Are we approaching the era of Marketing Mix Modelling Industrialisation?

Posted by Sirine Karray on Feb. 2, 2018


According to Forrester’s article “Vendor Landscape: Marketing Measurement And Optimisation Solutions”, today’s marketers live in a world dominated by more-powerful technology, growing data volume, and increasingly data-driven focus. In this complex multi-channel landscape, advertisers are pressurised to quantify the business impact of their investments and update their knowledge by taking a fresh look at current measurement approaches.

This is where Marketing Effectiveness Measurement Methods come in. Marketing Effectiveness is defined as the measure of how effective a given marketer's go to market strategy is towards meeting a specified set of goals.

Today’s advertisers generally turn to attribution and Marketing Mix Modelling methods (MMM) to help them understand and quantify the true impact of their marketing efforts be it online or offline.

The main difference between Marketing Mix Modelling and Attribution is that the former is an aggregated top down approach whereas the latter is a disaggregated bottom up approach.

MMM, when compared to Digital Attribution, is decades old but remains largely a service-driven exercise. In fact, one of the major hurdles of making MMM more accessible to advertisers is the lack of stand-alone affordable software products. This constitutes a major barrier to entry, restricting marketing effectiveness analytics to the consulting firms who developed their own approach and preventing the adoption of these methods by the advertisers themselves.

In the roundtable dinner debate event organised by MASS Analytics on the 18th of January 2018 that gathered marketing effectiveness experts in the UK, the perspective of MMM industrialisation was debated.

It transpired that three steps are essential to reach MMM democratisation:

  • Adopt a proven process
  • Ensure efficiency through using the right technology
  • Embrace the right modelling methods


Adopt a Proven Process

Cross Industry Standard Process for Data Mining (CRISP-DM) CRISP-DM breaks the process of data mining into six major phases:

CRISP-DM

MMM is part of the Data Mining/ Machine Learning family and should adhere to the above process. Therefore, the first step towards MMM industrialisation is to make sure to deploy a solution that respects this sequence so the whole process flows naturally and delivers to the business.

Ensure efficiency through using the right technology

Adopting a proven process is a necessary move in the right direction but is not sufficient. It is crucial to use a technology that encompasses the CRISP-DM process. This will ensure that the right layer of efficiency is added to guarantee a much quicker turn around and make the whole process immune to human errors. Four features are essential in the technology used to ensure MMM industrialisation:

  1. An end-to-end delivery
  2. Speed of Execution
  3. Smart Automation
  4. Ease of Updates


Embrace the Right Modelling Methods

The third requirement for MMM industrialisation is to use modelling methods that are capable of representing the true consumer behaviour and adapt to the changing consumers ‘path to purchase in a multi-channel world. In that regard, we think three leaps are essential to make when it comes to adopting innovative modelling methods that better represent the complexity of today’s markets:

Additive Modelling

  • Log-Linear Modelling provides more realistic interpretations and representation of reality as it distinguishes between relative variables (like price, distribution…) and incremental variables (like media, competitors’ activities). Log-Linear also allows to get a read on elasticity and measure the synergy between the different variables. In addition, Log-Linear models address the relative impact on the target performance indicator (say sales) which is the main priority of decision makers as compared to absolute levels. These models can also deliver contributions’ decomposition like the additive model when the right approximation algorithms are used.

National Modelling

  • Geo level Modelling typically adds more granularity and variability to the data which makes the models gain in robustness. Besides the statistical robustness, geo-level modelling (usually using pooled regression) allows to measure regional or store level activities that cannot be captured by aggregated National Models. One could also use this methodology to pool customer segments to assess whether the consumer response to marketing activities differs by segment.

Indep. Models
  • The move from independent modelling to Interactive Modelling (or nested modelling) is a must do in a world dominated by multiple screens, multiple touch points and a complex media landscape. In addition to identifying the main drivers for the modelled KPIs, Interactive modelling allows to quantify the relationship between Paid, Owned and Earned Media, adjust the sales contribution and give more realistic interpretations since it measures the omitted variables impact. The benefits of using interactive modelling is also to integrate digital planning across Paid Owned and Earned Media & optimise media spend based on a comprehensive analysis of all touch Points.

In this article, we tried to summarise our thoughts on MMM industrialisation. We do think that MMM industrialisation is possible providing one could act on the three cogs described above: Process, Technology and Methodology. This will make MMM not only affordable but also accessible to all advertisers regardless of their size and enable them to internalise the Marketing Effectiveness Measurement Capability.


by Dr. Ramla Jarrar, Co-founder and CEO at Mass Analytics

Learn more about MassTer, and let us help you uncover the bright side of your marketing ROI. Request your FREE TRIAL now.