Move from analyzing the performance of your media mix to making decisions about your next budget allocations.
What is Marketing Mix Modeling?
Marketing Mix Modeling is a mathematical modeling technique that helps companies quantify the impact of multiple components of their marketing mix (e.g., advertising, promotions, price, etc.) on sales or other key performance indicators.
The goal is to identify which tactics are most effective and how budgets can be reallocated to maximize ROI.
In practice, the data set of a Marketing Mix model is composed of a dependent variable (turnover for example) and X independent variables (your marketing channels, exogenous factors, etc.).
What are the benefits of Marketing Mix Modeling?
The MMM is essential for several reasons:
Budget Optimization: It helps businesses determine where they should invest their resources to get the best return.
Measuring Effectiveness: MMM evaluates the effectiveness of each element of the marketing mix, allowing companies to understand which tactics work best.
Forecasting: With historical data, MMM can be used to predict how future changes in the marketing mix might affect sales.
Cookie-less: Traditional attribution methods are challenged by the weight of technological and legal restrictions on the use of personal data. The unique identifier attached to individuals (cookie, CRM ID, email address) has always been the key to reconciling the customer journey and measuring the profitability of a marketing lever. Marketing mix modeling does not need personal data to work and is therefore a more resilient option. In addition, data relating to marketing campaigns is now much more accessible than in the past, both in terms of speed and granularity. This new situation allows MMM models to work on much shorter time dimensions than in the past.
What are the main components of the marketing mix analyzed in the MMM?
When it comes to analyzing marketing mix effectiveness using MMM, here are the commonly examined components:
Revenue: Overall measure of sales, often the primary performance indicator.
Leads: People or companies potentially interested in your product or service, an early indicator of sales performance.
Sales: Number of transactions or volume of products/services sold.
Paid Marketing Channels:
Online: Such as paid search engine advertising (PPC), social media advertising, and display marketing.
Offline: Such as TV advertising, radio, billboards and magazines.
Free Marketing Channels: Such as SEO, email marketing, and word of mouth.
Product Catalog: Variety and types of products available, as well as adding or removing products.
Price Change: How Price Changes Influence Demand and Sales.
New offerings: Launch of new products or services and their impact on the overall mix.
Promotions: Special offers, discounts or incentives to boost sales.
Seasonality: How seasonal variations (like holidays or weather seasons) influence sales.
Brand Awareness: Brand recognition and positive perception can influence sales.
Exogenous factors: Any other external elements that could influence sales, such as economic trends, competitor actions or world events.
By using MMM, companies can dissect the relative impact of each of these components on their overall performance, allowing them to adjust and optimize their marketing strategy accordingly.
How is MMM implemented?
The implementation of the MMM generally takes place in several stages:
Data Collection: Gather data on sales, marketing spend, and other external factors that might influence sales (e.g., seasonality, competition, economy).
Modeling: Use statistical techniques to develop a model that describes how various elements of the marketing mix affect sales or other key indicators.
Validation: Test the model on a separate dataset to verify its accuracy.
Interpretation: Analyze the results to determine which elements of the marketing mix have the greatest impact on sales.
Optimization: Based on the results, adjust your marketing strategy to maximize ROI.
What are the challenges of the MMM?
To ensure the reliability and accuracy of Marketing Mix Modeling, it is recommended to respect a ratio of 1 independent variable (or predictor) for 10 observations (or data points) in your dataset.
This has important practical implications for the design and analysis of your statistical models:
Practical implications
Sample size : If you plan to analyze the impact of multiple marketing channels (say 10) on sales, you will need at least 100 data points (10 channels * 10 observations per channel) to build a robust model. This requirement helps ensure that your model has enough data to learn the relationships between the independent variables and the dependent variable without overfitting.
Variable Selection : When planning your model, you should assess the number of observations available and determine how many independent variables you can reasonably include. If you have a relatively small data set, you may need to limit the number of variables to meet the recommended ratio.
Quality and robustness of estimates : Meeting this ratio helps ensure that your model's coefficient estimates are reliable and robust. This means that the predictions made by the model are more likely to reflect true relationships rather than artifacts of the particular data used for training.
Handling high-dimensional data : In scenarios where many independent variables are available or desired, this principle encourages you to use dimensionality reduction methods (e.g. through clustering) or variable selection to focus your model on the most significant predictors, in order to maintain a healthy ratio and avoid overcomplexity.
Here's why this recommendation is important:
Reducing the risk of overfitting: Overfitting occurs when the model fits the training data too closely, capturing statistical noise instead of the true underlying relationships. Having enough observations relative to the number of independent variables helps minimize this risk, as the model can generalize better from the observed data.
Improved stability of estimates: The more observations per variable, the more stable the estimates of the model coefficients. This means that the model's predictions are less likely to be influenced by random variations in the data, leading to better reliability and precision.
Increased statistical power: Statistical power is the ability of a statistical test to detect a true effect when that effect really exists. A high ratio of observations per variable increases the statistical power of the model, allowing the detection of significant relationships between the independent variables and the dependent variable.
Validation and cross-checking: A sufficient amount of observations allows for the implementation of validation techniques such as cross-validation, which is crucial for evaluating the performance of the model on unseen data. This helps ensure that the model will perform well in practice, not just in theory.
Reduced uncertainty in predictions: With more data per variable, the uncertainty or confidence interval around parameter estimates and predictions is reduced, leading to more informed and accurate decisions based on model outputs.
How does MMM compare to other marketing analysis techniques?
There are other techniques like multi-touch attribution, which analyzes specific touchpoints in the customer journey.
While MMM focuses on the overall impact of different elements of the marketing mix, multi-touch attribution looks at individual interactions. The two techniques can be complementary depending on the needs of the company.
Understanding the difference between correlation and causation:
Correlation : A correlation between two variables means that they tend to change together, but it does not necessarily mean that one causes the other. If A and B are correlated, when A increases, B may also increase (positive correlation) or decrease (negative correlation). But this does not prove that A causes this change in B.
Causality : Causality involves a cause-and-effect relationship between two variables. If A causes B, then a change in A will produce a specific change in B.
Example: Take ice cream consumption and sunscreen sales.
Observation : During the summer months, ice cream sales increase, and sunscreen sales also increase.
Correlation : There is a positive correlation between the sale of ice cream and the sale of sunscreen during the summer.
Causality : However, buying ice cream does not cause buying sunscreen. They are not directly related in terms of cause and effect.
Explanation : The real cause of this correlation is a hidden variable: temperature.
During the summer, it is hot. People want to cool down by eating ice cream and, due to the increased exposure to the sun, they also buy sunscreen to protect themselves.
Temperature is therefore the underlying cause of both observed increases.
This example shows how essential it is to distinguish between correlation and causation, because the simple coexistence of two phenomena does not mean that one is the cause of the other.