What is marketing mix modeling and when is it worth it?

Marketing mix modeling is how you divide an entire ad budget across every channel at once, including ones you can't cleanly isolate or tag. It takes two years of weekly spend and revenue data and splits your sales into named pieces: how much was base (revenue that would happen anyway), how much came from Meta, Google, quick-commerce, and so on.

The core insight is that MMM models two effects: adstock (the fact that an ad you run today sells for days after), and saturation (the tenth rupee spent on a channel converts worse than the first). Together these curves tell you where your marginal return crosses your margin-the point where the next rupee actually makes money.

When to use it: You earn MMM when you're spending across four or more channels, several of which you can't test one at a time, and you have roughly two years of genuinely varied spend history. Most Seed to Series A brands don't have that yet. If your spending is flat week to week, the model learns nothing.

The critical caveat: An uncalibrated MMM is just a correlation engine and will confidently hand you wrong numbers. It becomes trustworthy only when you feed it causal ground truth from incrementality tests-geo holdouts that prove what each channel actually causes. Without that calibration, don't bother building it.