Abstract [eng] |
Marketing Mix Modelling is used by advertisers to understand the effect of marketing channels spending on sales and optimize media budget. In most cases, marketing modelling accounts for lagged effects of advertisement and diminishing returns. This thesis proposed a marketing mix model with Bayesian approach, using prior knowledge and assumptions. In addition, Bayesian regression fits adstock and diminishing return parameters of functional forms in the regression which let decision makers learn about marketing channels’ features. The thesis goes through sensitivity check, posterior predictive check to assess model’s performance. The study finds that TV marketing channel is the biggest contributor to sales when comparing to other channels. However, TV has relatively low adstock rate indicating that impressions in previous weeks poorly directs to sales of current week. Moreover, according to learned diminishing returns parameters, TV is slow saturation channel which is effective as others only when high number of impressions is achieved. When calculating contribution to sales from the model, for advertisers it is applicable to calculate Return on Advertisement Spend which helps optimizing advertisement budget. |