CARBS.
We have seen plenty of Marketing Mix Modelling (MMM) projects. Some changed how budgets were set. Others were technically excellent and never cut through. The difference is rarely the methodology. If you are about to commission MMM, there are five things to get right before the model build begins. Think like a long-distance athlete – focus on the CARBS.
- C
Commercially focused.
Most Marketing Mix Modelling work we see is presented as a backward-looking evaluation, weakly linked to the commercial decision it is meant to inform. That usually comes back to the wrong question being asked in the first place. The first thing we ask of any MMM brief is the simplest one: what would actually change if the answer were different. If the honest reply is "not much", the next several months will be spent on a study that earns nothing.
- A
Accessible.
Not every analyst knows how to share insights honestly and well. Not every brand manager is comfortable turning a model output into a decision. Accessibility is the work that sits between the two – plain English, ranked findings, and the single chart that ends the meeting. Without it, MMM stays in the appendix; with it, MMM moves budget.
- R
Repeatable.
A one-off MMM was the right tool in the 1990s. Treating it like a one-shot study today leaves most of the value on the table. The interesting work is making Marketing Mix Modelling part of business-as-usual: refreshable, comparable across reads, and tied to the planning calendar rather than written for an award paper. Better processes around the model usually beat a better model.
- B
Beyond media.
MMM was developed as a way to explain demand. Media is one input. Pricing, distribution, promotions, the trading calendar, competitor activity, and the wider macro environment sit alongside. Models that ignore these tend to be overconfident about the variables they can see. The question is rarely whether TV is working – it is whether your model is being asked to explain the right things in the first place.
- S
Sceptical.
Scepticism is not cynicism. Cynicism assumes the answer is wrong before it looks; scepticism simply refuses to take it at face value. The useful questions are usually the ones everyone in the room has quietly agreed not to ask – why this adstock rate, why this control variable, why this number moved and that one did not. A model that has never been questioned has never really been tested.