The identification of inaccurate demand estimates

Most quantity-based airline revenue management systems rely on forecasting expected demand to prepare revenue-optimal capacity allocations. Inaccurate demand forecasts result in non-optimal allocations and hence, hurt revenue. Most airline revenue management systems also allow room for analysts to compare the accumulated bookings against the forecasts and to intervene, if they deem the demand forecast to be inaccurate. Systematically detecting outliers, where demand differs from expectations due to systematic market shifts, e.g., induced by special events, is an open challenge.

In a set of controlled simulation scenarios, we let demand systematically deviate from the general level. After transforming demand into bookings via a minimal revenue management simulation, we apply a variety of outlier detection techniques to determine whether the number of bookings can be classified as either normal or abnormal. The comparison includes a Euclidean distance-based approach, K-means clustering, and tolerance intervals and evaluates the ability to detect genuine outlying demand and false positive rates. We evaluate and discuss effects of unexpected demand on revenue, confirming previous findings that inaccurate demand estimates can result in a loss of potential revenue. We show that identifying instances of outlier demand and adjusting the forecast in a timely fashion has the potential to increase revenue. Hence, we show that the use of outlier detection techniques as an assistant to revenue management analysts can help to minimise such losses, and we provide best-case-scenario analysis of potential revenue gains.

Keywords: revenue management; simulation; demand estimation; tolerance intervals; decision support