Title

Decision models for wireless advertising

Date of Completion

January 2003

Keywords

Business Administration, General|Business Administration, Marketing

Degree

Ph.D.

Abstract

This dissertation develops models and solution methodologies for the delivery of wireless advertisements on mobile devices. Our models analyze the problem from the perspective of a wireless advertising service firm (WASF), an intermediary between the wireless carrier and sellers who wish to advertise their products and services. The WASF pays the carrier for capacity usage and charges sellers for ad deliveries. ^ Though there is literature on the importance of location and time of ad delivery to ad effectiveness, and literature on modeling mobility, there is very little that puts these streams of work together to improve targeting of wireless advertising. This is perhaps because wireless advertising is still in its infancy. The contribution of this research is in two areas modeling wireless advertising and developing solution methodologies to solve these models. We first consider alternative approaches to modeling constraints on wireless capacity and willingness to accept ads, prospect mobility, ad targeting efficiency and privacy concerns. For solution methodologies, we consider two models—a single day model, and a model for multiple days with response tracking and learning. In the first model, the WASF is paid for ad delivery, and in the second model, an additional bonus paid for responses. This differentiation between the two models is similar to recent developments in Internet advertising where the emphasis is shifting from best effort ad delivery to result oriented payment. ^ Both models use a Markov Decision Process (MDP) methodology and use information on the prospect's real time location and historical mobility information, and information on capacities, willingness to accept ads, and preference information to determine the time and location when ad is delivered. In addition, the second model updates the probability of future response using past history on contacts and responses. Since MDP models can only solve small problems, we develop efficient heuristics to solve larger problems. We test our heuristics against an upper bound we develop, and compare performance using Monte Carlo simulation techniques. We also analyze the impact of various parameters used in the model. ^