Adsmurai 
Gaining insight into the effectiveness of marketing strategies and optimizing your spend accordingly has always been a key task. By understanding which marketing inputs are most effective in driving sales, businesses can allocate resources more effectively and make better decisions about where to invest their budget.
TABLE OF CONTENTS
What is Marketing Mix Modeling?
One of the main challenges that marketers face is the number of channels that exist today and the direct impact they have on business results. Knowing how to quantify that impact, especially in strategies that include online and offline channels, is a complicated but essential task to know what works for a business and what does not.
This task is relatively easy when dealing only with digital media. Through the different conventional attribution models, such as a "last click", we can quantify conversions in a simple way. However, with these models we have a very limited view of advertising actions. They only analyze the impact of one channel and do not allow us to have a holistic view. Nor do they allow us to track all activity related to offline media.
In this context, it is necessary to develop new tools that will allow us to have more information about the effects of marketing actions and thus be able to make more informed decisions. This is where Marketing Mix Modeling can help us.
Marketing Mix Modeling or MMM is an advanced statistical methodology that, through the relationship between the different levers of a business, solves the doubts related to the impact of the different marketing levers, online and offline in the sales curve of a company.
Some of the questions that an MMM model can answer are:
- What is the optimal level of spend for each of the main marketing channels?
- What was the ROI for each marketing channel?
- How would sales be affected if I made a certain change to my marketing budget?
Through historical data, regression techniques and experimentation, Marketing Mix Modeling allows us to find out the contribution of each channel to a company's KPIs. By applying these models correctly, we will be able to know how changes in budgets, seasonality or even what is the optimal level of spending in each channel will affect them.
Why is Marketing Mix Modeling important?
As your company's marketing mix becomes more complex and incremental, understanding how each investment channel (such as Meta, Google, TikTok, etc.) contributes to improving target KPIs such as sales becomes more challenging. Each platform will report a certain amount of sales, but when we add them up, we often find that the total is greater than the sales recorded in our CRM, indicating discrepancies and duplication between channels. This is due to the attribution windows of each platform, which claim the same sale if the user interacted with ads in more than one channel.
Marketing Mix Modeling (MMM) emerges as an advanced solution to this problem. By correlating proprietary and historical data, MMMs predict with high reliability the real contribution of each channel to conversions. Not only does it allow us to better understand the contribution of channels such as Google, Meta and TikTok to sales, but it also evaluates the impact of other factors such as the holiday calendar, seasonality, events, promotions, competition, and our offline actions.
With this information, it allows much more informed and optimal budget allocation decisions to be made than with limited, static models such as last-click.
Measuring marketing effectiveness is essential for a company to know the impact of the different marketing actions it carries out. With this evaluation we will obtain valuable information that will help us to optimize marketing campaigns and improve the return on our investments.
One of the main advantages of measuring the effectiveness of a marketing strategy is that it allows companies to identify which actions are generating better results. Information from metrics such as lead generation or sales allows companies to determine which channels are generating the greatest return on investment and modify budgets accordingly.
Marketing Mix Modeling (MMM), through statistical analysis models, is crucial to measure the effectiveness of marketing campaigns and determine the impact of different marketing actions on sales. Here is a list of four advantages of incorporating an MMM model in your marketing plans:
-
Improved decision making
As we have seen, Marketing Mix Modeling provides companies with complete information on the performance of different marketing campaigns, enabling them to make more informed decisions.
-
More accurate attribution
Move beyond simplified models such as the "last cllic", integrating multiple channels and factors for accurate attribution.
-
Strategic planning
Insights facilitate long-term planning, enabling strategic adjustments based on historical data and projections.
-
Increased ROI
By identifying the most effective marketing actions and channels, Marketing Mix Modeling can help companies increase their ROI. For example, by increasing resources on those channels that are most effective and reducing them on those that have less impact on objectives.
-
Inclusion of offline data
Offers the ability to integrate offline data, which is not possible in most models, to measure overall marketing impact.
-
Incorporation of external factors
Allows the inclusion of external factors that affect results, such as economic context, weather, special dates, etc.
