Introduction for B2C Startups
As a Growth Marketing professional, this guide aims to illustrate what Marketing Mix Modeling entails, why it holds significance for startups, and it how works.
Our previous discussion delved into the AdStock model, an econometric technique valuable for assessing the enduring impact of advertising spending over the long term. For a hands-on blog post on measuring the AdStock model, you can refer to my earlier blog post.
Now, let’s delve into the world of Marketing Mix Modeling, where we’ll examine core concepts and real-world applications specific to B2C startups. This guide is designed to offer actionable insights, empowering you to grasp how this robust tool can significantly influence your marketing strategies.
Whether you’re familiar with the AdStock model or not, this blog post will guide you through the practical aspects of Marketing Mix Modeling, helping you navigate its applications and understand its potential for enhancing the marketing efforts of B2C startups.
Let’s dive into the world of Marketing Mix Modeling 😊
What is Marketing Mix Modeling?
Market Mix Modeling (MMM) is a statistical technique employed by both established companies and startups to quantify the impact of various marketing inputs on sales or market share. The primary purpose of utilising Marketing Mix Modeling is to discern the contribution of each marketing input to a specific output.
In simpler terms, MMM serves as a decision-making tool, aiming to understand how various internal and external factors (such as price, competition, inflation, seasonality, etc.) may influence the marketing Return on Investment (ROI) across different advertising channels like PPC, TV, radio, and flyers.
Within the landscape of multi-channel marketing campaigns, where each channel serves a distinct purpose, MMM becomes particularly valuable for measuring performance and optimising budgets across diverse channels. This enables businesses to make informed decisions about resource allocation, ensuring that marketing efforts align with strategic objectives.
By leveraging MMM, companies can identify high-performing channels, and make data-driven adjustments for enhanced overall performance in the competitive market environment.
What makes Marketing Mix Modeling so important?
‘Half the money I spend on advertising is wasted; the trouble is I don’t know which half.’
J. Wanamaker (1838-1922)
While Marketing Mix Modeling has its roots in the 1960s, its technique has recently experienced renewed interest in the digital landscape, driven by the challenges posed by Apple’s privacy policies (iOS) and Google’s deprecation of third-party cookies tracking.
In today’s omnichannel world, customers engage with multiple touchpoints and various channels before purchasing from a brand.
A multi-touch attribution model (e.g., Google Data-Driven) theoretically gives marketers the power to understand and evaluate the value of different touchpoints in the customer journey before conversion. However, in reality, it is simply not possible to accurately track every interaction that may have influenced a conversion. For example, Google Analytics cannot track 100% of user sessions, mainly because users deny consent to cookies and use ad blockers. If you can track 70%, that is considered a good result.
One of the biggest challenges for marketers with classic attribution models is ensuring accuracy. Incomplete tracking or data inaccuracies can skew results. For instance, if a customer interacts with a touchpoint that is not tracked, the attribution model may not give credit to that marketing channel, leading to an incomplete view of the customer journey.
Another common issue is the lack of a holistic view in tracking standardisation across channels. Each channel may have its own tracking methods. For example, Meta Ads and Google Ads track conversions differently, making it challenging to compare and combine data in the same reports, and thus harder to draw meaningful insights for the same buyer persona.
How to Overcome These Key Challenges in Media Optimisation with MMM?
As we swiftly transition to a privacy-first online world, businesses are increasingly turning to new econometric tools like MMM as a privacy-friendly preferred model for assessing marketing efficiency.
Marketing Mix Modeling is a privacy-friendly tool due to its focus on aggregated and anonymized aggregate and historical data to arrive at macro-level conclusions. Instead of delving into individual customer details, it analyzes overall patterns and trends, ensuring the protection of individual privacy.
On the other hand, multitouch attribution modeling assigns a specific value to each individual user and sales touchpoint; therefore, user data are not anonymized. Also, econometrics models, such as diminishing returns, can look into the future and not just in the rearview mirror. Marketing attribution models usually assess performance after a campaign concludes, often taking a few months. However, they struggle to effectively answer the question: where should your next best dollar go?
Key Components of MMM Model.
In this paragraph, I’ll analyze the primary objective of MMM, explore Marketing Mix Modeling components, and delve into the main steps for building a B2C startup model. The chart illustrates a comprehensive process. Starting with meticulous data preparation, incorporating advanced modeling techniques, and culminating in actionable insights to enhance media spending and overall effectiveness.
Marketing Mix Modeling Goals.
We will begin by reviewing three potential goals and the associated business questions addressed through MMM:
| Goals | Business Questions |
| 1. Measure efficiency | How efficient was our spending on PPC last year? |
| 2. Simulate | How would our sales change if we increase/reduce our budget on our channels? |
| 3. Optmise budget | How to optmise media budget to increase ROI and sales? |
The diagram below illustrates the three key steps of the Marketing Mix Modeling workflow:
Figure 2: Takeda, H. (2023). PyData. Media Mix Modeling: How to measure the effectiveness of Advertising.
