Media Mix Modeling helps brands and agencies determine which combination of marketing channels, and which tactics performed in each channel, ultimately produce the best results for their organizations. Media Mix Models typically include a time element — typically several years of current and historical data — allowing the best models to accurately predict which channels and channel tactics need to be expanded, pulled back, or modified for the best aggregate results.
Winston Li, founder of Arima, a Canadian company providing Media Mix Modeling as a service, has developed something called “The Synthetic Society,” which provides the data foundation for the models his company generates. Using a mix of more than 20 public and private data sources, a statistically accurate model of the Canadian population has been developed that allows highly granular modeling without impinging on user privacy. The availability of this model (now Canada-only but soon to come to the U.S.) will, according to Li, democratize media mix modeling by making the technique more accessible to more businesses who may have not have had the means to perform this kind of sophisticated modeling in the past.
Of course, even the best model will fail to deliver satisfactory results if its data inputs are unreliable, if the statistical data sample used to generate the model is too small, or if the model is not updated frequently enough to account for market or channel changes. Consequently, it’s necessary to review and update any model upon which one relies to ensure that the model accurately simulates actual media channel performance.
In the longer discussion below, Kevin Lee and Winston discuss other issues germane to media mix modeling, including Arima’s origin story, best practices for media mix modeling and statistical testing, the role of creative in media mix modeling, and other issues relevant to media mix modeling and digital marketing.