What is Marketing Mix Modelling?
Marketing Mix Modelling (MMM) is a powerful statistical approach used to measure the performance of marketing efforts by analysing the relationship between marketing inputs and business outcomes, such as sales, revenue or customer growth. Through marketing mix models, businesses can determine which marketing strategy channels — like TV, online video, press, radio or out-of-home — are driving results, and allocate budgets accordingly. It provides a clear view of which strategies yield the highest ROI, by breaking down each medium’s contributions.
The Importance of Data Quality
The success of MMM hinges on the quality of the sales data being fed into the model. Accurate, comprehensive, and clean data ensures that the model outputs are reliable and actionable. Inaccurate data, incomplete records, or poorly formatted datasets, can distort findings, resulting in misguided marketing decisions that can cost companies millions. So addressing data quality is not just an operational concern but a strategic imperative for businesses looking to gain an edge in an increasingly data-driven world. Mutinex has a number of solutions for this, but first, let’s understand a little more about the marketing mix modeling techniques and challenges landscape.
Understanding Marketing Mix Modelling
What is Marketing Mix Modelling?
A Detailed Definition of MMM
Where traditional Marketing Mix Modelling is based on regression analysis to understand how media channels have performed in the past, Mutinex GrowthOS leverages Bayesian inference. This technique is based on the Bayes theorem, which states that “the probability of an event occurring given certain conditions is equal to the probability of those conditions given the event occurred.” This means we can analyse historical data and predict how changes in investment will impact future results.
A Brief History of MMM
MMM has been around for decades, but its origins date back to the 1960s when companies first began using statistical methods to allocate their advertising budgets. Back then, the primary focus was on traditional media, such as TV, radio, and print. As digital marketing channels have grown in prominence, MMM has evolved to encompass online data, giving marketers a holistic view of their entire mix. Today, a marketing mix model is more relevant than ever as companies navigate increasingly complex omnichannel marketing environments.
Key Components of MMM
Data Inputs
The success of an MMM model depends on the data that goes into it. Key inputs typically include marketing spend data (how much was spent on each channel), media metrics (like impressions, clicks and reach), and business outcome data (like sales or leads). In addition, external factors like economic indicators, competitive activity, and even weather data can play a crucial role in improving model accuracy.
Many traditional MMMs use impressions, rather than marketing spend, as the key input variable representing channel activity. But impressions are not measured consistently across platforms; they can change over time, and they don’t directly reflect business outcomes.
By contrast, a good MMM should focus on marketing spend, which is closely tied to the bottom line. Mutinex — which has been operating since May 2022 and works with over 50 brands — recommends marketing spend as the key input variable for MMM modelling.
As an input variable, marketing spend yields considerable dividends — like financial alignment, cross-channel comparability, causal visibility, and data integrity. And even though spending more doesn’t always mean more sales, we account for this through a hierarchical time-varying model which delivers a more robust signal.
Statistical Methods
Traditional MMM delivered by consultants typically employs regression analysis, a statistical method used to model the relationship between independent variables (like media spend) and a dependent variable (like sales). More advanced models (like those deployed by Mutinex) might use Bayesian methods, time series analysis, or machine learning algorithms to handle complex relationships and improve predictive power. By leveraging these methods, companies can quantify the impact of each marketing channel and optimise their strategy accordingly. Mutinex deploys a generalised foundation model approach to MMM.
Types of Data Used in Marketing Mix Modelling
KPI or Metrics-Based Data
Metrics-based data includes key performance indicators (KPIs) such as sales volume, customer acquisition cost, and revenue growth. These data points are essential for measuring the effectiveness of a marketing campaign and understanding how marketing efforts translate into business outcomes. This data forms the foundation of the MMM analysis, as the model’s primary goal is to explain and predict these performance metrics.
Marketing Data
Marketing data refers to the input data related to the specific marketing activities a business has undertaken. This includes ad spend, impressions, clicks, conversions, media reach, and more. For MMM, it’s important to capture this data in granular detail so the model can accurately assign value to each channel. For example, marketing data for digital campaigns may include cost-per-click (CPC), impressions, and click-through rates (CTR), while offline marketing data could include TV airings, billboard placements, and radio spots.
Non-Marketing or External Data
Non-marketing data refers to external factors that can influence sales or business performance but are not directly related to marketing efforts. Examples include weather data (which can affect consumer behaviour), economic indicators (such as inflation or GDP growth), or competitor activity (which may affect market share). Including these factors in MMM helps account for variables outside of the company’s control that may otherwise skew results, leading to more accurate and reliable insights.
Common Data Challenges in Marketing Mix Modelling
Incomplete Data
Identifying Missing Data
One of the most common data challenges in MMM is incomplete data. Whether due to system errors, human oversight, or data collection gaps, missing data can significantly distort MMM results. It’s essential to identify where data is missing and determine if it’s random or systemic, as different types of missing data require different solutions.
