The 4 Steps of Measuring With Traditional MMM

4-steps-MMM-detail

Terence Einhorn, Sr. Solutions Consultant in Sales

Published 12/06/2023

Along with Incrementality Experimentation, MMM (Media Mix Modeling) is an attractive attribution approach in that it doesn't rely on cookies or user tracking and is not affected by changing privacy regulations.

While many marketers have considered using MMM to understand their media's impact on sales and optimize their budgets, not everyone understands the details of how MMM works. 

Traditional media mix modeling,  correlative models that don’t leverage causal experimentation, can be broken down into four steps: data collection, model design, model calculation, and results/output. Let’s dive into each of these steps below.

Step One: Data Collection and Validation

MMM aims to isolate the impact of media on demand. This means that it needs to account for not only media drivers but also any other factor that could impact demand, such as: 

  • Pricing and promotions
  • Seasonality
  • In-store activities
  • Weather conditions
  • Significant one-time events
  • Competitive activity, including advertising and new product releases
  • Macroeconomic factors such as inflation, unemployment rates, national income, and GDP
  • Industry-specific factors

The team will collect historical data, broken out weekly (or even daily), for any variable that may have a significant impact on sales. These data are then transformed and consolidated into one cohesive dataset, aligned by desired model dimensions (e.g., by regions, weekly vs. daily, product groups, etc.)

This can be challenging, as it often requires extensive processing to align different data sets to the same dimensions, fill in data gaps, and remove erroneous and outlier data. It’s time-consuming, but accuracy and consistency are key to achieving reliable outputs.

Once processing is complete, the datasets will undergo quality analysis (validation) before receiving the final sign-off, and modeling can begin.

Step Two: Model Design

Once the dataset is complete, the next step is model design.

A modeler (Data Scientist) will load the data set into a modeling platform (Python, R, SAS, or SPSS, an in-house platform, etc.) and design a model structure.

There are various variations of MMM, but most models follow a similar multiplicative model structure:

mmm structure

Here xi, w represents the spend, impressions, clicks, etc., for channel i in week w, and i is the “coefficient” or measured quantity of interest representing the degree to which a change in the channel is spend/impressions/clicks corresponds to a change in sales for any week w. For more on MMM model structure variations, click here

Step Three: Model Calculation

Once the model structure is defined and the final dataset is loaded, the modeling software will run a statistical regression (or machine learning) algorithm. 

A regression looks through the dataset and observes how week-to-week variations in the independent variables correlate to week-to-week variations in the dependent variable ( i.e., sales). 

The more closely an independent variable, or “driver,” is correlated with sales, the higher its coefficient (or average impact on Sales) will be.

Step Four: Results and Output

Finally, the model will leverage the coefficients calculated in the regression process and use them to calculate the historical “contribution” of each media driver. 

There are several mathematical variations for calculating contribution, but the general approach is to simulate the expected loss to total sales when a channel is removed from the mix. This hypothetical loss in sales is the channel’s contribution to the business for a given period of time.

These contributions will then serve as the basis for ROI calculations, optimizations, insights, and recommendations. Dividing this contribution by the channel’s spend for the corresponding time period yields its ROAS (Return on Ad Spend).

Traditional MMM: Considerations to Keep in Mind

While MMM can help drive data-driven budget allocation decisions, traditional MMM is unreliable if not supplemented with Incrementality testing. MMM measures the correlative impact of media on sales, but Incrementality unearths the causal impact. 

As many media drivers are inherently correlated with sales, and with one another, in-market Incrementality testing is critical to inform MMM reads wherever possible.

To see how Measured can help you with incrementality testing and MMM, schedule a demo today