Member-only story
Marketing Mix Models 101: When Your TV Ad Did Not Return 21x, but Christmas Did
I built a synthetic market where I set the real returns myself, then watched an ordinary regression hand the whole TV budget credit for the holidays
For this project, I told my simulated market that television returns 2.5 times what you spend on it. I know that is the true number because I wrote the market that produced it. Then I did what most analysts do with real data: lined up weekly sales against weekly spend, fit a regression, and read the television coefficient as a return on investment, the sales a channel brings in per unit spent.
The regression said 21.
That gap is the most common mistake in marketing analytics, and it is why channel returns collapse the moment someone moves budget on the strength of them. In my market, television spend climbs every December, exactly when sales climb for reasons that have nothing to do with advertising. The regression cannot tell those apart, so it gives television the credit for Christmas.
Below, I describe the model that recovers the right number, then shows three quieter ways the answer still fools you after the obvious bug is fixed. It runs on three years of weekly figures, 156 rows, and knowing the true answer in advance is what lets me measure how wrong each method is.
Let’s dive in!
