Mastering Media Mix Modeling With Google Meridian MMM

In today's marketing, understanding the effectiveness of each dollar spent is crucial for maximizing return on investment (ROI) and crafting strategies that truly move the needle.

Enter Google Meridian MMM (Media Mix Modeling), a potent tool for today's savvy marketing professionals and digital marketing analysts.

Positioned at the forefront of media strategy optimization, Meridian offers an evolved approach to media planning and performance analysis, drawing on the foundation laid by the pioneering Lightweight MMM model.

The Evolution from Lightweight MMM to Meridian

Google's introduction of the Lightweight MMM model in 2017 marked a significant advancement in Hierarchical Bayesian Modeling, addressing the complexities of measuring and optimizing media mix.

This open-source initiative was designed to provide marketers and analysts with a robust framework for dissecting the multifaceted relationships between media expenditures and significant business KPIs, such as weekly sales performance.

Explore the original Lightweight MMM research and its GitHub repository for a deep dive into its methodologies.

Meridian, now emerging in 2024 with limited availability, builds upon this foundation, offering enhanced capabilities for navigating the intricate dynamics of today’s marketing ecosystems. Its goal remains firmly rooted in maximizing the effectiveness of the Media Mix Modeling process.

Unpacking the Lightweight Model

The core equation of the Lightweight MMM,

$kpi = \alpha + trend + seasonality + M + C$

reveals a (non)linear framework where media-related KPIs ($M$) and control factors ($C$) combine to forecast outcomes like weekly sales. However, the real game-changer is understanding the multiplicative nature of these models and the profound influence of incorporating prior knowledge to refine predictions and strategies.

• Trend Analysis: Distinguishing between background sales levels sans media spending and regional minimum sales levels underscores the nuanced effects of market dynamics and consumer behavior over time.
• Seasonal Adjustments: Tailoring the model to capture seasonal variances accurately is critical, especially in industries with pronounced cyclical trends.

Fundamental Concepts in Measuring Media's Impact

Media Spend

Effectively modeling media spend starts with establishing clear definitions and baselines, focusing on the media channel expenditures over time and recognizing the multifaceted dimensions of advertising impact, including impressions and cost-per-click metrics.

Key Concepts Underpinning Effective MMM

• Lag Effect: Acknowledging the variance in how quickly different advertising mediums influence sales is essential for accurate modeling.
• Carryover Effect: Recognizing the accumulative impact of continuous advertising spending over time enriches the model's predictive accuracy.

Lag Effect

The lag effect is the speed at which sales react to media. In OOH advertising, such as billboard advertising, the spending effect may not become readily visible in sales. However, Sales respond quicker to Instagram ads.

Carryover Effect

The carryover effect describes how the impact of media advertising accumulates over time as a result of continuous spending.

Some models use the Attribution Window concept, which combines the Lag effect and the Carryover effect.

Diminishing Returns

The Diminishing Returns (DR) in Media Spend law explains that as media spending increases, the incremental value received from it declines. In other words, media reaches a saturation point, and as a consequence of DR, reaching a higher sales target by solely increasing media spend becomes increasingly harder (learn more about DR here)

The DR law applies not only to the channels but also to the total marketing spend.

Let's consider a firm that allocates $10 million in marketing & advertising. The company's total revenue is approximately $325 million, resulting in an ROI of $32 for every dollar spent.

At this level, they're delivering +16 ppt incremental value over the baseline—$275 m revenue, in this case.

If nothing else is changed (launch a new product, increased competitor pressure), how much additional marketing spend is needed to deliver the same incremental value? The answer is $100M as the result of DR. This is exactly marketing spend optimization in Performance Marketing.

Control Variables: The Catchall for Remaining Anomalies

After accounting for baseline, trend, seasonality, and media spend impacts, control variables serve as the analytical Swiss Army knife, capturing all other factors influencing the outcome.

Reach and Frequency KPIs

The use of reach and frequency is one of the innovative aspects of Meridian. Note that this feature is only applicable to modern media like Google Ads and is rarely useful for traditional media.

MMM models typically rely on impressions as input (like in Google Lighwtweight MMM), but ignore the fact that a person can be exposed to ads multiple times, and the impact can vary depending on exposure frequency.

Meridian offers the option to model the effect of any media channel based on its reach and frequency data. This method potentially produces a more precise measurement of marketing impact.

Also, it's important to note that the addition of reach and frequency leads to a more complex MMM model and, hence, the usual issue with the data size in MMMs.

Incremental KPI - A Causal Definition

One of Meridian's great aspects is its intersection with Causal Modeling. In Meridian, the MMM model generates Counterfactual predictions.

So, instead of performing experiments like holdout or A/B testing, Meridian offers a creative way to measure the causal impact of Media spend on incremental sales (target KPI).

However, this doesn't eliminate the need for experimentation. We should still perform experiments at regular intervals to ensure that we have sufficient evidence to calibrate the MMM model.

Incremental sales is defined as the difference of sales with and without channel spend:

$Y_{m=1} - Y_{m=0}$

The $Y_{m=1}$ is the value of sales when we spend on an arbitrary channel $m$. The $Y_{m=0}$ is the predicted sales when no spending was made on the channel. We use the MMM model to predict the value of $Y_{m=0}$. In Causal modeling lingo, $Y_{m=0}$ is known as potential outcomes

Choosing Control Variables

The Meridian uses a Causal Graph to formalize how we should choose the Control variable.

According to the Meridian's Causal Graph documentation, control variable Z must satisfy the backdoor criterion.

In the simplest terms, the variable Z's interaction with Spend and Sales must have an upward-pointing V-shape.

Let's look at examples of valid and invalid control variables:

• Valid: Black Friday sales. Because the increase or decrease in sales doesn't define when Black Friday happens.
• Invalid: PR and sales are correlated, but the causal direction is bidirectional (reciprocal causation in Causal language). A PR can lead to higher sales, and higher sales can lead to a PR.

MMM Data Platform

Google's MMM Data Platform provides access to various data, including Google Query Volume (GQV), reach and frequency, and paid search data. The GQV is added to the MMM as a Control Variables (See Control Variables section above).

Although Meridian does not require GQV data, including GQV helps to get an unbiased measurement of the ROI of the Paid Search channel.

The next figure shows the data schema for MMM Data Platform (see source):

Engaging with Google Meridian MMM

Now, as we stand on the cusp of broader availability, the opportunity to leverage Meridian’s advanced modeling capabilities beckons. To those starting their exploration or seeking to deepen their mastery, engaging directly with the Google Meridian MMM resources and community will be instrumental in unlocking the full potential of this powerful tool.

Conclusion

The world of digital marketing is rapidly changing, and Meridian presents an opportunity to measure the impact of your media in a better way. With the improved insights and efficiencies that it offers, marketing professionals can not only justify their budgets but also significantly amplify their impact. This ensures that every dollar spent is a step towards achieving their organization's goals.

Take the Next Step

Don't miss the opportunity to revolutionize your marketing strategy. Start integrating Marketing Impact Optimization into your arsenal today. Join the forefront of marketing professionals who are leveraging this advanced tool to make data-driven decisions that propel their organizations forward. Dive into our resources and begin your journey toward marketing excellence. Your path to impactful media mix modeling starts now.