What is the business facing these days?
In today’s competitive business environment, optimizing budget allocation to increase profitability is crucial for any company. Modern advertising not only represents a significant marketing investment but also influences audiences in multifaceted ways.
With recent technological advancements, the diversity of advertising channels has grown substantially, exposing customers to brand messages across numerous platforms and occasions. The data generated from these interactions is invaluable, especially for executing a data-driven marketing strategy. Therefore, we place a great emphasis on using this data to drive value for our clients. Our goal is to provide them with concrete guidelines on media channel investments, helping them uncover potential and improve their performance.
MTA? Or MMM?
As brands adopt increasingly diverse marketing tactics, the task of sorting, arranging, and analyzing data in this complex market has become more challenging. Two primary methods for this cross-channel marketing analysis are Multi-Touch Attribution (MTA) and Marketing Mix Modeling (MMM).
MTA is an approach that tracks user-level interactions to identify patterns in customer behavior. However, this method now faces significant hurdles due to the rise of privacy concerns and the phasing out of third-party cookies. It is heavily regulated, particularly under European data protection laws like GDPR.
Therefore, MMM has emerged as a popular alternative. By operating at an aggregated level, it avoids using individual user data while still providing valuable insights to guide business decisions. As a result, MMM is widely used today, despite having been developed in the 1960s.
What is Marketing Mix Modeling? How to use it?
In the context of complicated strategies, MMM is a powerful tool for understanding how various channels influence purchasing habits and quantifying their impact on performance. It is widely recognized as an econometrics framework for developing data-driven solutions. With a wide mix of online and offline channels available, understanding which marketing efforts are truly driving growth is critical for any business.
MMM is a powerful statistical technique for measuring the true contribution of each marketing channel to a company’s KPIs. However, measuring this impact is complex. The effect of an advertisement isn’t always immediate (the adstock effect), and spending more doesn’t always yield proportional results (the diminishing returns or saturation effect). This complexity is compounded by consumer behavior, which can vary significantly across industries such as FMCG, luxury, and retail. Despite the challenges of aligning models with this behavior, several advanced techniques exist to ensure a more accurate analysis.
Several ways of doing MMM
There are several options and packages available for implementing MMM. These options range from basic multiple linear regression, which is the simplest approach but requires significant manual effort to translate consumer behavior into model parameters.
To address this complexity, more advanced tools have been developed. For instance, Meridian by Google uses Bayesian MMM to better predict and measure behavioral patterns. Similarly, Meta offers the Robyn package, which employs Ridge Regression and algorithmic optimization for more robust analysis. Furthermore, other open-source libraries support these efforts, such as PyMC-Marketing and SciPy.
Why Google Meridian?
To tackle these challenges, we use Meridian, an open-source Bayesian MMM tool from Google. We chose this tool due to its seamless integration with the Google ecosystem and our use of BigQuery for data storage.
This approach is powerful because it directly models real-world complexities. Built on a Bayesian framework, Meridian allows users to customize the model with multiple settings that align with their specific data and industry context. Moreover, through MCMC simulations and model diagnostics, Meridian confirms model convergence, which is essential for a reliable interpretation of its output and it can:
- Quantify each channel’s incremental contribution, the significance baseline and ROI.
- Capture the time-delayed effects of advertising (adstock).
- Model the non-linear saturation of channel performance.
Despite its value as an analytical tool, there are still limitations. First, it operates at an aggregated level, which makes it difficult to measure individual customer journeys. Additionally, the model relies on historical data, posing a challenge for companies with limited data or for new market entrants. These limitations, however, can be addressed to inform future strategic planning and model improvements.
The results output and its strength (brand equity and budget allocation)
The following plots illustrate the key findings from our MMM analysis, revealing the contribution and relative Return on Investment (ROI) of each channel.
In MMM, the “baseline” represents the sales or KPI generated without any marketing spend. This is a powerful metric, as it serves as a strong proxy for the organic demand driven by brand equity. Furthermore, because the goal was to provide a data-driven basis for future strategic decisions, we have detailed the contribution percentage of each channel.
The model goes beyond simple attribution to:
- Decompose Sales: Clearly separating the baseline (brand equity, organic effects) from paid media contributions.
- Guide Future Investment: Combining contribution data with ROI and Marginal ROI (mROI) analysis to optimize budget allocation.
Moreover, due to measuring the actual consumer behavior across time, one of the most powerful techniques for MMM is the response curve. By modeling real-world complexities like adstock and diminishing returns, these curves provide a clear map of marketing effectiveness. Google’s MMM tool, Meridian, shows the response curve and provides budget allocation suggestions for each media channel. These curves are valuable for strategy because they show:
- Saturation and Current Position: The saturation level of each channel and its current position on the curve.
- Growth Potential: Which channels have room to grow and helps identify optimal allocation points for future strategy.
This level of analysis provides a solid, data-driven foundation for strategic discussions. After all, a true marketing strategy isn’t just about optimizing the next ad click; it’s about strengthening the foundation your brand stands on.
However, a key feature of this Bayesian MMM approach is that the ROI is presented not as an exact number, but as a range known as a “credible interval.” This feature quantifies uncertainty and reflects the reality that marketing performance is never guaranteed – a crucial point for strategic business discussions.
This deep-dive analysis, utilizing Google’s Meridian, represents a modern approach to marketing measurement. It offers significant insights that allow a company to build on its strengths and shape its future marketing blueprint. While there may still be limitations or areas for improvement worth discussing, these guidelines are invaluable for framing future engagement and prompting the right strategic questions, such as: “Do we invest aggressively in a high-risk, high-reward channel, or allocate more budget to channels with lower, but more stable and predictable, returns?”
To learn more about how we can optimize your data for precise marketing insights, check out our Applied Marketing Data Science Services. Our team is here to help you leverage advanced analytics, streamline your data processes, and enhance your marketing strategies.