AI: how it’s delivering sharper route planning

Planning a passenger aircraft’s route has historically been a manual process for the specialists in airline operations centers. Now some dispatchers have a new tool: Artificial intelligence software that can rifle through weather, flight congestion and other data faster than a person can. This could be a timesaver, not just for dispatchers but for the flying public. Karen Kwon tells the story.

A few months ago, I hopped on an American Airlines flight from Memphis, Tennessee, to Washington, D.C., that was scheduled to depart at 3:55 p.m. Once everyone was aboard, the captain announced that our departure would be delayed, and he offered an explanation: We’d have to divert around a storm brewing over Nashville, and that meant we needed more fuel.

As we sat on the tarmac waiting for the fueling to be finished, I started to wonder if this delay could have been avoided. 

It turns out that entrepreneurs in the artificial intelligence field have had scenarios like this one in mind for several years. As the technology is introduced into airline operations centers, the large rooms where dispatchers plan routes, it could reduce delays and missed connections while also making a dent in the carbon footprint of flight. 

This is the story of how Alaska Airlines began working with Airspace Intelligence, a San Francisco-based startup with an office in Gdansk, Poland, to become the pioneer in the field. The story is based on inquiries to three airlines, plus interviews with executives at Alaska Airlines and Airspace Intelligence, and an analyst who watches the industry closely. Under a licensing agreement, Alaska Airlines dispatchers have been using Airspace Intelligence’s Flyways software for two years now. The deployment has shortened average flight times, reduced fuel usage and carbon emissions, and contributed to its environmental sustainability goal, the airline says.

Reinventing the dispatching process

The three founders of Airspace Intelligence, Phillip Buckendorf, Kris Dorosz and Lucas Kukielka — one German and two Polish immigrants — come from an autonomous driving startup that Buckendorf created in Palo Alto, California, in 2017. Working on the front line of an industry that had gotten a lot of attention — from media to venture capitalists to engineering talents — the founders could see that their field was getting crowded. This made them wonder if their specialty of using artificial intelligence to predict the future movements of other vehicles could be applied elsewhere in the transportation sector. So they started to explore how maritime and aviation were handling similar issues. Then in 2018, Alaska Airlines’ head of corporate development, Pasha Saleh, who knew one of the founders, invited the trio to Alaska Airlines’ operations control center in Seattle, officially known as the Network Operations Center, or NOC (pronounced “knock”). 

“We expected to see science fiction-like systems, like we know from movies,” says Buckendorf, Airspace Intelligence’s CEO. Instead, in Seattle and at the centers of other airlines they visited later, they saw dispatchers looking at old-fashioned IBM green screens, weather charts printed out on paper and operating software a decade or two old. The founders realized that there was a chance to innovate, which led to a decision to start a new company focused on aviation, Buckendorf says.

Creating a route requires a dispatcher to answer a host of questions such as: “What is the wind today?”, “What is the best altitude for this flight?” and “Is there any military training?” Before the Flyways software, the 100 or so dispatchers at the NOC had to find answers to these questions by visiting multiple websites. These included FAA websites designed specifically for dispatchers, but that information was available only as strings of text that were hard to read. 

Here’s how Saleh describes the pre-AI days: “You have a tab open for the Weather Channel, a tab for CNN” and so on, he says. “So it’s just click, click, click. If you look at a dispatcher, they have, like, 19 tabs open that they’re flipping [through].” 

A single dispatcher would typically be assigned about 20 flights to route, and manually assembled that information for each flight into a proposed flight plan for FAA. Airspace Intelligence believed it could modernize this archaic system. 

Having decided to focus on the aviation industry, the team started spending an obscene amount of time at the NOC in an effort to understand how dispatching works and to create a user-friendly product — one that a real dispatcher could seamlessly operate when under pressure. Alaska Airlines’ employees would joke that the team was basically camping in their operations center with sleeping bags, Buckendorf says. 

The attention paid off. After two years of intense development, Alaska Airlines agreed to try out the cloud-based software. The result of their efforts? During the airline’s six-month trial period that started in mid-2020, dispatchers accepted 32% of the suggestions made by Flyways. Alaska Airlines then agreed to license Airspace Intelligence’s proprietary software for a fee under a multiyear contract that began in January 2021.

Now, dispatchers no longer need to scour for data across multiple websites. Instead, the Flyways software funnels and displays the information for them. Plus, when a dispatcher is in the midst of planning a route on a computer screen, Flyways sends alerts about potential improvements. For example, the software could tell the dispatcher that by slightly changing the flight trajectory, the wind would be more favorable and the overall flight time could be reduced by seven minutes.

These suggestions are possible because of Flyways’ machine-learning approach, in which the software improves itself by recognizing patterns between the input data — including weather and air traffic congestion — and the previous decisions that human dispatchers made based on that input. Then, once the software receives new data, it comes up with possible new routes. 

Even though the software is trained on historical data, it often presents options that are different from what dispatchers might otherwise have considered. 

Humans have a tendency to stick with the familiar when planning a route between point A and point B. “But in reality, there’s really an infinite amount of options of how you can travel from point A to point B,” says Buckendorf.

Flyways improves itself further by learning from a human dispatcher’s acceptance or rejection of its recommendations. When the dispatcher dismisses a suggestion, Flyways asks why: Was it because of the weather? Was the route putting an airplane uncomfortably close to somewhere it shouldn’t be? The idea is that Flyways learns from those decisions and evolves — though certain data points need to be filtered out so that the software does not simply emulate human dispatchers’ choices, stifling innovation. 

“Now we’ve been using [Flyways] for over a year, the model is just getting better and better,” Saleh says.

