Using Bayes Networks to Estimate Return on Marketing Investment

Presented at the 4th Annual BayesiaLab Conference in Nashville on September 29, 2016.

Abstract

Companies spend a large proportion of their marketing budgets on advertising without fully quantifying the sales return that results from their investments. This presentation uses Bayes Networks to estimate Return on Marketing Investment (ROMI). XYZ Company made available weekly sales and advertising figures for this study. In addition, VSA collected a large array of non-advertising factors that also potentially drive consumer sales of the product.

The first model concentrates on estimating store sales at the week level in the absence of advertising. Using batch inference, the model estimates what the store sales should have been if XYZ Company had not done advertising. The second model estimates ROMIs from direct effects of a model that includes only advertising factors as predictors of store sales, controlling for store sales in the absence of advertising.

Results show estimated ROMIs for different advertising channels for XYZ Company, which in turn have some strategic implications for advertising allocation decisions.

Presentation Video

About the Presenter

Charles Hammerslough is Director of Data Science at VSA Partners in Chicago, where he oversees a wide range of projects involving marketing and predictive analytics for a small advertising agency. Prior to VSA, he was Director of Modeling for the 2012 Obama for American campaign, and before that a VP of Research and Development at Nielsen. He obtained his Ph.D. in Sociology and Demography at Princeton University.