Adding more campaigns to a Marketing Mix Model increases statistical uncertainty around each estimate, making results less actionable. Start with roughly 30 channels to balance granularity with reliability, not hundreds of micro-segments that blur the signal.
You probably think that the more campaigns and channels that you can estimate with your MMM, the more useful it'll be to drive your business. But actually it's often the opposite. Here's the granularity trade-off that every MMM faces. The more channels that you add to the model, the more uncertain it will be about each one. Now, of course, everybody using an MMM wants to be able to make as many decisions with as much precision as possible about how they allocate spend. But it's an iron law of statistics that the more predictors that you had, the more uncertain it will be about every single one. So if you take this to the limit and add in every single campaign on its own, every single sub-channel on its own, you will end up with an MMM that is uncertain about everything and has unacceptably wide uncertainty intervals. So the first step in building an MMM is Finding the zone of agreement between what breakdown of your spend is going to lead to effective decision-making such that the MMM can get an effective read on each channel. At Recast, we find that 30 channels is a really good place to start. It provides a lot of detail about how your media mix is performing, and it can drive concrete actionable decisions. And the small channels, the 30th, the 29th channel, they're still getting a pretty good read. So that's where we typically like to start. It's not a hard rule. Some models have less, some models have a little bit more, but that is where we think the granularity trade-off is best fit.
The Continuous Testing Loop That Separates Great Marketing Teams From Average Ones
