Machine Learning and AI at Delta Air Lines

Understanding how AI and machine learning are affecting Delta and the world's airlines.

Right in the middle of the company’s 2017 Investor Day presentation, Delta highlighted the following: “Expand machine learning and artificial intelligence.”[1] As Anastassia Fedyk writes in Harvard Business Review,  “Machine learning is, at its core, a set of statistical models meant to find patterns of predictability.”[2] If large data is critical for machine learning than Delta is no stranger to it. Delta Air Lines—with more than 180 million passengers flown annually, more than 15,000 flights per day, and more than 80,000 employees worldwide—is no stranger to the AI and machine learning revolution. [3] In fact, a 2017 survey of the industry reported that half of all global airlines plan to make significant investments in their AI and machine learning capabilities within the next three years.[4]

Airlines generate immense amount of data and information which often goes unused. To better explain, Virgin Airlines reported that a single flight generates roughly half a terabyte of data.[5] Until very recently, that was such a huge amount of information it was unusable. However, “with machines as sidekicks, though, people can more quickly find valuable insights buried in big data,” writes James Wilson in a May 2016, Harvard Business Review Article.[6] Finding these valuable insights is critical in reducing delays and disruptions, improving customer experience, and increasing the bottom-line.

Delta has been investing in the AI space for more than a decade. As of 2003, the company was using predictive fleet maintenance programs to “filter and integrate data from [the company’s] physical assets, contextualize it and provide actionable insights on their current technical condition.”[7] Maintenance personnel are relayed signals on problematic parts long before these parts fail. In the twelve months ending March 2018, Delta’s predictive analytics software prevented nearly 1,200 aircraft delays and cancellations.[8]

AI and machine learning in the airline industry—especially with Delta—is extraordinarily complex. For example, airlines have been using AI algorithms in the ticket pricing space for over a decade now, but have only recently made large inroads in fully autonomous flying. First generation technology included flight management systems and automatic cabin pressurization technology. Under current development, with near-term deployment are machine learning initiatives such as predictive analytics around passenger behavior to better model revenue management, enhanced biometric screening to eliminate the need for boarding passes and baggage tags and improved rerouting due to maintenance and weather issues.[9] Delta’s Director of Innovation, Nicole Jones, stated, “In the long-term, passengers could transfer from the curb to the gate without the need of a printed or mobile boarding pass.”[10] Furthermore, long-term initiatives that the company has started to invest in include fully autonomous flights, and entirely self-serve experiences in which customers never interact with a Delta employee, from a check-in agent, to a gate agent to a flight attendant, machines and AI would usher him or her through the process.

Airlines—Delta included—only have limited means to process this immense data and fully understand it today. Though systems have become increasingly good at collecting and storing critical data, how best to put it to use can remain elusive. Furthermore, existing legacy systems can often clash with new-modern systems. An Oliver Wyman report on the space states, “legacy systems not flexible enough to accommodate more sophisticated analytics and artificial intelligence systems,” will inevitably create issues. Further investment and hiring the right people who understand these issues can greatly reduce friction as the industry grows into itself.

One major roadblock remains: are customers truly comfortable with a fully autonomous airline industry or will humans always be necessary? As fliers, will customers feel safe knowing that there may not be a pilot on the plane, or that a mechanic never checked the engine because a machine did instead? Will we feel comfortable handing our luggage with our valuables to a robot that scans our face and promises delivery of the baggage to a final destination? That final question is particularly potent because Delta is currently using machine learning algorithms to tie a customer’s face to his or her bag. Gone away are the day of luggage tags.[11]

The end goal of all of this is to create a seamless passenger experience while delivering maximized profit to the company. That means fewer delays, a more enjoyable traveling experience, near-zero aircraft issues and perfect customer service. The airline industry hopes in many ways machine learning and AI can deliver on these long desired of goals.


Word Count: 732

[1] “Investor Day, 2017,” Delta Airlines Investor Day,

[2]  Anastassia Fedyk, “How to Tell if Machine Learning Can Solve Your Business Problem,” Harvard Business Review (November 25, 2016)

[3] “Worldwide Service,” Corporate Stats and Facts, Delta,

[4] “Airlines Turn to AI as They Up IT Spending,” Business Travel News,

[5] “Boeing 787s to create half a terabyte of data per flight, says Virgin Atlantic,” ComputerWorld UK,

[6] James Wilson, Sharad Sachdev and Allan Alter, “How Companies are Using Machine Learning to Get Faster and More Efficient,” Harvard Business Review (May 3, 2016)

[7] “7 Ways Airlines Use Artificial Intelligence and Data Science to Improve Operations,” AltexSoft: Software R&D Engineering,

[8] Ibid

[9] “Airlines Turn to AI as They Up IT Spending.”

[10] Ibid

[11] Kristina Velan, “How Artificial Intelligence Will Change the Airline Passenger Experience,” Apex, January 4, 2018,


Organovo: bioprinting tissue to speed up drug development


Darktrace: Can Artificial Intelligence lead the fight against Cyber Crime?

Student comments on Machine Learning and AI at Delta Air Lines

  1. Great article! It is refreshing to see how airlines like Delta are leading the way in machine learning, as current public discourse has largely been about the industry’s trend towards shrinking seats and amenities for customers. In response to your question about when customers will be comfortable with machines in airlines replacing humans – I agree that the rollout will have to be staggered, with perhaps the last step being self-flying planes. I imagine that the first step will be the low stakes positions, where safety is not compromised (or interpreted as being compromised); perhaps roles like customer service, ticketing, and even some flight attendant roles. However, I also see a potential reputational issue with this, as airlines like United have recently come under fire because of poor company policies and actions by flight attendants [1]. I imagine that it will be even harder to integrate machines into these roles, when trained humans can’t navigate the intricacies of customer service grey areas.


Leave a comment