Marketing Media Mix Models: Towards a Bayesian Framework

Abstract

Assessing advertising profitability is relevant for marketing decision making. However, this task is not trivial and no available methods until now can estimate advertising effect on sales perfectly. Marketing Mix Modelling (MMM) is a kind of modelling that applies accumulated lag effects (adstock) and non-linear transformation (saturation) for the estimation of how spending on advertising (adspend) impacts sales. On this Thesis we study different formulations for modelling those effects and we check the available open-source alternatives, namely Meta®’s Robyn and Google®’s Meridian. Moreover, we also propose our own solution, that we named 4M_MesioMMM. We confront those solutions through simulated data and present lines for further research on this field. We also review the transference this research has led to since it was performed under a Research Introduction scholarship and founded by the R+D project “Assessorament en el desenvolupament de models de Marketing Mix Modeling” under LOSU’s article 60 in collaboration with Adsmurai SL. All this leads us to conclude the need to be cautious with the solutions proposed by big tech companies, even if they are open source, and that the apparently most promising path for the future advancement of Marketing Mix Modelling (MMM) runs through the Bayesian framework.