I’m Fortunate to Work in a Company of Geniuses.

I’m fortunate to work in a company of geniuses. In fact, PROS has been at the forefront of AI and Machine Learning for almost 20 years, and our team of 40 data scientists (including +20 PhDs) have built some of the most sophisticated algorithms in the industries in which we operate. And with our Microsoft partnership, we are now building cool new generative AI functionality on top of Azure OpenAI. It’s a very exciting time to be working in an AI company, and I consider myself deeply privileged to be working here with such a great group of people. One notable aspect of our AI team is that they publish and present like crazy, and I thought this paper on revenue management in the Airline industry is a particularly good example. Enjoy!

In this paper, we pose a completely different question of interest and a new approach to solve the Revenue Management (RM) problem: Is it possible to directly prescribe control parameters of the RM system using historical data without any demand forecasting?

We propose a novel methodology that does not rely on the abovementioned framework with demand forecasting and optimization. It utilizes historical booking data to directly generate the required RM output, skipping demand forecasting altogether. Our approach also alleviates the need of extensive historical data and eases the implementation of RM in practice. That makes our proposed framework accessible to industries relatively new to RM in addition to industries with demand volatility. We focus on generation of marginal opportunity costs (bid prices), which are an important and commonly used output of an RM system, as we will review later in this section.

Through a 3 comprehensive numerical study, we show that our method generates a robust bid price output even in the scenarios of unreliable historical data. Some unique characteristics of our approach that set it apart from conventional RM methodologies, can be summarized below:

• Direct: We generate bid prices directly from historical booking data, skipping demand forecasting completely.

• Adaptive: As we illustrate with an extensive simulation study in Section 4, our methodology is robust with good revenue performance under various scenarios with mis-specified demand.

• Machine Learning (ML) Based: We utilize a deep Neural Network (NN) model as the predictive algorithm, which provides flexibility with respect to data sources and features used for prediction.

It is well recognized that incorporating additional factors such as competitor data, shopping data, and market indicators can improve prediction accuracy in RM and pricing (Gautam, Nayak, & Shebalov, 2021). Although our approach does not require any inputs other than the historical booking data, it can easily incorporate additional data when available.