How is Facebook measurement different when modeled over a shorter time period than over a longer period in MMM: Abstract
In the rapidly changing ecosystem of digital touchpoints, it’s of utmost importance for the Marketing Mix Modeling (MMM) providers to be able to quickly deliver the most up-to-date results of media efficiency. This allows for a close to ongoing media-mix optimization. Shortening the modeled period from the standard 2 years to 9 months could be a promising way to make MMMs more capable of this. In this paper, we investigate the consequences of shortening the modeled period on statistical and predictive properties of the estimated econometric models, as well as on business outcomes.
This is a case study from a particularly fast-paced telecommunications category in a European country. It appears that in this case, models estimated over shorter periods got as solid statistical properties and better predictive properties compared to models estimated over longer periods.
Also, the business outcomes of the shorter period models were more up-to-date than of the longer period models.
We conclude that models estimated over the shorter periods might provide an additional boost to clients’ businesses compared to standard MMMs. The shorter period models might become an industry standard in a few years to come. In order to estimate shorter period models the right way, analysts should pay close attention to the methodological frameworks in which the modeled period is shortened. This includes ensuring enough degrees of freedom, using prior knowledge to avoid accidental estimates of some variables, being particularly cautious with AdStock level estimation and following other MMM best practices mentioned in this paper.