Forecasting Online Advertising Campaigns in the Wild

Online advertising is a high and a constantly growing business with an elaborate complexity. An important position is occupied by intermediaries, such as ad networks, marketing agencies and companies specialized in delivering software for managing and displaying advertising campaigns. Efficient use of advertising budgets and maximizing profits from ad impressions are the common goal of all players. This goal is usually achieved by manual management of advertising campaigns. Such management requires a very good prediction of Web users' behavior and web site traffic. We have tested the performance of an ad emission simulator for various campaigns and different types of constraints (e.g. capping or targeting). Real data from completed campaigns has been used for the experiment. It has been shown that predicting performance of advertising campaigns that have constraints is considerably more difficult than predicting simple campaigns. Our result applies even if we use a Web traffic sample from the current period (idealized Web traffic prediction) and a simulator that practically emulates ad server operation (the most realistic prediction approach).