MMM: Is there enough variation in channel activity?
Is the advertising being well managed in the model?
MMM detects patterns by matching changes in media activity over time to changes in sales.
Imagine media spend in a channel, such as TV, which is the same for three years. Plotting this onto a graph would remain flat. Adding sales data to the same chart, which are likely to bounce around because of the many factors influencing sales (hopefully trending upwards!)
The job of the model is to link those sales movements to the media. Without changing spend in the media, there are no patterns to detect or connect, so the model looks for other data for a better explanation.
If you OOH advertising always runs at the same weight, timing and format, the model has nothing to connect to sales movement. No variation means no learning.
Does each channel show meaningful variation in spend, reach, timing or geography to estimate effects accurately?
YES!
There is strong variation.
Good - the model has contrast to work with.
Confirm the variation is genuine and not coincidental and request a visual check, such as a chart or graph, which will help give a feel for how the model is working.
NO!
Spend levels are relatively flat.
Without variation, impact can appear artificially low, or even invisible.
You can plan intentional variation for future campaigns or identify historical differences to add variation in the data to aid the model.
To understand more: