We’re creating a more connected travel industry, underpinned by sustainability and long-term investor relations.
Research Engineer, Amadeus IT Group
Revenue management (RM) is the science that looks at how to maximise revenue for airlines. The traditional approach, which has prevailed since its inception 40 years ago, has been based on forecasting traffic flows (customer volumes and willingness to pay) followed by an optimisation procedure that prioritises among customers by selecting optimal availabilities, or prices. However, RM makes many (and unrealistic) assumptions, such as monopoly or stationarity, and often disregards cross-price elasticities between products.
One of the areas where we are trying to help airlines improve their business is focused on Deep Reinforcement Learning (RL) applied to revenue management. Amadeus recently presented a research paper on RL, prepared by Anh-Quan Nguyen, Thomas Fiig, Rodrigo Acuna Agost and myself.
Reinforcement learning (RL) is an area of machine learningfocused on how machines take actions in order to optimise a given reward (e.g. revenue) by interacting with its dynamic environment. One of the featured examples of applications of RL in the industry comes from driverless cars. By a learning process of action (A), state (B) and reward (C), a car can take action, either by continuing straight, turning right or left (A); measure its state by using sensors to see what elements surround it (B), and obtain a reward, which in this case is not crashing (C). Another example is computers beating humans at highly complex games, such as chess or go.
Essentially, what we showed with our work is that classical revenue management techniques are no longer sufficient. As a matter of fact, RL is an alternative that opens the door to a radical new approach based on direct price testing – that’s right, no more demand forecasting, passenger behaviour modelling or optimisation based on old-fashioned models. With RL, more information is added to the way prices are calculated, such as competitors, state of the market, etc. This only comes to show the disruptive impact that machine learning could have on many industries, especially in travel.
The fruit of our efforts has been recognised by the international community. As a matter of fact, our research paper on RL received the best innovation award at The Airline Group of the International Federation of Operational Research Societies (AGIFORS) conference 2017, one of the most prestigious professional societies dedicated to the advancement and application of operational research within the airline industry. At the event, attendees discuss the latest innovations in airline operations research and analytics.