At JetBlue, machine learning may help you book your next vacation

In a world of abundant travel possibilities, JetBlue is leveraging machine learning to personalize and simplify the itinerary planning process for customers.

More options, more complexity

You have just made the exciting decision to plan a summer vacation. Your immediate next steps are to choose a destination and book flights. There are many factors to consider – which dates can you travel? What types of destinations do you prefer? Do you have budget constraints?

As the airline industry continues to grow, there is an ever-expanding abundance of options and decision points around booking travel. While these options create new possibilities for customers, they can also make planning a vacation a daunting and time-consuming task. In fact, recent research suggests that around 43% of leisure passengers and 51% of business passengers want to spend less time researching flights [1].

Role of machine learning in itinerary planning

How might an airline save customers the time – and, potentially, stress – associated with this research? JetBlue, an airline that prioritizes technological innovation throughout its business, is leveraging machine learning to predict customer behavior in the itinerary planning process and offer personalized flight and vacation recommendations. Last year, JetBlue announced a partnership with Utrip, a destination discovery and planning platform that helps travelers create a highly personalized vacation itinerary in minutes. The portal “uses artificial intelligence and locally curated recommendations to save customers the legwork that planning a vacation requires.” [2]

 

Innovative booking platform uses customer inputs to provide personalized trip recommendations [2].

JetBlue can access rich passenger data and combine it with machine learning to profile customers and predict future behavior. In 2017, researchers from the Amadeus Research and Innovation Division conducted an experiment demonstrating how this machine learning process might occur. They also compare the effectiveness of machine learning to algorithms more traditionally used to predict itinerary choices. The experiment and resulting findings are outlined below [3]:

  • Training: The team used a dataset of ~27K choice situations to train the model (each situation contains a search, between 1 and 50 itinerary options, and the option actually booked by the customer)
  • Testing: Another ~7K choice situations were used for testing. The algorithm ranked various features on their importance to customers (e.g., price, connections, trip duration, arrival and departure times). Segmenting business and leisure travelers was an important step, as behaviors differ significantly between these travelers.
  • Findings: The machine learning model yielded better performance in predicting bookings than the traditional algorithm, primarily due to its ability to allow non-linear modeling of a high number of variables.

JetBlue’s focus on innovation

As airlines continue searching for ways to differentiate themselves and retain valuable customers, personalization of the passenger experience is becoming an important initiative that can be enhanced through machine learning. JetBlue has designed its organization to align focus on this priority and ensure the company remains an industry leader in technology innovation more broadly.

Chief Digital and Technology Officer Eash Sundaram explains that, while the digital function used to be split between commercial and technology, it is now a blended organization, reinforcing the linkage between technological innovation and customer experience. In addition to his executive role, Sundaram leads an innovation lab and serves as the Chair of JetBlue Technology Ventures – a subsidiary established by the company in 2016 to identify, invest in, incubate, and partner with innovative startups in the areas of artificial intelligence, predictive analytics, and other emerging technology. These recent bold moves reflect JetBlue’s and Sundaram’s commitment to building the next generation of customer experience [4].

In an increasingly cost-focused industry, JetBlue must now consider the resources it will devote to technological innovation over the next decade. While the organization appears to be designed for success structurally, the talent it continues to recruit will play a critical role in the success of innovation initiatives. For instance, few airlines today are hiring data scientists; recruiting this talent from across industries could unlock substantial value for JetBlue, especially if it does so before other airlines and gains a first-mover advantage [5].

Future impact on air travel

As machine learning continues to advance, and as it enables airlines to profile customers in ways they were unable to do previously, a world of hyper-personalization in the near future does not seem out of the question [5]. Beyond customized itineraries and vacation packages, could airlines offer personalized pricing at the customer level? What challenges might arise with machine algorithms – for instance, bias leading to unintended passenger discrimination – that would need to be considered from a regulatory perspective?

Even with these unknowns, machine learning will be an important component for JetBlue going forward as it strives to create a personalized, streamlined customer experience. As Umang Gupta, President of JetBlue Vacations, points out, “When you hear artificial intelligence, it’s easy to envision a far-off future seen in the movies. . . But AI is ready to change how we now plan travel” [2].

 

(Word Count: 777)

 

Sources:

[1] Harteveldt, Henry H. “The Future of Airline Distribution, 2016-2021,” 2016, written by Atmosphere Research for IATA, [https://www.iata.org/whatwedo/airline-distribution/ndc/Documents/ndc-future-airline-distribution-report.pdf], accessed November 2018

[2] “JetBlue Vacations and Utrip Launch Artificial Intelligence-Based Trip Planning Portal,” press release, June 13, 2017, on JetBlue website [http://mediaroom.jetblue.com/investor-relations/press-releases/2017/06-13-2017-141427979], accessed November 2018.