Buckendorf underscores that humans remain in control. “The machine is really good at crunching huge amounts of data in an incredible fast amount of time,” he says. “What the human is really good at is judging the situation.” 

He suspects that this dynamic likely will not change for a long time.

This hybrid structure between humans and AI has an added bonus: With humans in the loop, the software is only providing assistance under an existing process, so there is no need for additional oversight by FAA regulators, says Buckendorf. 

Because route planning isn’t a mission-critical activity and the routes are reviewed by FAA, the use of AI probably draws less scrutiny from regulators than, for instance, if AI were applied to avionics, says Peng Wei, an engineering professor at George Washington University in Washington, D.C.

Buckendorf and Saleh say that neither company has plans for a fully autonomous version of Flyways. The goal of implementing AI isn’t to transfer a human job to a machine, Saleh says, especially when dispatchers are unionized. “We made sure from day one that the union realizes [Flyways] is not trying to remove dispatchers’ jobs,” he says. “It instead is a decision-support tool.”

Reducing carbon emissions

Alaska Airlines calculates that between January and September 2022, Flyways saved an average of 2.7 minutes per flight, meaning that the airline avoided 6,866 metric tons of carbon dioxide emissions.

The reduction is important, says Saleh, adding that Alaska Airlines turned to AI in large part to reduce its impact on climate change. When the airline’s board met two or three years ago, it decided to place environmental sustainability as Alaska Airlines’ top priority. From there, the board looked into ways that the airline could achieve its goal — such as using sustainable aviation fuel. Another item that came up during that discussion was improving operational efficiency.

Some of the effects are also felt directly in the NOC. Traditionally, even when two dispatchers were sitting right across from each other, one would not be aware of what the other is up to. For instance, if both were handling flights landing at, say, Boston’s Logan International Airport, they could inadvertently schedule the two flights to arrive at the same time, creating a conflict for local air traffic control to solve. In this scenario, the flights could be ordered to circle around Logan, resulting in unnecessary fuel usage and carbon dioxide emissions.

Flyways solves this problem by having all flights by the same airline on a single software, giving dispatchers a means to consider flights other than their own. “[At the] end of the day, as an airline, you are operating an entire system of flights, and they all impact each other,” Buckendorf says.

To gauge the overall impact this emission reduction strategy might have in the fight against climate change, I emailed the International Council on Clean Transportation, a research nonprofit based in Washington, D.C. While the numbers Alaska Airlines reports are representative of what is to be expected, “it is not a huge reduction in fuel burn,” writes Jayant Mukhopadhaya, an aviation researcher at ICCT.

The same statement can be applied more generally to using AI to plan routes. “[B]ut every little bit helps,” he adds. “These operational improvements are low-hanging fruits that require minimal investment when compared to things like developing hydrogen-powered airplanes or fueling aircraft with sustainable aviation fuels.”

That “every little bit helps” philosophy motivated the formation of the France-based company OpenAirlines by founder and CEO Alexandre Feray. Using various types of AI, the company’s SkyBreathe platform examines almost every step of an airline’s operation — including aircraft maintenance, flight preparations and flight operations — and finds areas where the airline can reduce fuel consumption.

But Feray cautions that “you won’t solve everything with AI.” Sometimes, physics-based models are needed instead of data-driven approaches like machine learning. Also, SkyBreathe currently doesn’t consider contrails, and neither does Flyways. Feray explains that there currently isn’t a clear, actionable item driven from sound scientific evidence that SkyBreathe can reference. And Saleh says Alaska Airlines’ current use of Flyways is focused on reducing carbon emissions, but he believes the program is capable of taking contrail data into account for its decision- making process if the airline decides to pursue that direction in the future.

Feray predicts the use of AI to expand across the industry, especially when it comes to reducing its carbon footprint. 

We are lucky to be in an industry where you can be more environmentally friendly and, at the same time, improve your bottom line,” he says. He also suspects that, while the current technology relies on historical data, the advancements toward providing real-time data will enhance AI’s reach within the industry.

Buckendorf tells me that Airspace Intelligence is in talks with other airlines, though he and company aren’t ready to announce anything yet. And while he is unaware of other airlines using AI to help dispatching, Wei of George Washington University speculates that bigger airlines with their own research and development teams could be developing their own AI dispatching solutions while exploring options for hiring an outside vendor to develop such technology. Wei himself was scheduled to give a talk at American Airlines, his former employer, to the airline’s operations researchers about using machine learning to aid decision making.

As for my flight from Memphis to Washington, D.C., American Airlines says the company currently does not use AI in route planning. Could Flyways have created a new route in time for my plane to receive the necessary amount of fuel and avert the delay? 

“100%,” Buckendorf says. 

About Karen Kwon

Karen is a science journalist based in the Washington, D.C., area, and is the associate editor of Optics & Photonics News. She holds a doctorate in chemistry from Columbia University and previously wrote about physics, space science and technology for Scientific American.

“The machine is really good at crunching huge amounts of data in an incredible fast amount of time. What the human is really good at is judging the situation.”

— Phillip Buckendorf, Airspace Intelligence
This rendering illustrates how dispatchers in an airline operations center would see recommendations from Airspace Intelligence’s Flyways software. The crisscrossing gold lines represent existing flight routes. When Flyways identifies a conflict, such as the weather front at the bottom right of that map, the algorithm internally simulates possible alternative routes before making its recommendation: the popup message at right. Credit: Airspace Intelligence
On any given day, about 100 or so dispatchers plan flight routes in Alaska Airlines’ Network Operations Center, located on the sixth floor of “The Hub,” a large building on the airline’s Seattle campus. Credit: Ingrid Barrentine/Alaska Airlines

AI: how it’s delivering sharper route planning