The growth of diverse digital platforms and extensive datasets have presented challenges in the field of advertising. When combined with signal resiliency issues, reduced tracking capabilities and data-sharing legalities, advertisers are turning to Marketing Mix Modeling (MMM), a widely accepted cross-channel measurement solution.
In this white paper, “Mixed-Methods MMM: The Utilization of Ridge Regression in Marketing Mix Modeling”, bynd Consulting (OMG Germany)* and Snap Inc. propose a two-stage mixed-methods approach to MMM, which combines traditional methods (OLS) with contemporary statistical and ML-related methods, such as ridge regression techniques.
This modeling approach would produce statistically robust output results while effectively breaking out digital channels in the process. For instance, a conducted database analysis by bynd Consulting reveals that social/messaging platforms are both competitive and efficient when compared to other digital channels.
Fictive ex.: Quantifying sales impact levels of individual online channels in MMM
Technically applying Mixed-Methods MMM, requires the creation of two separate models in a two-stage modeling process:
[1] The first model (reference model) uses traditional OLS methods. It quantifies non-media control variables and includes modeled, high-level digital bucket output results containing all available digital channels and platforms.
[2] The second separate model (auxiliary model) applies ridge regression techniques in order to break out digital channels/platforms in the process [2a]. We then combine both models. [2b].
This two-stage modeling approach ensures that the final model results meet typical industry-standard MMM requirements with respect to statistical robustness.
Modeling process of Two-Stage Mixed-Methods MMM, Source: bynd Consulting (OMG Germany)
Based on this developed MMM approach, a database analysis, based on 17 sales modeling cases containing social media channels, was conducted by bynd Consulting. These cases spanned four sectors (beverages, pharmaceuticals, FMCG, retail), with an annual gross total media spend ranging between 2.8m and 109m in the German market.
The results showed that social media and messaging platforms [ROAS index score of 189] are very efficient when compared with other online channels [OL Display [118] and OL video [100]] and only rank behind SEA [465]] which occupies a special position in the complex customer journey.
Average impact per contact, n = 17 cases; Source: bynd Consulting (OMG Germany)
Put simply, Mixed-Methods MMM is an effective way to quantify sales impact levels of various online drivers while ensuring statistical robustness which is crucial when running marketing mix modeling analyses. It can also be used to break out individual channels/platforms as well as digital ad products and activities such as AR, video, organic, etc in the process.
*bynd [bɪˈjɒnd] is the strategy and technology consultancy of Omnicom Media Group Germany.