-
Adaptation to change
Uses real-time data and the flexibility of these models allows for updates with new data, facilitating adaptation to market and brand changes.
-
Continuous improvement
Encourages constant optimisation of marketing strategies, identifying opportunities to improve results over time.
-
Understanding the contribution of branding
Enables you to better understand and explain the contribution of actions and channels focused on branding or different conversion objectives. For example, assess how TikTok campaigns contribute to brand awareness and reach objectives, providing a more complete assessment of their effect beyond direct conversions.
-
Actionable results
From the MMM model we will obtain a cost curve, in which we will be able to see where each of the investment channels is and if it still has room to deliver positive results. By analyzing historical data and simulating different marketing campaigns, companies can predict the impact of different strategies on future performance.
-
Better targeting
Marketing Mix Modeling allows us to incorporate into statistical models all investment channels, including offline or traditional ones, and other external factors that may have an impact on results (economic moment, weather, special dates, etc.).
MMM models are a powerful tool for companies seeking to improve the effectiveness of their marketing campaigns and optimize their budgets. By providing information on the impact of different marketing actions and campaigns on business results, these models can help companies make more informed decisions on how to allocate their marketing resources for maximum impact.
Data collection and preparation for Marketing Mix Modeling
Both to feed the model with data and to be able to squeeze the most out of the information returned by the model, it is essential to collect and prepare the data we are going to use.
Collecting quality data will ensure that both the information we extract from the model and the decisions we make as a result will be based on reliable and accurate data. Poor quality data can increase the risk of error and lead to incorrect conclusions, reducing the effectiveness characteristic of MMM models.
Proper data preparation streamlines the data analysis process, reduces the time and effort required to analyze the data, and improves data understanding leading to more informed decisions and better results. Choosing and collecting data is very important, but don't forget to normalize the data to fit the model as well.
Data normalization is the process of transforming raw data into a standard format that eliminates redundant or inconsistent data and reduces data redundancy. The goal of normalization is to eliminate data anomalies and improve data consistency, accuracy and reliability. In this way, the data can be easily queried and analyzed.
In short, data collection and preparation is vital to ensure the accuracy and reliability of the information obtained in MMM models. Now that you know all the previous process, find out what are the main types of data needed for Marketing Mix Modeling:
- Conversion data
This includes information about the main objective of the campaign. For example, sales data, leads, etc. It is important to have these to accurately measure the impact of marketing actions.
- Ad spend data
This includes information on how much money you have spent on different marketing channels and actions, such as marquee advertising, browser ads, social media campaigns, etc.
- Relevant market data
Includes external factors that can affect sales, such as economic indicators, competition and consumer trends. This type of data will help us understand the overall market environment and how it affects your sales.
- Brand context data
In the data collection process it is interesting to include information about the brand context: events, promotions, discount campaigns, number of stores, etc.
- Experimental data
In this case, it is the information that will allow us to calibrate the model and improve its accuracy. We are talking about Geolifts or Conversion Lifts, for example, which help us to train the model and improve it.
The more complete and accurate the data you enter in the model, the better you will be able to understand the effectiveness of your marketing actions.
Data modeling techniques in MMMs
As we have seen MMMs have a statistical approach used to analyze and quantify the impact of various marketing actions on business results, such as sales, revenues and profits. Several modeling techniques are used in an MMM, the most common of which are:
- Machine Learning
Machine learning algorithms can analyze complex relationships between marketing inputs and sales and identify patterns of behavior, aiding in process automation and decision making.
- Regression
Regression analysis is a technique widely used in MMM models. It consists of analyzing the relationship between a dependent variable, which will be our KPI (e.g. sales), and one or more independent variables (e.g. advertising expenditure, price, promotions, etc.). In this case, Ridge regression is used to minimize the overfitting of the model.
- Time series analysis
A statistical technique used to analyze data points that are collected over time. It can be used to examine the impact of various marketing activities on business performance over time, as well as to predict future sales based on historical data. In the case of MMMs, although we have data at the daily level, it is usually aggregated to minimize noise and to be able to focus on the periodicity of the signals.
