running the new tool, American has already reduced its annual fuel consumption by 1 million gallons, the equivalent of removing 2,000 cars from the road. As Batra’s team expands the technology to more airports, it hopes to continue realizing environmental impact alongside improvements in efficiency and customer satisfaction. Says Batra, “Automation is addictive. Once people are accustomed to higher-order work and better results, they don’t want to go backward.”