At Mutinex, our MMM is robust for small amounts of missing data, with sophisticated predictive modelling to fill in the gaps and make accurate estimations.
Solutions for Incomplete Data
Data Imputation
Data imputation is the process of filling in missing values with plausible data points, using statistical techniques. This method can help maintain the integrity of the dataset without discarding valuable records, but it needs to be done carefully to avoid introducing bias.
Forecasting
Another approach is to use forecasting models to estimate missing data based on historical trends. By analysing patterns in the data over time, businesses can predict the missing values, which can be particularly useful for seasonal industries.
Deletion
In cases where the missing data represents only a small fraction of the total dataset, deletion of these incomplete entries may be a viable option. However, this approach should be used sparingly, as too much deletion can lead to biased results.
Proxy Creation
When no marketing spend data is available, Mutinex will typically advise creating a proxy marketing spend metric that’s aligned with the customer’s finance team. And as a viable alternative, we can also suggest choosing the most holistic signal for each channel from multiple inputs, including not just impressions but reach and engagement.
Lack of Measurement Standards
Short- and Long-Term Effects
Another challenge in MMM is distinguishing between the short- and long-term effects of marketing activities. While some campaigns might yield immediate sales, others (like brand-building initiatives), may take months or even years to fully realise their impact.
At Mutinex, we accounted for long-term effects by measuring the impact of brand equity, something that’s been notoriously hard to do — until last year, when we released a new Brand Equity feature into our GrowthOS platform.
Solutions for Measurement Standards
Unobserved Component Modelling can help isolate the long-term effects of marketing from short-term fluctuations by using statistical techniques to model them — separating short-term gains from long-term brand equity, so that businesses can make more informed marketing decisions.
Difficulty in Measuring Ad Content Performance
Challenges with Current Metrics
What’s more, current metrics (like impressions, clicks, and engagement) are often insufficient to fully capture the effectiveness of ad content, especially when the goal is to build brand awareness rather than drive direct sales.
Solutions for Measuring Ad Content
Surveys and Questionnaires
One way to gain a deeper understanding of ad performance is through surveys and questionnaires, which can capture qualitative insights about consumer perceptions and preferences. These insights can complement traditional metrics to give a fuller picture of how effective a piece of content is.
Model Bias
Introducing Bias
A bad model occurs when the model creator introduces bias, like assuming that display ads only have a short-term effect and TV ads only have long-term impact. By applying this kind of arbitrary data transformation, you can create a biased model that aligns with preconceived notions, rather than reality. Such assumptions lead to misspecified models, as the model’s structure predetermines which channels appear most effective.
Avoiding Bias
At Mutinex, we let the data speak for itself, and apply rigorous statistical method to identify the true drivers of marketing performance.
Multicollinearity
Identifying Multicollinearity
MMM is also challenged by a phenomenon called multicollinearity. This occurs when two or more independent variables in a regression model are highly correlated, making it difficult to determine which variable is actually driving the result. It’s a common issue, because marketing channels often work together and influence each other.
Solutions for Multicollinearity
Deleting Common Predictors
If two variables are too highly correlated, it may be necessary to remove one of them from the model to reduce multicollinearity and improve the accuracy of the results.
Ridge Regression
Ridge regression is a type of linear regression that applies a penalty to large coefficients in the model, which helps reduce multicollinearity by shrinking the coefficient estimates.
Advanced Data Solutions for Marketing Mix Modelling
Machine Learning Algorithms
Automating Data Processing
Machine learning algorithms can automate many of the data processing tasks that would otherwise be done manually, speeding up the MMM process and reducing the likelihood of human error.
Mutinex DataOS ingests data simply and quickly, so that the model creator can regularly enter clean, accurate data without having to burden their team or pay their media company for campaign tracking and data collection. This also avoids the need to give dirty data to a consultancy to analyse after-the-fact. We also run automated checks on the data entry to make sure it’s sufficient for analysis, and so it’s easier to see how best to label it.
Reducing Errors
By using machine learning, businesses can develop more accurate models that continuously learn and improve as more data becomes available, leading to fewer errors and better predictions.
Time Series Analysis
Routine Data Assessment
Time series analysis allows businesses to monitor their marketing data over time and track patterns, trends, and outliers. This can help marketers identify when changes in performance are due to marketing efforts or external factors, such as seasonality or market disruptions.
At Mutinex, we do Routine Data Assessment as part of our onboarding process — and our DataOS product can quickly flag an anomaly (like an unexpected spike in sales), allowing us to work with the model creator and identify missing data.
Converting Time Series Data
By converting marketing data into time series format, companies can analyse changes over time and make more informed decisions about how to adjust their strategies based on evolving conditions.
