← Back to Curriculum

For the last decade, Digital Marketing has been obsessed with tracking. We dropped cookies, we appended UTM parameters, and we built massive data warehouses to stitch together the path of User ID #12345 from a Facebook Ad to a purchase.

This is the "Bottom-Up" approach (Attribution). It assumes that if we can't see the link, it didn't happen.

But the world has changed. iOS14, privacy regulations (GDPR/CCPA), and cross-device fragmentation have blinded us. Tracking is dying. This is where Econometrics—specifically Marketing Mix Modeling (MMM)—steps in.

1. The Fundamental Shift

MMM does not care about User ID #12345. It does not care about cookies. MMM is a "Top-Down" approach. It looks at the world in aggregates.

2. The Core Equation

At the heart of every MMM, no matter how complex the Python code becomes later, is a simple linear regression equation. We are trying to explain Sales (the dependent variable) using a set of Media and Control variables (independent variables).

Let's break this down:

• Salest: Your KPI (Revenue, Orders, Installs) at time t (usually weekly).
• α (Alpha): The Baseline. How much would you sell if you stopped all marketing today? This represents your brand equity and organic demand.
• β (Beta): The Coefficients. This is the magic number we are solving for. It tells us: "For every $1 spent on Facebook, how much incremental Revenue is generated?"
• Controls: External factors. Seasonality, Price changes, Competitor spend, Macroeconomics.
• ε (Error): The noise. The part of reality our model cannot explain.

3. Correlation vs. Causation

The danger of MMM is confusing correlation with causation. Just because you spent more on TV in December and sales went up in December, does not mean TV caused the sales. It might just be Christmas.

To build a "Good" model, we must ruthlessly separate these signals. This is why Feature Engineering (which we cover in Phase 2) is critical. We cannot just dump raw data into a model; we must transform it to represent reality.

The Input Reality

Before we write code, we must visualize our dataset. In Python, your ABT (Analytical Base Table) will look like this:

Date        Sales    FB_Spend    TV_GRPs    Price_Index    Holiday_Flag
2023-01-01  $50k     $10k        150        1.0            1
2023-01-08  $35k     $12k        0          1.0            0
2023-01-15  $42k     $11k        120        0.9            0
...

Notice there are no User IDs. There are no clicks. There is only Time and Intensity.

4. The Goal of this Masterclass

Over the next 11 modules, we will build a production-grade MMM pipeline. We will not use "black box" automated tools. We will write the Python code to:

1. Transform raw spend into Adstocks (Memory).
2. Model diminishing returns using Saturation curves.
3. Solve the equation using Ridge Regression and Bayesian priors.
4. Optimize the budget to maximize ROAS.