Marketing mix modeling (MMM)
Marketing mix modeling (MMM) is a statistical model that estimates what each marketing channel adds to sales from weekly totals, without tracking customers.
How it works
MMM takes weekly totals: sales, spend on each channel, and other factors such as prices, promotions, season and search demand. A regression then estimates how much each channel added and how its return falls as spend grows. It needs no user-level data, so it doesn’t depend on cookies or click tracking. Google’s Meridian and Meta’s Robyn are open-source MMM packages.
Both packages want at least two years of weekly data, and Meridian asks for three years for a national model. Spend also has to vary: if a channel’s spend stays flat, the model struggles to tell what it added. In our data, Performance Max runs in 141 of 146 stores and takes a median 95.7% of their Google Ads budget (GetProfit data, June 2025 – June 2026). So inside Google Ads the model has little to separate. Our guide to marketing mix modeling for an online store covers when the effort pays off.
Example
Example store, not client data.
The tableware shop has two years of weekly data: 104 rows. Robyn’s rule of thumb asks for 7–10 rows per input variable, so 104 rows carry about 10 to 15 variables (104 ÷ 10 = 10.4; 104 ÷ 7 = 14.9). Google Ads, Meta, email, price, promotions, season and search demand already make 7.
Not to be confused with
- Attribution model — splits the credit for each order between the clicks a customer made. MMM never sees single orders, only weekly totals.
- Incrementality — the sales that would not have happened without the ads. MMM estimates it from history. An experiment such as a geo lift test measures it and can calibrate the model.
Right and wrong readings
- Wrong: “In weeks with more search spend, sales were higher, so search drove them.” Right: when demand rises, people search more, so search spend and sales rise together. Meridian’s documentation calls organic query volume an important confounder for search ads.
Sources
- Meridian (Google) — open source, calibration with geo experiments. Checked 2 October 2026.
- Collect and Organize Your Data (Meridian) — weekly data, two or three years; query volume and seasonality as control variables. Checked 2 October 2026.
- Paid search modeling (Meridian) — query volume as a confounder. Checked 2 October 2026.
- An Analyst’s Guide to MMM (Robyn, Meta) — aggregated data, price and promotions as context variables, the need for variation in the data, two years, 7–10 rows per variable, calibration. Checked 2 October 2026.
- GetProfit data: 146 online stores, June 2025 – June 2026 — Performance Max share of Google Ads budget.