The general problem of the revenue management department is to maximize revenue through filling the fleet with the most profitable customers. Flight analysts today are doing a great job filling flights with the customers willing to pay the most. This depends on e.g. seasonality, duration of the trip or the day of the week people are flying. However, the demand and the willingness-to-pay of the customers on a flight is significantly affected by all kinds of events, trade fares, etc. This customer behavior is difficult to predict with classic airline data.
In this talk we will present a deep learning based approach to extract so-called “data signals” from social media data (such as Twitter, Facebook feeds, etc.) event pages and search engines. The data signals are directly incorporated into the models that are used to predict demand and willingness-to-pay.
The method was tested in a proof-of-concept during three months. We showed that the models augmented with data signals clearly outperformed the standard models. As a next step, we want to unlock the potential from adding contextual data from various sources to our models to describe customer behavior during events more accurately.