Marketing Mix Modeling without the Coca-Cola budget

Marketing Mix Modeling used to live in two places: at Coca-Cola, and in pitch decks selling you a six-month consulting engagement. Open-source libraries and cheaper compute have changed that. MMM is now a tool a mid-sized ecommerce team can actually run.

What MMM is really for

Attribution tells you which click closed the sale. MMM tells you which spend created the demand. Those are different questions, and last-click attribution has been answering the wrong one for a decade.

Attribution measures credit. MMM measures cause.

The lean MMM stack

• Two years of weekly spend, by channel, including offline.
• Weekly revenue or, even better, weekly gross profit.
• External signals: seasonality, promotions, weather if relevant, category trend.
• An open-source library, Meta's Robyn or Google's LightweightMMM, to do the heavy lifting.

Three traps that kill MMM projects

• Modeling revenue instead of profit. You will over-credit discount-heavy channels.
• Ignoring saturation curves. Doubling spend on a channel rarely doubles the return.
• Treating the output as truth instead of a hypothesis to validate with incrementality tests.

What MMM is good at

Answering the questions your CFO actually asks: what happens if we cut Meta by 30%? What is brand actually worth? Where is the next dollar best spent?

If you already have two years of clean spend and revenue data, you are closer to running MMM than you think. If you do not, that is the first thing to fix.