Marketing mix modeling looks at the business from above. Instead of following individual clicks, it takes weekly or daily totals (spend by channel, revenue, promotions, seasonality, pricing, sometimes weather or macro factors) and uses regression to estimate how much each input moved sales. It needs no user-level tracking, so cookie loss and iOS privacy changes do not break it.
Modern MMM tools, including open-source options from Meta (Robyn) and Google (Meridian), also model two effects that click-based attribution misses: carryover, where this week’s spend keeps driving sales for weeks, and diminishing returns, where each extra dollar in a channel buys less than the last.
Why it matters for eCommerce
Platform dashboards each claim credit for the same orders, and last-click attribution undervalues channels that create demand, such as video or upper-funnel Meta campaigns. MMM gives one view of all channels against total revenue, including organic and offline effects. That makes it useful for annual and quarterly budget decisions: how much to put into Google versus Meta, and where the next dollar earns less than it costs.
How it works in practice
A model needs history and variation. Most practitioners want at least one to two years of consistent weekly data, and spend that actually changed over time; a channel that spent the same amount every week cannot be measured. The output is a contribution estimate and a response curve per channel. Strong teams then validate the model with a geo holdout or conversion lift test and feed the result back in as a calibration point.
Common mistakes
- Running MMM on a few months of data, or on a store spending in only one or two channels.
- Treating model outputs as exact. They are estimates with ranges, and they should be read that way.
- Never calibrating the model against an incrementality test, so errors go unnoticed.