Automated revenue management systems with complex and protected algorithms have significantly changed the revenue management profession. Hotels have become increasingly reliant on advanced computer... Show moreAutomated revenue management systems with complex and protected algorithms have significantly changed the revenue management profession. Hotels have become increasingly reliant on advanced computer systems that continually reforecast hotel demand. However, performance of these systems is not flawless, and revenue managers adjust their computer system forecasts using their professional judgment. These judgmental adjustments can improve accuracy, but if performed poorly, can result in a cascade of suboptimal decisions. This dissertation explores the influence of judgmental adjustments on the accuracy of system-generated hotel occupancy forecasts. To this purpose, three studies have been conducted. The dissertation first evaluated whether it matters which measure is used to assess the forecast accuracy. It was then investigated whether the timing of judgmental adjustments influenced the forecast accuracy. Finally, the characteristics of judgmental adjustments (e.g., direction, size, frequency) and their effects on forecast accuracy were analyzed. The dissertation shows that (1) the choice of accuracy measure is complex and important, (2) the accuracy of occupancy forecasts improves considerably as the forecast horizon becomes shorter, and (3) the effect of judgmental adjustments differs and differs per hotel. Using a disaggregated reforecasting perspective, this dissertation improves understanding of how judgmental adjustments impact the accuracy of system-generated occupancy forecasts. Show less
This thesis applies data mining in commercial settings in the telecommunications industry. The research for this thesis has been performed at T-Mobile Netherlands B.V. and the methods described in... Show moreThis thesis applies data mining in commercial settings in the telecommunications industry. The research for this thesis has been performed at T-Mobile Netherlands B.V. and the methods described in some of the chapters have been also applied in Deutsche Telekom subsidiaries in other countries. We had a rare opportunity to work on real commercial data sets and have the results of our research deployed in practice. Throughout this thesis we describe some of the challenges that data miners (or data scientists) meet when working on business problems and our solutions to these problems. The complex data sets we were analyzing contained in certain cases millions of records. In this research we were using simple methods combined in innovative ways to achieve results that were either an improvement on how the business was previously solving these problems or solving important business problems that were not addressed before in such detail. We address the stages of CRISP-DM (CRoss Industry Standard Process for Data Mining), and our main focus is on the stages least covered in literature. Show less
