Transforming the airline customer experience
Melissa Hartwick Melissa Hartwick
March 21 8 min

Transforming the airline customer experience

Delays and cancellations consistently rank as two of the top pain points for airline customers. According to the Air Travel Consumer Report, released by the U.S. Department of Transportation, both are regularly listed in the top 5 formal passenger complaints.

It is not hard to understand why these complaints continue to be filed. Of the $60 billion that is estimated as the cost of airline disruptions, a staggering $25 billion is due to lost productivity and downtime.

Airlines are starting to invest in tools to improve communication with passengers. Approximately 85% of airlines report that they plan to implement a chatbot by 2021. However, as long as human agents and chatbots are not supported behind the scenes with more advanced tools to enhance insight and decision making, airlines won't realize a step change in customer satisfaction.

In a time when customer-centricity is core to airline strategy, airlines need to better equip personnel with tools that are able to predict disruptions in advance and better manage disruptions in real time.

Real-Time Disruption Management with AI - A Competitive Advantage

There is a window of opportunity to lead the way in a new era of real-time disruption management using artificial intelligence (AI).

AI is not new to the airline industry. It has been discussed at length for various use cases, from dynamic pricing and flight scheduling to predictive maintenance. It is now time to apply AI to real-time operations, to build a competitive advantage and drive real value for customers.

Building this competitive advantage can in part be accomplished by augmenting the role of operations personnel with AI tools that:

  • Enhance predictions around the probability of delay
  • Identify how the delay will propagate through the network
  • Offer intelligent recovery options during a disruption that also take customer lifetime value into account

Customer Perspective

As mentioned, delays, missed flight connections, and cancellations are keenly felt by flyers. How an airline responds to these disruptions influences a customer’s perception and ultimately the airline’s bottom line. AI tools can give airline operations personnel new information to help them offer customers the best possible flying experience.

Scenario #1

Meet Ben and his family of four. Ben has planned his family’s first trip from Toronto to Orlando, and their flight leaves in three days. Behind the scenes, the airline’s AI-enabled disruption forecasting tool is predicting an 85% chance of a 2hour+ delay due to flash freezing.

Ben receives a notification from Awesome Air on his phone:

“Hi Ben, your flight on Sunday February 24th at 17:00 from YYZ to MCO is likely to be delayed due to an incoming storm. Would you like us to re-book your four seats on an earlier flight at 11:00 to avoid the storm? This service is free.”

Ben clicks the accept button and is automatically emailed 4 boarding passes for the earlier flight.

Scenario #2

Meet Gabrielle. Gabrielle is a frequent flyer of Awesome Air and is waiting at Chicago O’Hare for her return flight to Boston. Due to a mechanical issue, the flight will likely be cancelled. While it is the last flight out that night for Awesome Air, their alliance partner, Team Player Air, has a flight in 45 minutes that is at a 76% load factor. Because Gabrielle is a highly valued Awesome Air customer, the airline’s AI recovery tool recommends rebooking her on their alliance partner’s flight.

The operations team agrees with the AI recommendation. Gabrielle receives a notification on her phone:

“Hi Gabrielle, we are very sorry, your current flight from ORD to BOS is delayed. In the spirit of Awesome Air, we would like to offer you seat 15A on Team Player Air’s flight at 21:45 at no extra cost to you. Alternatively, we can offer you business class seat 2B on flight YY3329 tomorrow morning at 6:30.”

Gabrielle clicks the button to accept the flight on Team Player Air and automatically receives a new boarding pass.

How are these seamless experiences enabled? By an operations team augmented by AI tools that:

  • Predict disruptions with greater accuracy further in advance
  • Provide better recovery recommendations as disruptions unfold

Proactively Mitigating Disruptions

To be useful, a disruption forecasting tool must:

  • Predict likely delays of 30 minutes or more (e.g. “There is a >80% chance of a 1.5h delay on flight YY4445 from CDG - JFK”)
  • Identify how delays might impact downstream flights (e.g. “This delay will likely cause a 1.25h delay on flight YY4564 from JFK - SFO”)
  • Provide this insight with enough advance notice to allow for it to be actioned
  • Be accurate enough for operations staff to have confidence in the tool’s predictions and recommendations (i.e. minimizing percentage of false positives and false negatives)

Machine learning, specifically a time series model, is ideal for forecasting disruptions. By leveraging real-time and historical data of flight movements, weather and airport congestion, among a whole host of other data sources, a well-trained time series model can yield timely predictions at a sufficient level of accuracy to be actioned. Accuracy will continue to improve as the model learns from human responses to its previous predictions on the likelihood of delays and cancellations.

