Marketing Mix Modeling (MMM) is a statistical method used to assess the impact of marketing campaigns on key performance indicators like sales, conversions, and traffic. Traditional MMM uses regression techniques, but the new Bayesian approach, exemplified by Google’s Meridian, offers significant advantages. Bayesian MMM provides distributions over parameters rather than point estimates, allowing for a better understanding of uncertainty. This method is particularly useful with smaller datasets, as it can incorporate prior knowledge to compensate for data sparsity. Meridian also models adstock, where marketing spend effects accumulate over time, and saturation, where the impact of spend diminishes at higher levels.
In a practical example, Meridian was used to analyze weekly sales data, considering holidays as control variables and marketing spends from five channels: Newspaper, Radio, TV, Social Media, and Online Display. The model showed a clear seasonality with sales peaks around Christmas and a trend of decreasing newspaper spend with an increase in social media spend. After fitting the model, it was found that the highest ROI came from Social Media, followed by TV, although with significant uncertainty.
The model suggested a 3% revenue increase by reallocating marketing budgets, reducing spends on Newspaper and Online Display while increasing on Radio, Social Media, and TV. However, this optimization does not account for channel interactions or future baseline changes, highlighting the complexity and potential of MMM in guiding marketing strategies.
Source: medium.com















