Introduction
Despite being an old technique that is already well known in the market, Marketing Mix Modeling (MMM) has made a comeback in the Business Intelligence environment. This is because Machine Learning techniques have made it possible to greatly improve the accuracy of these models. In addition, the Multi Touch Attribution (MTA) methods that were at the forefront of marketing analysis, with the huge increase in digital media in companies’ media share, began to present some limitations, mainly related to user privacy.
In view of the return of MMM to the vocabulary of marketing analysis, we have tried here to give a general explanation of how this model works, and some of its advantages, especially in a cookieless environment.
How MMM works in general
Marketing Mix Modeling is an analysis technique that uses Machine Learning to understand the relationship between a collection of events (independent events) and a specific event (dependent event). For example, if I have, in a given period, investments in Online, Out of Home (OOH) and TV media, and in the same period a Y number of sales, the MMM algorithm will seek to understand the relationship between the variation in the investment values of each of the media with the variation in the Y value. If the Online media is one of the main media responsible for convincing people to buy that product, the MMM model will give it a greater “weight” than the other media, and so we will know that it has a greater influence on our Y variable.
What the model actually identifies is how changes in online media investment data, over time, have a greater effect on sales than the other media (OOH and TV). With this result, we could come to the conclusion that Online media is more relevant when we think about the distribution of media investments.
The specifics of how the MMM works
Beyond this simple conceptualization, the MMM stands out for three other characteristics:
- The inclusion of context variables,
- Consideration of the delayed effect of the impact of events
- Measurement of saturation in the effect of an event.
The inclusion of context variables
Context variables refer to external factors that interfere with our dependent variable. If a technology company, for example, launches an innovative product, and the mere fact that this product is launched greatly increases the interest of potential buyers in the brand, then MMM would be able to understand this impact, without mistakenly attributing it to the company’s media mix. On the other hand, we can think of a retail company that is impacted by a truckers’ strike, thus hindering deliveries; this would be an effect with a negative impact on sales that is not related to the media. The inclusion of external factors in an MMM model therefore helps us to understand the effect of the media in isolation, without there being any noise in the attribution of this effect.
Consideration of the delayed effect of the impact of events
The second characteristic concerns a fundamental premise: the effect of independent events on the dependent event is not necessarily immediate. In practical terms, we can think of awareness campaigns for a new brand. Generally, such campaigns don’t have an immediate impact on sales, but as they make the brand better known, making potential customers consider its product as a viable solution to their problems, there will probably be an increase in sales. This delay in the effect of a campaign, called the carry-over effect, is taken into account by the MMM model, and helps in understanding the effects of top-of-the-funnel media, such as awareness or branding campaigns.
Measuring saturation in the effect of an event.
The third characteristic is already known to many media operators and concerns the fact that the relationship between media investment and return is not linear. This means that there comes a time when an increase in investment in a medium no longer brings an equal increase in return. We call this effect investment saturation, which indicates the point at which the increase in return from this media, given the increase in investment, is no longer advantageous.
In summary, the MMM is a model that takes into account both media campaigns and external factors to understand the influence of each of these events on the dependent event, which will usually be sales or revenue. In addition, the model takes into account the complexity of a campaign’s effects over time, including the possible delay in its impact and its saturation point.
Conclusion
MMM is a very advanced and robust technique. Although it has some requirements in terms of data maturity and technical knowledge for its application, it has significant advantages, mainly because it goes beyond the digital environment and takes external events into account. In addition, as we have seen, there is an effort to model the effect of each media, so that we take into account the particularities of each one in terms of its impact on sales.
The market offers ready-made solutions for applying MMM, which can make its implementation simpler and faster. However, such solutions are not usually reliable for more complex environments, because the productization of a solution such as MMM requires, to some extent, standardization in the way it is applied. This standardization can prevent particularities related to the market context, the business rules and the data used from being taken into account. Therefore, these more complex contexts usually require a model to be built by a technically trained team, otherwise the results may be less relevant or reliable.
Other content:
Profile of the Author: Renan Trindade | Graduated in social sciences and now a data scientist. A constant researcher of the universe of data and machine learning, but no less curious about other branches of the sciences, be they exact or human.
