Southwest Airlines (Southwest) is one of the United States’ largest, lowest cost airlines. The company flies nationally and internationally and is known for its friendly, creative customer services, low air fares, and generous treatment of its employees. Southwest flies to 121 airports in 10 countries, has over 56,000 employees, and takes over 1 million flights per year.
Value Creation through AI & Machine Learning
Southwest has invested in AI and machine learning capabilities in four major areas: chat bot technology, customer segmentation, dynamic pricing, and automated airplane data collection to assess operational efficiencies. Of these, chat bot and customer segmentation create value and dynamic pricing and data collection capture value.
AI Chat Bot Technology: Southwest has started Data Science and Machine Learning teams to develop its AI chat bot, Southwest Bot. The major distinction between the Southwest Bot and the traditional chat functionality is that Southwest Bot uses natural language processing to respond to any questions asked by customers. The previous chat could only process pre-approved questions. Southwest Bot was trained on existing chat and phone call data sets and tested via partial release to gather real-time data and ensure an acceptable customer query resolution rate. Southwest Bot’s flexibility, allows Southwest to handle a higher volume of customer queries. This improves the customer experience, as a higher percentage of queries can be resolved without human intervention and leads to shorter wait times and lower staffing costs for Southwest.
Machine Learning Generated Customer Segmentation: The second area where Southwest has leveraged AI to create value is in enhanced customer segmentation. Southwest has gathered a massive amount of data on its customers: their location, flight preferences, flight frequency, spending habits, communication response habits, in-app activity, etc. Southwest’s Data Science and Marketing teams leverage these data with machine learning methods to improve how they target customers with offers and reminders about Southwest features. These teams train AI models to create differentiated customer personas not previously known to the Marketing team leveraging the massive data sets and iterative statistical methods like K-means clustering. After a lengthy training period, these teams have seen increased target customer response rates, indicating that the updated personas actually improve the relevance of offers served to customers. This improved accuracy not only benefits the bottom line, but also customer’s satisfaction with Southwest’s offering.
Value Capture through AI & Machine Learning
Dynamic, AI Pricing: Dynamic pricing in the airline industry has been around for a while. Most travelers are familiar with the most basic form of dynamic pricing, time based. Flights get more expensive leading up to the date of departure, because travelers are willing to pay a premium to get to their destination. See below fares for flying from Boston to San Francisco with a tomorrow departure date. Fares clearly increase as the departure date approaches. Southwest, and other airlines, are now dedicating AI resources to building algorithms that can discriminate on more factors than just timing (e.g. in-app behavior, previous flights booked, multiple page visits, customer personas, and competitor flight offerings). These models allow airlines to update prices more frequently and make pricing customer specific. The goal of these models is to identify a customer’s willingness to pay, and extract as much as possible in the fare. As these models continue to improve, Southwest will improve its value captured.
Operations Data Collection: The last area where Southwest is implementing AI is in-flight recommendations to pilots. Similar to client segmentation, Southwest was capturing data on its flight operations long before machine learning was well-known. Each flight offered data on the amount of time in-air, fuel spent, number of passengers, and decisions made in-air (i.e. going into a holding pattern, altitude selected, etc.). The Southwest Machine Learning team now leverages the data to develop high confidence estimates of the cost of each decision and applies them to current flights. These models work by using historical data to assess the cost of different flight attributes and make recommendations for trade offs across the fleet (e.g. how high a plane should fly and which plane out of two should enter a holding pattern). As these model mature, Southwest will reduce operating costs.
As Southwest implements these advancements in AI and machine learning, they have several challenges to overcome:
- Resistance to adoption of tools created by a centralized AI and Machine Learning team – Firms initially stand up AI and Machine Learning teams as central teams, to ensure the members are learning from each other and are working on top priorities. This structure causes limited adoption of tools, as other departments (i.e. customer experience, marketing, pricing, and operations) are unwilling to risk their role on new technology.
- Lack of data sharing across areas or links across databases limiting model effectiveness – Without good data sharing across multiple departments, the likelihood of AI and machine learning models will outperform Analytics teams is lower. A tangible example of this is the link between in-flight purchase data and customer segmentation data. Since Southwest is a low cost carrier, a disproportionate amount of its revenue comes from in-flight purchases. Without linking the segmentation data with in-flight purchase data, both the AI customer segmentation and pricing decisions will be ineffective.
- Regulation against AI and machine learning in a high risk field – Regulators have been slow to allow AI to make decisions in an industry where mistakes can cause crashes.
To avoid these pitfalls, AI and machine learning at Southwest needs to be structured intentionally. After initial success, the CEO should embed AI and machine learning capabilities in each department and set department goals for initiatives completed that leverage AI. While AI and Machine Learning teams should not be centralized, they should be supported by a centralized data infrastructure team that ensures data sets across departments are structured compatibly and are shared appropriately. Lastly, the legal and compliance teams should begin recruiting experts to advocate for progressive regulation of airline’s AI and machine learning capabilities.