Marketing Mix Modeling: a Journey Towards Awareness
Date
13 May 2025
Marketing Mix Modeling (MMM) is a methodology created to quantify the contribution of each marketing channel to business objectives, such as sales, in-store traffic, and customer acquisition.
In the past, MMM was mainly used to analyze television and print advertising, but today it has become an indispensable tool for companies operating in a multi-channel ecosystem and in an increasingly privacy-first regulatory context.
In recent years, the return to MMM has been driven by three main factors:
The end of third-party cookies and restrictions on user data collection.
The growing complexity of the media mix, with the integration of many different digital channels, influencer marketing, and retail media.
The advancement of AI and Machine Learning, which have made MMM models more accurate and granular.
Why is it important to talk about it? Because every company doing digital marketing today should adopt this methodology in order to optimize its investments and improve its reporting capabilities.
The History of Marketing Mix Modeling
Marketing Mix Modeling was born in the 1960s as a tool to evaluate the effectiveness of advertising campaigns through the analysis of sales and other business data. Between the 1980s and 1990s, it became a standard in the world of fast-moving consumer goods (FMCG).
In the 2000s, with the advent of digital, MMM lost centrality in favor of multi-touch attribution (MTA), which promised a more granular measurement of the user journey. However, with new privacy regulations and the increasing fragmentation of the media mix, MMM has regained relevance in recent years, adapting to new contexts thanks to the use of machine learning and causal analysis.
The CRISP-DM Process applied to MMM
To structure a Marketing Mix Modeling project effectively, the CRISP-DM framework (Cross-Industry Standard Process for Data Mining) can be adopted, a consolidated method in the data analysis sector. This is a very important framework because without method there is a risk of making gross errors, or even worse, abandoning the path because one gets lost in the labyrinth of complexity.
The 6 phases of the CRISP-DM process:
- Business Understanding
- Definition of objectives: what do we want to measure? Sales, brand awareness, lead generation?
- Identification of the marketing variables to analyze.
- Data Understanding
- Collection of data sources: CRM, sales data, advertising investments, market trends.
- Exploratory analysis to identify patterns and correlations.
- Data Preparation
- Cleaning and normalization of data.
- Creation of variables for the model (e.g., adstock, seasonality, promotions).
- Modeling
- Choice of statistical technique: multiple regression, Bayesian models, machine learning.
- Model training on historical data.
- Evaluation
- Comparison of model results with real data.
- Validation through incremental tests (e.g., geo-experiment).
- Deployment and Optimizationone
- Application of the model to marketing decisions.
- Periodic updates to improve accuracy.
MMM and Multi-Touch Attribution are complementary, not alternatives
One of the most common mistakes is to think that Marketing Mix Modeling (MMM) can replace Multi-Touch Attribution (MTA). In reality, the two approaches serve different purposes and should be used together.
| MMM | Multi-Touch Attribution (MTA) |
| Analyzes aggregated data | Analyzes user-level data |
| Great for long-term strategic decisions | Great for short-term tactical optimizations |
| Measures the impact of all channels, including offline | Measures only digitally traceable channels |
| Not affected by the loss of cookies and user data | Increasingly limited by the lack of user data |
So, what is the best approach? Take the best from each: use MMM for strategic decisions and MTA for daily optimizations.
The Fundamental Elements of MMM
To obtain reliable results, MMM uses key concepts such as:
- Adstock: measures the delayed effect of advertising.
- Saturation: identifies the point beyond which further investments no longer generate value.
Adstock can be modeled with different mathematical functions, including:
- Exponential Decay – Classic model, assumes that each day the effect of advertising decreases by a fixed percentage. Formula:
Adstockt=Spesat+λ×Adstockt−1Adstock_t = Spesa_t + \lambda \times Adstock_{t-1}Adstockt=Spesat+λ×Adstockt−1
Where λ\lambdaλ is the decay factor (between 0 and 1).
- Weibull Decay (Robyn and other advanced models) – More flexible model, allows modeling both slow dispersion and stronger delayed effect. The decay follows a Weibull distribution function, which better adapts to long-lasting advertising phenomena.
- Custom carryover models – Use machine learning techniques to dynamically estimate adstock for each channel.
One of the most common mistakes in marketing strategies is to believe that increasing the budget always produces proportional results. In reality, there is a point beyond which advertising effectiveness decreases: this phenomenon is known as the saturation effect.
How to calculate it? Thanks to Saturation models:
- Exponential Saturation – Follows a curve that shows decreasing returns as spending increases. Formula:
- Hill Curve (S-shape) (model used by Robyn and Meridian) – More realistic model, simulates a threshold behavior: the impact of advertising initially grows, then stabilizes. Formula:
An important aspect to keep in mind is that saturation is not simply a metric linked to the target or reach, nor is it an immutable metric over time. On the contrary, by working on creativity and strategy it is possible to reduce it and make investments more efficient.
Advanced Models: Robyn and Meridian
In recent years, Machine Learning has greatly improved the accuracy of MMM. Today, for example, the two main advanced open-source models are:
- Robyn (Meta) – Employs Ridge regression, a linear regression technique that helps manage multicollinearity among independent variables and prevent overfitting, thus improving the predictive performance of Marketing Mix Modeling (MMM) models.
Automation: minimizes human intervention thanks to automatic variable selection.
Advanced management of adstock and saturation: uses Hill and Weibull curves for better modeling.
Validation through incremental tests: allows verifying the validity of models with experimental data. - Meridian (Google) – Uses causal inference to analyze the true impact of marketing channels. The open-source framework for Marketing Mix Modeling (MMM) developed by Google uses a Bayesian causal inference approach. This probabilistic method combines prior knowledge with observed data to estimate the impact of marketing activities and quantify the uncertainty associated with forecasts.
