Offer management is the process of selling the right bundle – airfare and air ancillaries – to the right customer at the right price at the right time. It is an extension of the well-established revenue management process of optimizing inventory controls for airfares to also include airline ancillary products.  This is the first of three articles describing how machine learning methods can generate targeted, ancillary bundle offers by customer segment to delight customers and maximize revenue for airlines.

Retail merchandising of branded fares (pre-defined, static bundles) and ancillaries is growing in importance for airlines. Average airfare in the airline industry has declined annually by -0.9% over the past decade (IATA 2018 [1]), but sales of air ancillaries, like checked bags, extra leg room, wifi and food and beverage, have grown +40%. In 2018, the total sale of ancillaries was $93 billion (US) (Ideaworks and Cartrawler 2018 [2]) and expected to cross the $100 billion threshold in 2019. That’s a lot of in-flight cocktails!

Effective airline offer management requires an understanding of customer preferences and purchase-behavior patterns across all channels of distribution, from online travel agencies, corporate booking tools or directly from the airline’s website. Thanks to Amazon.com, Netflix, iPhones and other high-quality consumer applications, traveler demands are changing, and they now expect simplification and personalized services throughout the travel planning process.  Sabre’s framework for offer management decision support can elevate both simplification and personalization for travelers. See this earlier Sabre posting about our dynamic ancillary bundles prototype [3]).  The core components of an offer management solution are:

  • Part 1: data collection and customer segmentation based on context for their travel (trip purpose segmentation),
  • Part 2: a recommendation engine to recommend bundles to customers based on such segments or persona, and an offer engine to customize and price the offer for a segment of ONE (1:1 personalization) and
  • Part 3: a test-and-learn experimentation engine using reinforcement learning to continuously adapt the product recommendations to changing consumer behaviors and new competitor products.

The following diagram depicts these offer management processes:

Sabre has implemented selected models of the above framework for both travel agency and hotel applications, and beta-versions are in-work for the end-to-end process on airline.com websites.  Our initial experience using these methods has been generally positive from both implementation and usage perspectives. Practicality has been an important consideration guiding the design of our offer management applications. Here are some of our design choices, and why.

Data Availability

A common theme across the travel value chain is to display targeted content to a given customer based on implicit or explicit information we know about the customer. This type of deep understanding can be derived from a range of data such as transactional booking and ticketing data, demographics, social media, surveys and more.

Our framework is based primarily on historical air ticketing data and ancillary sales (because this information is generally widely available).  In addition to selling volumes, such data contains rich information about the selling channel, booking and departure times and dates, fare types purchased, etc.  Our test-and-learn experimentation framework also uses information on website impressions and conversion rates for each offer coupled with metadata about the context (i.e. pages visited, request details, any bundle customization performed by the user, etc.).  A future data requirement (currently being tested in a research mode) involves use of agency low-fare search shopping results; which can provide further insights on current competitive offerings in the marketplace for use in dynamic pricing decision making.

Customer Segmentation

Customer segmentation based on context for travel is important to support customer acquisition, maximize revenue-generation and retain the most profitable customers. In our experience, segmentation of customers based on purchase-behavior patterns is helpful in determining relevant offers; however, segmentation is never perfect and should be refined periodically with information from new data sources.

Our proposed segmentation approach recommends segmentation based on trip-purpose and revealed preference models. Trip purpose segmentation uses classification based on characteristics of the trip itself (e.g. advance purchase, travel days-of-week, length-of-stay, number of travelers, distance traveled, purchase time-of-day, flight departure and arrival times, O&D market, selling channel, etc.).  These characteristics can reasonably segment customers into similar types (e.g. business vs. leisure-like travel), and they work even when the customer’s identity is not known. For example, a trip request several months in advance for five passengers tells a different story than a last minute, weekday trip request for one traveler. Furthermore, non-overlapping classification based on trip attributes is mutually exclusive and collectively exhaustive (MECE) which simplifies management of rules used and analysis by airline marketing analysts. We have successfully applied techniques such as hierarchical-clustering and CART (classification and regression trees) for trip segmentation.

In the next part of this blog series, we will talk about recommendation and offer engines.

References

  1. Airlines Financial Monitor. https://www.iata.org/publications/economics/Reports/afm/Airlines-Financial-Monitor-Jan-2019.pdf, December 2018–January 2019.
  2. Ideaworks and Cartrawler. Airline Ancillary Revenue Projected to Be $92.9 Billion Worldwide in 2018. https://www.ideaworkscompany.com/wp-content/uploads/2018/11/Press-Release-133-Global-Estimate-2018.pdf.
  3. Airline Offer Management – Sabre Insights https://www.sabre.com/insights/innovation-hub/prototypes/airline-offer-management/