I increasingly see that more is being talked about ‘How MMM can be done in 1 day’ rather than how the problems of multicollinearity and endogeneity are solved.
Saas-ification in parts is good and even desirable. We ourselves have a SaaS platform ‘ArymaEdge’. But our SaaS platform does not try to fully automate the MMM process.
We still build the MMM the old fashioned way because the problems of multicollinearity and endogeneity can only be addressed through careful econometric modeling.
Why is the problem of multicollinearity and endogeneity so crucial to address?
Lets take an analogy.
Marketing attribution is a problem of identifying the Most valuable player/s (MVPs) in a team.
If you have multicollinearity, then each of the players ‘shout’ and say ‘it was their effort that led to the win’.
Now because of all this ‘shouting’, you don’t know who are the real ‘MVPs’.
So you decide to ‘omit’ some players. Now the shouting has reduced but it turns out that ‘omitted players’ were very crucial in various assists and each of these assists were in fact what led to the team’s win.
So now when you are trying to figure out who are the MVPs, the information that you are missing are the ‘omitted players’ who assisted the most.
So your attribution process is not robust.
This is what endogeneity and multicollinearity does to MMM. The very purpose of accurate attribution becomes difficult.
The problems of multicollinearity and endogeneity requires a solution that also has an human element (domain knowledge).
Having a well specified MMM model is the most important step in the attribution process because saturation curves, scenario planning and budget optimization etc. are all artefacts of MMM.
If you have a misspecified MMM model, then none of the saturation curves, budget optimization dashboards will really matter.
There are many ways to solve endogeneity and multicollinearity (check our website’s FAQ section and link in resources).
Resources:
- FAQ’s: https://arymalabs.com/marketing-mix-modeling/
- Multicollinearity and Endogeneity : https://www.linkedin.com/posts/ridhima-kumar7_marketingmixmodeling-marketingeffectiveness-activity-7023899326837858304-RtO3?utm_source=share&utm_medium=member_desktop
- How to solve multicollinearity through regularization : https://www.linkedin.com/posts/ridhima-kumar7_marketingmixmodeling-marketingeffectiveness-activity-7024606573452828673-jZCv?utm_source=share&utm_medium=member_desktop
