Personal recommenders to improve online user experience

Custom Predictive Modelling // Divolte




Rob Dielemans

Our role

Custom Predictive Modelling // Divolte

Background info about Transavia

Personal recommenders on Transavia website

Every year, Transavia, one of the largest European low-cost airlines and a wholy-owned subsidiary of KLM, transports over 12 million passengers. The number of routes that the low-cost carrier offers hit 220 in 2016, and continues to grow. Transparency and service are key value drivers for the airline. In order to improve the user experience of their website, Transavia has introduced personal recommendations based on machine learning algorithms. For the collection of clickstream data, the airline implemented open source clickstream collector Divolte. For the development of the recommenders Transavia engaged in a close collaboration with the data scientists of GoDataDriven.

Distinction by better service

In an industry where price pressure increases exponentially, Transavia aims at making the difference with a transparent approach where all conditions and costs are made clear directly from the start and by pro-actively offering the right service at the right moment during the full customer journey.

Personal recommenders play an important role in Transavia’s strategy towards a fully personalized customer journey. This experience starts with the recommendations of destinations based on individual website behavior. “Viewed destinations are real-time compared with historical website data. Based on this information we recommend destinations that have the strongest match with the search behavior of every single website visitor”, says Charles Verstegen, Senior Revenue Development Manager at Transavia.

Personal recommenders on the Dutch Transavia website

Personal recommenders are not limited to the destination finder, Transavia sees opportunities to increase the relevancy other product and service offerings as well. “Take a business traveler, generally speaking this type of customer benefits more from an offer for more leg room than for bigger luggage allowance, while this is exactly opposite for leisure travelers. By taking these characteristics into account, we are able to recommend the right bundle to different types of visitors. For our customers this leads to an immediate improvement of the user experience”, exemplifies Verstegen

Not far from now, Transavia expects to add real-time price information to the flight pages, allowing customers to easily compare prices between different destinations that are probably relevant to the current session. This could mean that based on price or availability, a visitor that actively searches for flights to Valencia, might be persuaded to book a flight to Malaga instead.

The ultimate goal is to offer the right information, service and offers to every individual customer in every step of the customer journey, from finding their destination to returning home.

Better service by collecting customer information

In order to get to know customers better, it is essential to gather as much relevant information as possible. Transavia uses a wide range of customer related information to find correlations and to train models. The machine learning models are fed with data like group size, duration, group composition, season, moment of the day and of course destination.

Various data sources used to profile website visitors

In order to collect all website data properly, Transavia implemented open source clickstream collector Divolte. Gathered data is then added to a storage layer within the Azure cloud. Subsequently, the data is combined with other data source and ingested in a machine learning model. Finally, to personalize the website, the generated recommenders are made available for the Sitecore website via an API. GoDataDriven implemented Divolte and worked together with the data team of Transavia to develop the recommenders.

Architecture of the Transavia recommenders

Divolte was not the first clickstream collector to be implemented at Transavia. Various solutions are installed for various purposes. Quickly after the initial implementation of Divolte, Transavia decided that they will consolidate as many clickstreams into Divolte as possible. “The big advantage of Divolte is the flexibility. Because of the open source character of Divolte, the clickstream collector lacks the limitations that we experienced in other solutions”, says Verstegen. “Our data team is very excited about the possibilities they now have with Divolte. Our web team, for example, now is able to tag data exactly as the data team prescribes”.

The increased importance of the data team

Transavia’s data team was introduced in the spring of 2016 and is composed of both professionals from business and IT departments. Historically, the airline has always had a strong foundation for data driven modelling of flight optimization. The rapid expansion of data and data sources, created opportunities for more advanced features that they could not seize with the organization structure and existing processes. Hence, the introduction of a dedicated data team.

The data team has full autonomy for innovation, development and operations of data applications. To reconcile and exchange information, on a daily basis, the data specialists are in touch with other teams that are involved, like architecture or e-commerce. For projects the team adopted an agile process, which divides complex projects in clear sprints.

Smart data applications are crucial for the long term strategy of Transavia. Besides projects to optimize revenue, the data team will also be increasingly occupied with operational excellence. These can be projects to develop models that optimize crew planning or the prediction of potential disturbances, like maintenance.

Technology we used

GoDataDriven developed recommenders that provide our visitors with a relevant experience in every step of the customer journey by comparing individual sessions with historical website data in real-time.

Charles Verstegen
Senior Revenue Manager, Transavia