Don’t get the wrong impression! Assessing the best input variable to reflect Meta in MMM: Abstract

Background

Meta campaigns have multiple buying options. This affects the cost of media, campaign delivery and the impact on business outcomes. The combination of buying options makes accurately estimating Meta in Marketing Mix Modelling challenging.

Previous research by Business Science Warsaw How to split Facebook in MMM? has demonstrated the value of granular modelling for Meta. The research recommended including splits by campaign objective and/or platform. They drive the most actionable insights for the media planners and resulted in the biggest improvement in model.

Granular modelling is the gold standard, but it isn’t feasible for every project. Sometimes we must use one input variable to estimate the impact of Meta’s family of apps. In this research we assess the best input variable for advertising on Meta’s family of apps (i.e. Facebook, Instagram, Messenger, Audience Network).

Meta’s MMM feed provides two variables, impressions and spend broken down by numerous dimensions (day, DMA, etc…). We have also created and assessed a new variable, equivalent impressions, which adjusts impressions by the cost of media for key buying options. This research evaluates how the input variable affects model performance.

The Marketing Mix Modeling meta-analysis scope

Our analysis of 145 models shows nuanced findings. Any of the 3 variables considered

  • Impressions
  • Equivalent impressions
  • Spend

can perform best in some situations. The new variable, equivalent impressions, seems to be the safest choice.

When splitting is not possible, our recommendation would be to test all 3 types in any specific case and choose the one that performs best in your model. If that is not possible, or you wanted to use a consistent variable across all projects, then equivalent impressions are the safest choice.

Recommendation

Thus, we recommend the following cascading choices for selecting the Meta variable:

  1. Build granular models and split Meta impressions variable by campaign objective and/or platform.
  2. When not possible , compare how total impressions, total equivalent impressions and total spend perform in your model. Choose the one that’s best in your specific case.
  3. If for some reason you can’t make a separate decision for any given model or compare all 3 options, then the safest choice is equivalent impressions.
Recommended Posts

Contact Us

We're not around right now. But you can send us an email and we'll get back to you, asap.

Not readable? Change text. captcha txt