Mutinex’s DataOS product is designed to take in data and structure it for time series models. We make column mapping and labelling easy, and our API integrates seamlessly with existing media databases, even automating for refreshes.
Implementing Marketing Mix Modelling
Steps to Developing an MMM Model
Data Collection
The first step in implementing an MMM model is collecting accurate, high-quality data. This includes gathering all relevant marketing inputs, business outcomes, and external factors that might affect the business results. Proper data collection ensures that the model has a strong foundation on which to base its findings.
Model Building
Once the data is collected, the next step is usually building the model. This involves selecting the right statistical techniques, such as regression analysis or machine learning algorithms, and defining the relationships between variables. The goal is to quantify the impact of each marketing channel on the business outcomes.
Most MMM companies build the model from scratch for each customer and business state, but when your business changes (and it will) you’ll need to rebuild the model to remove built-in bias and overfit. So Mutinex takes a different approach: a generalised ‘foundation’ AI model that understands the principles of marketing across all businesses and business states, learns about the company’s marketing and business, and saves time and money.
Validation and Calibration
After building the model, it must be validated to ensure it accurately represents the real-world dynamics of the business. This often involves using historical data to check the model’s predictions against actual results. Calibration is also necessary to fine-tune the model and account for any changes in the market or business environment.
Good model creators know that complex models can fit data well within the modelling framework but fall apart in the real world. If a model doesn’t have a clear way to validate the results, it’s a bad model. Which is why Mutinex always validates the model’s results in a way that reflects reality. We make sure we can get results outside the model’s framework, through methods like backtesting, experiments, or lift tests.
Tools and Software for MMM
Overview of Popular Tools
There are several popular tools for running MMM, including software like SAS, R, and Python for custom modelling (completed in house by highly trained data scientists and engineers), as well as dedicated MMM platforms like Nielsen Visual IQ, Google’s MMM solutions, and Mutinex. Each tool has its strengths and is suited to different business needs, from basic regression analysis to advanced machine learning techniques like the ones Mutinex employs across their suite of platforms.
Integration with Marketing Platforms
Many MMM tools can integrate with existing marketing platforms, such as Google Ads, Facebook Ads, and CRM systems, enabling businesses to streamline data collection and automatically update their models with new data. This integration allows for faster, real-time insights into marketing performance. Mutinex DataOS is the only platform designed specifically to automatically ingest and organise marketing data for MMM.
The Benefits of Addressing Data Challenges in Marketing Mix Modelling
Improved Accuracy
Better Predictions
Addressing data challenges like incomplete data, multicollinearity, and the lack of measurement standards leads to more accurate predictions in MMM. By improving data quality, businesses can get a clearer understanding of which marketing channels are truly driving value.
Enhanced Decision-Making
Optimised Campaigns
With improved data accuracy and model reliability, businesses can make better-informed decisions about how to allocate marketing budgets, optimising campaigns for maximum ROI. This enables companies to reduce waste in underperforming channels and double down on the tactics that work.
Future Trends in MMM
The Role of AI and Machine Learning
As AI and machine learning technology advances, it will play an increasingly important role in MMM by automating data analysis, improving the accuracy of models, and allowing for faster insights. This will enable businesses to stay agile and adapt their strategies in real time. Which places companies who already use Mutinex’s AI-based solutions ahead of the curve.
Churn and retention modeling
Current MMM solutions only model incremental revenue and sales, which means they’re largely focussed on new customers. But there is huge demand from brands to be able to model customer churn and retention value against revenue and sales. Mutinex is currently exploring solutions to meet this demand.
Emerging Data Solutions
As new data sources emerge, such as real-time social media insights and behavioural data from IoT devices, businesses will be able to enhance their MMM models with more granular and precise data. This will lead to even more refined marketing strategies and improved ROI.
Conclusion
Key Points
Addressing data challenges in MMM is crucial for improving the accuracy and effectiveness of the model. By overcoming issues like incomplete data, multicollinearity, and lack of measurement standards, businesses can make better marketing decisions and optimise their strategies for better results.
Final Thoughts
As the marketing landscape continues to evolve, companies that are able to address their data challenges with advanced solutions, such as machine learning and time series analysis, will be well-positioned to thrive in an increasingly competitive market.
FAQs
In a nutshell
What is MMM?
Marketing Mix Modelling (MMM) is a statistical approach used to measure the effectiveness of marketing efforts by analysing historical data and correlating marketing inputs with business outcomes.
Why is data quality important in MMM?
Poor data quality can lead to inaccurate model results, making it difficult to optimise marketing strategies effectively.
What are the common data challenges in MMM?
Incomplete data, multicollinearity, difficulty in measuring ad content performance, and lack of measurement standards are common issues that need to be addressed.