The threshold at which operations personnel are notified of a potential disruption could be adjusted according to a customer’s lifetime value, loyalty standing, fare class or purpose for travel (if known). For example, a frequent business flyer, or a customer with multiple layovers, might prompt a flight change recommendation at a lower threshold (e.g. 55% likelihood of delay) as there would be greater negative impact if the delay occurred. This example points to how AI tools can enable and further enhance the experience for passengers by leveraging readily available contextual clues.

Handling Disruptions in Real Time

A comprehensive disruption management recovery solution needs to take into account three interdependent and complex components:

  • Aircraft re-routing
  • Crew re-assignments
  • Passenger re-accommodations

While the focus of this article is on re-accommodating passengers, this cannot be effectively accomplished unless aircraft are correctly positioned throughout the network and crews have been correctly assigned based on their duty constraints.

Classical Operations Research (OR) models work well for scheduling problems that are solved one to two months before a given flight. However, they do not perform well in real-time conditions. Tens of thousands of constraints (100s of aircraft, 1,000s of crew, 10,000s of passengers) and the time-sensitive and dynamic nature of the problem do not align well with the rule-based approach taken by OR models.

Current software offerings in the market provide a solution for a portion of the problem (e.g., only crew); however, orchestrating across all three interdependent components in real time is still a challenge.

In recent months Element AI’s applied research lab has demonstrated the viability of using machine learning to reduce recovery time and cost by combining machine learning with OR. Early results are promising. In technical terms, an innovative approach using graph networks allows a machine learning model to shrink the problem space for an OR solver, allowing for a quality recovery solution that, in parallel, solves for aircraft, crew, and passengers.

Why Explainability Matters

Any innovation that is put into operation must be easily understood and trusted. ‘Explainability’, a key topic in AI research, looks at developing approaches to provide the user with greater insight into what is driving an AI model’s recommendation. This is akin to opening up the hood of a car to get a better understanding of its inner workings. For an AI model, a multitude of factors could be influencing a recommendation, including input data that is being ingested in real time as well as data the model was trained on.

For disruption management, Explainability could come in the form of insight into:

  • Why a given flight is likely to be delayed
  • Why the recovery model is recommending option A over option B

While applied research on Explainability is still in its infancy, advancements will only further the adoption of AI-driven products in the airline industry.

Benefits to Real-Time Disruption Management

Over time, AI-augmented decision-making tools focused on real-time disruption management will spread throughout the industry. Early adoption and continuous innovation will enable pioneering airlines to achieve a significant head start over competitors in building this customer-centric competitive advantage. Further, in an age of regulation that requires compensating passengers for delays, there is even more pressure to resolve disruptions, not only quickly and efficiently, but also economically.

The benefits of adopting AI to augment decision making in airline disruption management are clear:

  • Operations personnel would receive notifications of potential disruptions, empowering them to proactively engage with customers in advance - before they arrive at the airport - to enhance their travel experience.
  • Having a clear understanding of how a flight delay will impact downstream flights provides operations personnel with the necessary insight to mitigate delays across the network.
  • During a disruption, an AI tool recommending multiple recovery solutions faster, along with costs for each, allows for passengers to be re-booked much more quickly and get one step closer to their destination.
  • Layering in loyalty status or ticket class allows for loyal customers and frequent flyers to be reassigned first. This may increase the likelihood that a valuable passenger will continue to fly with the airline in the future, an important KPI for many airlines.

In conclusion, it is time for airlines to handle disruptions in a more customer-centric way. AI-augmented decision making is key to transforming the passenger experience by enabling operations personnel to mitigate disruptions in advance and manage disruptions in real time.