Market Mix Modelling

Market mix modelling is a statistical analysis technique use to asses the impact of different factors on sales of the product or service and leverage it for most efficient resource allocation with improved marketing effectiveness.

Factors here are mostly trade factors such as price, features, geography; promotional factors in terms of media and non-media channels of target and competitors product/service are taken into consideration.

In Marketing Mix Modeling (MMM), establishing a robust relationship between sales volumes and various marketing activities involves several critical steps:

Data Collection:

Gathering comprehensive data at both national and geographic levels is essential. This includes information on sales volumes, trade promotions, media expenditures, and other relevant variables. Collecting data across different regions and time periods allows for a more accurate analysis of how various factors influence sales.

Feature Engineering:

Marketing activities often exhibit non-linear relationships with sales. To model these effectively using linear regression techniques, appropriate transformations of the variables are necessary. For instance, media spend variables can be transformed using methods like the 'adbudg' transformation, which accounts for diminishing returns of advertising spend or 'adstock' transofmations which account for carry over effect of media spend. This process ensures that the linear regression model can capture the true impact of media promotions on sales.

Modeling:

To quantify the contributions of media, non-media, and trade variables to sales, regression models are employed:

• Ordinary Least Squares (OLS) Regression: This method estimates the parameters by minimizing the sum of squared residuals, providing unbiased and consistent estimates when the error terms have constant variance and are uncorrelated with the predictors.

• Mixed Linear Regression Models: These models extend OLS by incorporating both fixed effects (common to all observations) and random effects (specific to individual units, such as different geographic regions). This approach is particularly useful when dealing with hierarchical or grouped data, allowing for more precise estimation of the impact of marketing activities across different segments.

By following these steps, businesses can develop a nuanced understanding of how various marketing efforts influence sales, enabling data-driven decisions to optimize marketing strategies.

Automated Feature Selection Process:

Feature selection in market mix modelling is very crucial step as the product/service sale is impacted by different factors which are highly correlated with each other. 

We can use forward/backward feature selection process with careful addition or removal of features. The careful selection process can be evaluation of set of features with metrics like target correlation, sign, priority, regression coefficient and its significance, VIF.   We can also use causal AI techniques with directed graphs to select the features which are directly contributing to sales volume.