[3] A. Lheritiera, M. Bocamazoa, T. Delahayea, R. Acuna-Agosta. “Airline Itinerary Choice Modeling Using Machine Learning,” March 2017, [http://www.icmconference.org.uk/index.php/icmc/ICMC2017/paper/viewFile/1178/393], accessed November 2018.

[4] High, Peter. “JetBlue’s Head of Technology And Digital Also Runs The Company’s Venture Arm,” September 2017, Forbes, [https://www.forbes.com/sites/peterhigh/2017/09/05/jetblues-head-of-technology-and-digital-also-runs-the-companys-venture-arm/#33554cf96e48], accessed November 2018.

[5] R. Boin, W. Coleman, D. Delfassy, and G. Palombo. “How airlines can gain a competitive edge through pricing,” December 2017, McKinsey, [https://www.mckinsey.com/industries/travel-transport-and-logistics/our-insights/how-airlines-can-gain-a-competitive-edge-through-pricing], accessed November 2018.

Previous:

Overcoming Information Asymmetries in Credit Markets: Machine Learning at LendingClub

Next:

Intel’s move towards self-driving cars

Student comments on At JetBlue, machine learning may help you book your next vacation

  1. As someone who is constantly trying to plan my next vacation, the idea of simplifying the planning process is incredibly attractive. I typically search for and book flights through an online flight search or online travel booking website – rarely do I actually go to an airline’s website to book a flight. I’m curious how this process of booking influence the volume and type of data that JetBlue has access to and uses in its machine learning algorithms. I know you mentioned JetBlue used a dataset of 27K situations, is that enough? Is it representative of customer booking scenarios?

    I’m also particularly interested in your point about personalized pricing. I’m always searching for the best deal, and I’ve heard that the prices shown to me differ on a number of factors (location, computer type, how many times I’ve searched). Is that considered a type of machine learning that the airlines are already using? I also know that airlines segment their passengers and prices on a number of similar factors. Can machine learning be used to help make that process more efficient and accurate for airlines?

  2. I can total resonate this when I spent hours to search the right plane tickets for vacation.

    As illustrated in the article, the application of AI in the airfare business could largely reduce the time customers spend on the “treasure-hunting” experience. While I totally support such application, I’m wondering if JetBlue is the right platform to implement. JetBlue represents only a fraction of the routes available, while machine learning results based on the customers’ data and preferences may exceed the capability of JetBlue. In this case, JetBlue may provide the service they may have to accommodate the customers using promotions or other solutions, and this new data could tamper the customers’ selection model.

    So I think the best application platform is a third-party platform, like Expedia. They could provide all the information based on the customers’ preference, and give them the best match as well.

  3. This was such a cool article! I would love to have the opportunity to plan my itinerary while I am booking my tickets. It is definitely painful planning vacations especially to new countries for more than a few days. However I was surprised to see JetBlue pursuing this application. Like the comment above I would have expected third parties to have more access to historical data that could utilize ML to develop trip recommendations. I am very excited to see how things progress with ML at JetBlue in the future!

  4. Thanks for your article, William!

    You raised an interesting point related to the unintended consequences of personalized pricing at the customer level. From a purely economic perspective, price discrimination is “efficient” in the way that maximizes the total surplus in the economy. Nevertheless, in reality, if you find out you are being price discriminated people usually get mad at the company and probably end-up losing brand loyalty.

    Just to add on your question on regulatory issues. Charging different prices to different customers is generally legal. “The practice could be illegal, however, if the reason for the difference were reliance on a “suspect category” such as race, religion, national origin or gender” [1]. Therefore, Jetblue would have to be extremely cautious to make sure the relevant variables to the price discrimination relate only to the customer’s willingness to buy and not with any racial or religion variable (or perfectly correlated variables).

    [1] CNN Law Center, http://www.cnn.com/2005/LAW/06/24/ramasastry.website.prices/

  5. Travel is a very personal experience and as an avid traveler, I often find it a problem to find a website that can use the limited data they collect for me to give me the right and fair priced recommendations. I think machine learning innovations work when they have multiple data sets to pull from and have the ability to analyse this data to give the most relevant results. How is JetBlue trying to solve for this problem? What are the data sources that they are relying on apart from the historical travel and the preferences they ask the customer in short forms? What is the personalisation actually driven by? There is also the question of tying in the actual budget constraints with the preferences. How does machine learning solve for this?

  6. The pain points associated with searching and booking plane rides are legitimate. However, I am curious whether this could actually lower revenue rather than increase it. Say certain flights were recommended based on past travel history, typical destinations, connection preferences, etc. If consumers begin utilizing this recommendation feature to book cheaper flights on average, then revenue will likely not be maximized. JetBlue has to be careful to balance customer interests with fiduciary responsibility in utilizing machine learning.

Leave a comment