- Bayesian analysis
Bayesian analysis is a statistical technique that involves using Bayes' theorem to calculate the probability of an event occurring based on prior knowledge and new information. It can be used in an MMM to estimate the impact of various marketing actions on business results and to make predictions based on previous data.
- Multichannel attribution model
This technique involves assigning sales credit to the different marketing channels that contributed to the sale. It can help identify the most effective channels to drive sales and optimize advertising spend accordingly.
- Econometric modeling
This involves using economic theory and statistical models to analyze the impact of marketing inputs on sales. It can help identify the long-term impact of marketing campaigns and optimize marketing strategies accordingly.
The choice of modeling technique for MMM will depend on the specific needs and objectives of the company, as well as the data available to carry out the analysis.
Interpreting Marketing Mix Modeling results
As we have seen, MMM models provide information on how various marketing actions, such as advertising spending, promotions and pricing, influence sales. Still, depending on the data modeling technique used and the objectives of the analysis, MMM results can be interpreted in a variety of ways. Here are some common ways to interpret the results:
- Coefficient estimates
In linear regression and other statistical techniques, coefficients represent the estimated impact of each marketing input on sales. A positive coefficient indicates that the input has a positive impact on sales, while a negative coefficient indicates a negative impact. The magnitude of the coefficient represents the strength of the impact. The larger the coefficient, the more significant the impact of the input on sales.
For example, suppose that after performing a regression analysis, a company finds that the estimated coefficient for the value of a channel is -0.5. This means that for every €1 increase in ad spend investment in that channel, the KPI is expected to decrease on average by €0.5.
Thus, with a coefficient estimate, informed decisions can be made on how to adjust prices to maximize profit and minimize lost sales.
- Elasticity
Elasticity represents the change in sales of a product or service resulting from a change in a marketing input. For example, a company wants to know if increasing its advertising investment would have an impact on the sale of its products. It discovers an elasticity of its products of 0.5, which means that increasing its advertising investment by 10% would increase its sales by 5%. Therefore, if the company decides to increase its investment from $10,000 to $11,000, it could expect to sell 10,500 products instead of the 10,000 it was selling previously. Elasticity can help identify the most effective marketing actions or channels and optimize advertising spending accordingly.
- Sales contribution
Marketing Mix Modeling can provide information on the contribution of each marketing action or channel to total sales. This can help to identify the most important contributions and prioritize for future strategies.
- Return on Investment (ROI)
Represents the amount of revenue generated for every dollar spent on marketing. MMM models can help calculate the ROI of different marketing actions and identify those channels that are most effective in terms of ROI.
- Scenario analysis
They can also be used to analyze different marketing scenarios and predict the impact of changes in inputs on sales. This can help identify the best marketing strategy to achieve specific objectives.
In general, the interpretation of MMM results will depend on the specific objectives of the analysis and the modeling technique used. It is important to be aware of the limitations and assumptions of the modeling technique and to validate the results with real data before making decisions based on the analysis.
Conclusion
Identifying which marketing channels drive the most sales is a crucial step in optimizing a marketing strategy. Through data collection, the development of a Marketing Mix Modeling model and the subsequent analysis of the results obtained, you will be able to identify the most effective marketing channels. When you work with MMM models, it is ideal to propose them on a recurring basis, so that they can be fed with data and learn from them in order to make them more and more accurate.
Through this analysis, which will allow you to measure any type of marketing action, you will obtain information to optimize your marketing strategy. Allocating more resources to the most effective channels and reducing or eliminating advertising spending on those that are less effective. You will also be able to experiment with different marketing scenarios to identify the best strategy to achieve your objectives.
Marketing Mix Modeling may seem a complex environment to handle, that's why at Adsmurai we have a team specialized in Marketing Science that through our data and benchmarking infrastructure has managed to develop a proprietary model. This completely customized MMM for each client and based on transparency is always accompanied by a Marketing Scientist, an analyst specialized in data modeling.