Given the near 85% fail rate in corporate artificial intelligence projects, it was a pleasure to visit with Alaska Airlines, which launched a highly successful AI system that is helping flight dispatchers. I visited with Alaska to see what the “secret sauce” was that made its AI project a success. Here are some tips to help your company execute AI as well as Alaska Airlines has. 

SEE: Hiring Kit: Video Game Programmer (TechRepublic Premium)

Don’t over-sell to your executive management

Initially, the idea of overhauling flight operations control existed in concept only. “Since the idea was highly conceptual, we didn’t want to oversell it to management,” said Pasha Saleh, flight operations strategy and innovation director for Alaska Airlines. “Instead, we got Airspace Intelligence, our AI vendor, to visit our network centers so they could observe the problems and build that into their development process. This was well before the trial period, about 2.5 years ago.”

Saleh said it was only after several trials of the AI system that his team felt ready to present a concrete business use case to management. “During that presentation, the opportunity immediately clicked,” Saleh said. “They could tell this was an industry-changing platform.” 

Define a compelling business use case

Alaska cut its teeth on having to innovate flight plans and operations in harsh arctic conditions, so it was almost a natural step for Alaska to become an innovator in advancing flight operations with artificial intelligence.

SEE: Digital transformation: A CXO’s guide (free PDF) (TechRepublic)

“I could see a host of opportunities to improve the legacy system across the airline industry that could propel the industry into the future,” Saleh said. “The first is dynamic mapping. Our Flyways system was built to offer a fully dynamic, real-time ‘4D’ map with relevant information in one, easy-to-understand screen. The information presented includes FAA data feeds, turbulence reports and weather reports, which are all visible on a single, highly detailed map. This allows decision-makers to quickly assess the airspace. The fourth dimension is time, with the novel ability to scroll forward eight-plus hours into the future, helping to identify potential issues with weather or congestion.”

“We saved 480,000 gallons of fuel in six months and reduced 4,600 tons of carbon emissions.”
Pasha Saleh, flight operations strategy and innovation director for Alaska Airlines

The Alaska Flyways system also has built-in monitoring and predictive abilities. The system looks at all scheduled and active flights across the U.S., scanning air traffic systemically rather than focusing on a single flight. It continuously and autonomously evaluates the operational safety, air-traffic-control compliance and efficiency of an airline’s planned and active flights. The predictive modeling is what allows Flyways to “look into the future,” helping inform how the U.S. airspace will evolve in terms of weather, traffic constraints, airspace closures and more.

SEE: 9 questions to ask when auditing your AI systems (TechRepublic)

“Finally the system presents recommendations,” Saleh said. “When it finds a better route around an issue like weather or turbulence, or simply a more efficient route, Flyways provides actionable recommendations to flight dispatchers. These alerts pop up onto the computer screen, and the dispatcher decides whether to accept and implement the recommended solution. In sum: The operations personnel always make the final call. Flyways is constantly learning from this.”

Get staff involved—and supportive

Saleh recalled the early days when autopilot was first introduced. “There was fear it would replace pilots,” he said. “Obviously, that wasn’t the case, and autopilot has allowed pilots to focus on more things of value. It was our hope that Flyways would likewise empower our dispatchers to do the same.”

SEE: Graphs, quantum computing and their future roles in analytics (TechRepublic) 

One step Alaska took was to immediately engage its dispatchers in the design and operation of the Flyways system. Dispatchers tested the platform for a six-month trial period and provided feedback for enhancing it. This was followed by on-site, one-on-one training and learning sessions with the Airspace Intelligence team. “The platform also has a chat feature, so our dispatchers could share their suggestions with the Airspace Intelligence team in real time,” Saleh said. “Dispatchers could have an idea, and within days, the feature would be live. And because Flyways uses AI, it also learned from our dispatchers, and got better because of it.”

Define the relationship between human and machine

While Flyways can speed times to decisions on route planning and other flight operations issues, humans will always have the role in route planning, and will always be the final decision-makers. “This is a tool that enhances, rather than replaces, our operations,” Saleh said. Because flight dispatchers were so integrally involved with the project’s development and testing, they understood its fit as a tool and how it could enhance their work.

Capture business value

“With the end result, I would say satisfaction is an understatement,” Saleh said. “We’re all blown away by the efficiency and predictability of the platform. But what’s more, is that we’re seeing an incredible look into the future of more sustainable air travel.

“One of the coolest features to us is that this tool embeds efficiency and sustainability into our operation, which will go a long way in helping us meet our goal of net zero carbon emissions by 2040. We saved 480,000 gallons of fuel in six months and reduced 4,600 tons of carbon emissions. This was at a time when travel was down because of the pandemic. … We anticipate Flyways will soon become the de facto system for all airlines. But it sure has been cool being the first airline in the world to do this!”

Data, Analytics and AI Newsletter

Learn the latest news and best practices about data science, big data analytics, and artificial intelligence.
Delivered Mondays

Sign up today

Also see