Personalization continues to be a mantra for many travel suppliers. As technology matures, traveler expectations evolve to include a greater desire for targeted search. This is especially true when sharing personal data delivers a better experience for the traveler. Yet, even mature technology faces the hurdle of inter-operability: how do you share data across systems so that a traveler’s past behavior and profile can inform future results?

Machine learning is the process that applications learn from past behavior. This learning promises a “magical” search experience, automatically taking personal preferences into account. This preference-driven shopping framework is a true personalization in travel, helping suppliers more efficiently match trip itineraries to travelers.

One example of this emerging practice in travel is a concept called Preference-Driven Airline Shopping. The concept is driven by a display algorithm for preference-driven shopping, which takes traveler preferences into account when displaying search results. Rajeev Bellubbi, Senior Principal Scientist, wanted the application to focus on personalization beyond the standard flight search filters, which have their own inherent limitations:

“Rather than giving a standard response to all requests from all customers, such as fastest, cheapest or most popular, we use each customer’s own preferences. We can easily store these preferences in their profile and then tailor results based on those specific travel criteria.”

Preferences currently include fare, outbound and inbound flight duration, departure date, length of stay, departure time, arrival time, connection quality, the transit airport, and the carrier.

How preference-driven shopping works

The preference-driven algorithm uses existing profiles — say general preferences of business travelers — alongside the machine learning that evolves a traveler profile based on what was shopped before. Based on this learning, preferences update automatically in the background. This continuous improvement increases conversion by targeting the right itinerary to the right customer.

Bellubbi shares more about how it works:

Our algorithm maps every preference that exists for an airline on a multi-dimensional scale. It assumes that there is an ideal itinerary for every person. It might not exist in reality, but there is an ideal itinerary that a traveler prefers to fly. By plotting these preferences, the algorithm scores the distance of each itinerary from the ideal. And then it sorts according to the shortest distance to the ideal itinerary.

The tablet application in the video below demonstrates how the algorithm works. In this case, the sliders represent preference weights of the algoirthm that automatically adjust according to traveler search behavior. The sliders could exist as consumer-facing, but the idea is that the algorithm learns and delivers options based on profile and past behavior. The algorithm and interface itself is transparent to the traveler.

Another compelling option for the traveler is to group itineraries with similar attributes. When a traveler considers a standard display, similar literariness are generally clustered together. What a travel agent or customer might want to do is collapse similar itineraries and just show the top two or three in each group. The grouped itineraries can be ranked based on their individual preferences and viewed as needed, making it easier to browse flights with different attritbutes and preference priorities. It’s a much more efficient way of displaying itineraries, says Bellubbi:

For example, the first ten itineraries in a result might have the same outbound flight but each with different inbounds. So we can collapse those into one view: here are all the itineraries with the same outbound flight. We call that the anchor. We can also group them according to carriers, all collapsed into one.

Preference-driven shopping helps agents deliver faster, more personalized service. For example, you call the agent and tell them you want to fly from Dallas to Sydney on this date. The agent gives you a couple of responses, and tries to pick the best fit. If the traveler wants to leave a bit later, then the agent responds with another flight. It’s usually one itinerary at a time. That’s the kind of thing that we want to help with – it allows the agent to better serve the customer through machine learning and deeper personalization.

Up next?

The potential to add new preference types makes the product extensible. In consideration are on-time departure time, so travelers can weight on-time departures in the mix. Seat map integration would allow travelers to search by available seat type. Equipment type is another useful preference, so flights can be grouped according to which aircraft flies a specific route.

For travel suppliers, and the technology providers that support them, more efficient shopping both reduces the time spent for travelers to find what they need and reduces costs related to the shopping process. In a world where shopping volume is outpacing passenger booking volume, personalization-driven efficiency is a win across the ecosystem.

Tom Klein As Conference (2)