Marketing Mix Modeling is not the only answer

We are currently transitioning into the cookieless era, prompting marketing scientists to revisit Marketing Mix Modeling for addressing marketing effectiveness. Often hailed as the ultimate solution to marketing effects, Marketing Mix Modeling is undeniably valuable, but it falls short of being the perceived holy grail. To achieve reliable results we need to combine more methods, and discuss with marketers. Does this mean Marketing Mix Modeling is a guaranteed disappointment? Definitely not, but we need high quality modeling: Data Modelers need to step out of there data bubble and incorporate marketing knowledge in models. Marketing mix modeling is not your only answer.

Marketing mix modeling requires a lot of variance.

One significant challenge with Marketing Mix Modeling lies in its requirement for substantial variance. Whether using classical regression or Bayesian models, Marketing Mix Modeling necessitates periods with both high and low marketing spend. This poses a considerable difficulty, especially for brands employing an ‘always on’ strategy. The shift to Marketing Mix Modeling raises the question of how a model can learn the effect of paid search when there are no data points during certain periods.

Marketing choices are not random

Unlike random decisions, marketing choices are not arbitrary. Models require varied marketing decisions to accurately assess the added value of your marketing efforts. For instance, an online retailer might increase marketing spend before the Black Friday season due to market growth, and this can significantly impact marketing channels.

Channel spend correlates with market size

Moreover, certain marketing channel spends are correlated with sales, such as Google search spend, particularly in branded search, which tends to increase significantly with a larger market or a more popular brand. This correlation can lead to inflated estimates in Marketing Mix Modeling, as effects from different marketing sources become intertwined with branded search, distorting the model’s accuracy.

Marketing channels are executed simultaneously

For a lot of marketing strategies, marketing spend is executed simultaneously. Think of campaigns that utilize Youtube, Television and Radio at the same time. The marketing mix model has no observations of campaigns solely depending on one channel. And thus channel effectiveness is hard to obtain via marketing mix modeling.

Creative matters

Markting mix modeling uses information on a channel level. If there is a variance in creative, this can alter the effectiveness of a channel greatly. Since a Marketing Mix model is a time based model, not all variables can be added to the model. Even with a hierarchical structure this becomes undoable at a certain point.

To summarize

Markting mix modeling is a powerful tool, but also has strong disadvantages. Relying solely on a marketing mix modeling is definetly a step forward compared to linear or last click attribution. However it should just be your first step in the field. Combine more methods, experiment, examine user paths, use the touchpoints available.

Next to that, marketeers can greatly help modelers by experimenting. Changing setup of their campaigns, experimenting with new creatives, new channels, new setups and new spend methods. If Marketeers and Modelers can work together, incorporate the right models, the sun will start shining on marketing effectiveness.

The combination of these methods is the road to true marketing succes! How to use these methods can be found in our blog.