Machine Learning Applied to Revenue Management

The article summarized below was published by Marius Radu in the 44th number of Today Software Magazine.

The author’s objectives were to analyze the current state of Machine Learning (ML) software applications from the business and product owners’ perspectives and to provide support to those looking for answers to questions regarding the data volumes and computational power required by ML. Three types of stakeholders can take part in or influence the application development of ML solutions, each of them with their specific agenda. The data scientists are interested in the tools themselves, the software developers pay attention mainly to the performance of the applications and business people want solutions for well-defined real life situations.

Machine Learning Applied to Revenue Management

Marius points out that nobody questions the popularity of ML algorithms, which have been frequently included in solutions / platforms for analytics and cloud solutions by industry players of all sizes, including the likes of IBM, Oracle and Amazon. The ML business oriented stakeholders have played a critical role in designing such applications, their domain know-how representing a key ingredient to successful implementation, and judge ML based on how well it solves real life problems.

In the second and most consistent part of the article, the author delves into three applications of ML in hotel management that aim at maximizing revenues: market segmentation, demand forecasting and pricing tactics and positioning. All these examples indicate that ML is an iterative process in which the programs learn how to reach their goals through trial and error. Without producing perfect results, these algorithms help us improve decision making in complex processes, where controlling all the variables at play is beyond human capacity. ML algorithms have been around, constantly improved and kept expanding their areas of application for over 15 years. Major industries like hospitality, travel, real estate and online advertising rely today on ML for their most complex decision making and forecasting tasks.

The article ends with answers to the questions it raised in its introduction. The relatively small databases in a hotel property management system (PMS) prove that ML algorithms can function without big volumes of data. While it is true that data from online booking systems and recommendation engines are also fed into the PMS, current technologies would produce reliable results even without it. At the other end of the scale lie the computational power, which ML require in substantial quantities. This is no longer a barrier, as the power of computers has been rising exponentially over the last years and keep improving at a fast pace.

According to Marius, solving current problems often depends more on innovative methodologies than on cutting age technologies. A well-chosen question and a robust methodology are critical elements in the process.

The entire article can be read here.

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+