Media Mix Modeling, commonly known as MMM, has been a tool used by marketers to improve their marketing strategy by applying a statistical and heuristic approach to predicting sales as a result of changes in by blending various marketing channels across mediums.

MMM, unlike tracking and attribution, is not standardized across companies or even brands within the same company. Unlike marketing attribution, a media mix model will try and measure the impact of untrackable marketing mediums such as TV, Out of Home, Radio and so on.

What makes media mix modeling useful is that it does allow marketers to predict what will happen if they make certain changes within their marketing activities, as well as assist marketers in understanding the return over investment (ROI) across channels.

Media mix modeling rely on finding correlations across a large set of data. Fortunately though, MMM can survive without any user level data, which is why it gained so much popularity within the past years, as user-level data is becoming less available with privacy restrictions and regulations.

To create a solid media mix model, a marketer will need to gather as much historical data as possible including:

  • Any and all advertising activities that happened, broken down to the lowest possible granularity and timeframe (day level is ok).
  • Any other data that would have possible relevancy as an impact over marketing performance. This may include anything from Weather, product price changes, any major external influencers such as holidays, macro and micro economical factors, competitor activities, press coverage, PR activities and so on.

Each broken down by day or at least by week.

  • Sales or other consumer level KPI data to be used as the Output of the model, broken down by day.

Once the data has been obtained, an analyst will be required to conduct some exploratory analysis creating a correlation matrix (finding correlation between values for each pair of parameters within the data), as well as work with the marketer to create an hypothesis for feature values.

Feature Values in a media mix model refer to assigning weights to various factors, as both standalone, as well as combined factors.

For example – if the marketer believes that promotional discounts to consumers pose a higher influence on sales than the influence of weather – the model would need to accept this hypothesis in the form of creating a scoring system where discounts are factored higher than weather changes data.

Digesting the correlative data and enriching this with hypothetical scores will allow marketers to create “what if” questions and output a prediction for sales.

An example to such question would be: If ad spend would to be increased by 2x across TV during April 1st – 14th , while the product price to consumers will be discounted by 15% – what should be the impact over sales ?

The amount of historical and external data points needed to create an accurate media mix modeling is typically the main reason marketers often abandon the idea of building one.

MMM requires so much data and analysis, while proving that the model is correct, requires continuous testing of advertising campaigns which consumes enormous resources.

MMM does not work for new product launches, as a model requires large sets of historical data to create a prediction.

Causal data science methods (incrementality) have been giving up a fight to the older media mix modeling methods. Incrementality measurement does not require so much historical data, and more importantly – incrementality measurement can be standardized.

Incrementality may not be capable of answering what will happen if a marketer would make a change, but is able to analyze past data and create a retrospective prediction indicating if certain marketing decisions contributed to sales, had influenced positively (or negatively) other marketing activities, or if certain marketing activities had no impact over sales.

INCRMNTAL is one of the first software companies to offer incrementality as a service, utilizing causal data science, allows marketers to find the answers they looked for in media mix modeling, within the scope of a platform that performs measurements without any need of experimentation.

Data integration happens once, and takes less than a day. Once that is done, Marketers can use the platform regularly and measure any marketing activity that happened, in minutes.

By Manali

Leave a Reply

Your email address will not be published. Required fields are marked *