Course contentsShow
← Data Science
Lesson 1738 of 2,14541. Marketing and Growth AnalyticsFree lesson

The Core MMM Regression Model

Building a regression framework where sales are predicted from marketing spend across channels over time.

The Core MMM Regression Model

What you'll learn: How to structure a regression model that predicts sales based on multiple marketing channels' spending over time.

At its heart, Media Mix Modeling uses multiple regression to answer: "How much did each marketing channel contribute to sales?" Unlike attribution (which tracks individual user journeys), MMM works with aggregate data—weekly or monthly totals across your entire business.

The Basic Framework

The core equation looks like this:

Sales(t) = β₀ + β₁·TV(t) + β₂·Radio(t) + β₃·Digital(t) + ... + ε

Where:

  • Sales(t) is your outcome variable at time t (weekly sales, conversions, etc.)
  • TV(t), Radio(t), Digital(t) are your marketing spend amounts in each channel at time t
  • β₀ is the baseline sales (what you'd get with zero marketing)
  • β₁, β₂, β₃ are coefficients showing each channel's effectiveness
  • ε represents unexplained variation (noise)

Think of it like baking: if you track how much flour, sugar, and butter you use each week and measure how many cookies you produce, regression helps you learn how each ingredient contributes to output.

What the Coefficients Tell You

A coefficient of β₁ = 2.5 for TV means: "For every $1,000 increase in TV spend, we expect 2.5 additional sales (holding other channels constant)." This lets you compare channel efficiency directly.

The model is fit using historical data—typically 2-3 years of weekly observations—using ordinary least squares or related techniques.

Key Takeaway: The core MMM regression model predicts aggregate sales as a linear combination of marketing channel spend over time, with coefficients revealing each channel's marginal contribution to outcomes.