The Science of Internet Advertising

Given the number of requests that I have received lately from firms, especially from start-up firms, to write about them in this blog, my daughter suggested that I start charging for such "services."

Since I am a researcher and educator, for the time being, I will instead note some research of ours, which has just been published. The research is on the general theme of the Science of Internet Ads, which we have been pursuing with co-authors. Our latest paper, hot off the press is, "An Integrated Approach for the Design of Optimal Web Banners," and it appears in volume 11 of the journal Netnomics (2010), pp. 69-83. My co-authors are Professors Lili Hai and Lan Zhao of the Department of Mathematics and Computer Information Science at SUNY Old Westbury.

In this paper, we target the very first step of Internet advertising -- that of banner design using the tools of statistical analysis and optimization.

While banner advertising has become prevalent, consumers have also become more selective. Indeed, the banners' click-through rate, which is the ratio of the number of click-throughs to the number of exposures (times that the banner is shown to Internet surfers), has declined precipitously to an average of less than one-half percent.

In order to be fully effective with banners, a scientifically sound approach using real time data is needed to determine an optimal design.

Our paper demonstrates how to use a statistical predictive technique and optimization methods to exploit the richness of data collected on banner visitors' activities in order to achieve the goal of optimizing the banner designs. The optimization procedure begins with establishing a banner information repository that complies with database technology.

The above data may change for each advertising cycle. After the establishment of the information repository, the statistical predictive model is constructed based on the data. It not only identifies the significant components, but also quantifies the contribution of each component to the click-through rate of a banner. The predictive model sets click-through rate as the function of banner components. Finally, mixed integer programming is used to maximize the click-through rate as a function of the feasible set of components.

The major benefits of our method are that it allows one to systematically improve banner advertising by capturing the dynamics of browsers and to "unintrusively" personalize web advertising at the cluster-level.

In our first paper on Internet advertising, published also in Netnomics, in 2005, Zhao and I modeled optimal Internet advertising strategies for allocating an ad budget to websites as a network optimization problem, and constructed an efficient special-purpose algorithm for the determination of the optimal solution. We then explained two different paradoxes that occur in this setting.

In a paper of ours published in the European Journal of Operational Research, the network optimization modeling framework was expanded to model Internet advertising competition in which multiple firms maximize their own ad effects within their limited marketing budgets. In that paper, we introduced an elastic Internet marketing budget in order to conjoin the online and offline marketing strategies. Consequently, the multifirm competitive equilibrium problem was modeled using game theoretic constructs as a Nash equilibrium with network structure.

Professor Lan Zhao is a Center Associate of the Virtual Center for Supernetworks that I direct, and she also consults for flowers.com so our research results have immediate practical applications.

For reprints of various articles, conducted by researchers at the Virtual Center for Supernetworks please click here.