What are these analyses practically used for?
We are often asked what these analyses are concretely useful for. The answers are multiple:
1. Measuring the efficiency of advertising channels
Let’s evaluate a practical example. The chart below shows a comparison between spend share and effect share for different advertising channels. This type of visualization is fundamental in Marketing Mix Modeling (MMM), as it allows evaluating the effectiveness of each advertising platform relative to its contribution to business metrics.
2. Evaluating the incrementality of advertising channels compared to exogenous factors
Another practical case, this chart instead shows the percentage contribution of different factors – both advertising and exogenous – relative to the overall result (for example, sales or conversions). This type of visualization is fundamental in Marketing Mix Modeling (MMM), as it allows quantifying how much each advertising channel and other external factors affect business performance.
Through this analysis, we can evaluate the efficiency of each advertising channel, separating the real effect of campaigns from the impacts due to seasonal trends, organic baseline, and holidays.
3. Calculating Adstock and Saturation for each channel
Let’s now see how this chart shows the delayed effect of advertising over time (adstock effect). This type of visualization is fundamental in Marketing Mix Modeling (MMM), as it allows evaluating how long an advertising action continues to influence user behavior.
- In the example of the chart, the curve highlights that the maximum impact occurs around the fifth day, suggesting that advertising has a cumulative effect that grows over time before reaching its peak.
- The Half-Life concept represented in the chart indicates the point at which the advertising response is reduced to 50% of the maximum effect, allowing the estimation of how long it takes for the advertising effect to wear off.
- This type of analysis is useful to optimize campaign frequency, avoiding excessive investments in periods when the advertising effect is still active.
The chart represents the relationship between daily advertising spending and revenues generated. This type of visualization is essential for Marketing Mix Modeling (MMM), as it allows evaluating the marginal effectiveness of advertising investment and identifying the optimal budget allocation point.
- The curve shows that as advertising spending increases, revenue grows, but with decreasing marginal returns.
- The Median Spend point marks the average investment threshold, providing a reference for testing budget variations.
- This chart is useful for analyzing advertising saturation, i.e., the level of spending beyond which further investments do not lead to a proportional increase in sales.
- It helps plan budget allocation scenarios, optimizing resource distribution on channels with the best incremental impact.
4. Future scenarios by making media budgets more efficient
The screenshot of the table above shows a comparison between the initial advertising spend and the optimized spend for different advertising channels, highlighting variations in ROAS (Return on Ad Spend) and daily budget. This type of visualization is fundamental in Marketing Mix Modeling (MMM), as it allows evaluating the efficiency of each advertising platform and identifying opportunities for budget reallocation to maximize investment returns.
The analysis clearly shows which channels could benefit from increased spending (such as Search, which shows significant ROAS growth with increased budget) and which instead are overspending (such as PMAX and other channels), which could be reduced to improve overall efficiency.
This analysis helps marketers optimize media budget distribution, avoiding waste and maximizing overall advertising strategy performance.
For example, from this forecast, an increase of €14,000 would lead to an increase of almost €90,000.
Marketing Mix Modeling Pills: rules to remember
MMM is a journey
Marketing Mix Modeling is not a static analysis, but an evolutionary process. It requires gradual implementation, based on historical data, incremental tests, and continuous optimizations. It is not enough to build a model and leave it unchanged: it is necessary to update it periodically to reflect changing market dynamics and consumer behavior. The goal is not only to measure the impact of advertising, but to create a data-driven culture within the company.
Know yourself
Before adopting MMM, it is essential to understand your own business model and key KPIs. Not all advertising channels play the same role: some generate direct conversions, others work on brand awareness and consideration. MMM helps answer questions such as: Where is my investment going? Which channels are really influencing my business? How much do discounts impact sales? But to obtain precise answers, one must start from a clear framework of objectives and metrics.
MMM and KPIs must speak the same language
One of the most common mistakes in MMM is to use metrics that are not aligned with business objectives. For example, optimizing a model based only on sales could lead to underestimating channels with a strong impact on brand equity. It is essential to build a model that takes into account all funnel stages, from impressions to conversions, in order to have a complete view of advertising impact.
Average saturation is your enemy, efficiency is your goal
Each advertising channel has a saturation point, beyond which further investments do not bring proportional benefits. MMM helps identify the optimal investment level for each channel, avoiding budget waste. Identifying the saturation point allows redistributing investments more efficiently, maximizing return. (Let’s remember, however, that creativity and strategy can help).
It’s not just about channels, but about context
Advertising effectiveness does not depend only on the chosen channel, but also on external factors such as seasonality, market events, macroeconomic trends. MMM takes these variables into account (if entered), helping to distinguish the real effects of advertising campaigns from those influenced by industry dynamics. This allows more accurate forecasts and the ability to adapt strategy based on future scenarios.
Experiment
MMM is not just an exercise in passive analysis, but an experimental approach. Budget simulations, A/B tests, and incremental experiments are fundamental to validate model hypotheses. A good approach to MMM requires a culture of continuous testing: Which changes in media mix lead to improved ROAS? Which channels can be scaled down without impacting performance? Only by testing is it possible to obtain concrete answers.
At SAY we do not just measure the past, we build the future
MMM is not only useful to understand what worked in the past, but above all to plan the future. Thanks to scenario simulations developed by SAY’s Performance team, it is possible to forecast the impact of different media strategies and optimize investments in advance. The ultimate goal is not only to maximize the efficiency of the current budget, but to create a data-driven and predictive approach to marketing decisions.
Do you want to implement MMM in your company?
Contact us for a personalized consultancy!