- Data Preparation: Preparing data for Media Mix Modeling involves defining business objectives, collecting time series data, and identifying key variables such as sales, media spending, and external factors. Technical tasks include splitting datasets for training/testing, custom scaling for uniformity, and addressing missing data. Pre-processing ensures data integrity, setting the stage for effective model training and analysis.
- Modeling: In the Modeling phase of Media Mix Modeling (MMM), key steps involve selecting suitable algorithms, evaluating model performance through techniques like hyperparameter search, and fine-tuning for optimal results. The focus is on linear or multilinear regression with considerations for saturation and ad stock principles, ensuring an effective representation of marketing impact on sales.
- Analysis: Insights gained in the previous steps guide adjustments to media spending strategies. The model’s ongoing maintenance involves assessing R-squared and MAPE metrics for accuracy. Optimization decisions based on Bayesian or frequentist methods and hierarchical approaches contribute to continual refinement, enhancing overall marketing effectiveness.
How the MMM Model Works?
Building a robust and objective model means using accurate input data to feed the model. Let’s break down our data variables in two categories:
Figure 3, Takeda, H. (2023). PyData. Media Mix Modeling: How to measure the effectiveness of Advertising.
Sales and media spending are essential input data for predicting sales based on past media expenditures. The charts above display additional input data that proves valuable in forecasting potential revenue based on these input values
Now, let’s analyse the main components and statistical concepts behind MMM modeling, the third step of the workflow.
Regression Analysis.
Regression analysis is an important part of model building. It is a powerful methodology used to measure the relationship between a set of variables and a specified KPI and predict future outcomes. This makes it very useful and powerful. We can use linear regression or multilinear regression if using more than one output (independent variable).
Saturation (Shape effect).
Saturation is a phenomenon wherein the incremental escalation of advertising expenditure corresponds to diminishing effectiveness of the advertisement. This trend suggests that as marketing investments increase, the impact on target audiences reaches a point of diminishing returns.
Figure 4: Menzies, T., 2002, Saturation effects in the testing of formal models
AdStock (Carry-over effect).
AdStock refers to the lingering effect or carryover impact of media exposure on sales, manifesting with a temporal delay after the initial advertisement exposure. This temporal persistence is characterized by a decay function, reflecting how the influence of media diminishes over time yet still maintains a residual impact on consumer behavior. Explore my previous blog post to learn more about AdStock.
Seasonality.
Seasonality, in the MMM model, involves analysing cyclic patterns in consumer behavior tied to specific times, like holidays or seasons. This technical aspect entails incorporating time-dependent variables, adjusting for regular fluctuations, and employing sophisticated algorithms to accurately capture and model seasonal effects, enhancing the precision of marketing strategies. For instance, a retail MMM may reveal increased sales during the holiday season or fewer in-store visitors during the pandemic.
Time Series.
To build the model we need weekly aggregated data shown in time series (Column A). Column B shows aggregated weekly sales (output). This is what we are trying to predict. Columns C, D, and E represent media spending by channels (input). You can use pandas in Python to perform the same analysis.
Figure 5: Taylor, M. Vexpower, 2023
Assessing Media Mix Modeling (MMM) results is a key step for ensuring the accuracy and effectiveness of the model in predicting sales. Using (multi)linear regression metrics like R square, Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) provides valuable insights into the model’s performance.
- R square, or the Coefficient of Determination: This is a measure of how well the model explains the variability in the dependent variable. A higher R square indicates a better fit, showcasing the proportion of the variance in sales that can be explained by the independent variables in the model.
- Mean Absolute Error (MAE): it gauges the average absolute difference between the predicted and actual values, providing a straightforward measure of prediction accuracy. It’s particularly useful for understanding the average magnitude of errors in the model.
- Mean Absolute Percentage Error (MAPE): it is a percentage-based metric that calculates the average absolute percentage difference between predicted and actual values. It is helpful fo understanding the relative accuracy of the model across different levels of sales.
As for using Google Sheets for these calculations, you can indeed leverage its functionalities. The =LINEST function in Google Sheets performs a linear regression analysis, returning an array that describes the line that best fits your data. This can be applied to evaluate R square, providing a quick overview of the model’s explanatory power.
Conclusion
In the contemporary marketing landscape, marked by the evolving privacy landscape and challenges posed by the deprecation of third-party cookies, Marketing Mix Modeling (MMM) has resurged as a practical and privacy-friendly econometric model. Its renewed relevance is further fueled by the mounting concerns surrounding mobile iOS 14 attribution. By embracing MMM, marketers not only adapt to these shifting paradigms but also gain a comprehensive perspective, enabling them to fine-tune budgets and strategically enhance ROI. In the face of an ever-changing marketing landscape, MMM stands as an indispensable addition to the marketer’s toolkit, providing actionable insights for informed decision-making and sustained success.
Did you find this blog post useful? Do you have any questions? I am happy to schedule a virtual coffee with you 🙂 ☕
Bibliography
Coleman, Alex (2016, Sep. 30). The Path to Purchase in the Context of Time [Web Log Post]. From https://www.wolfgangdigital.com
Danaher, P. (2017). Advertising effectiveness and media exposure. In Wierenga, B. & van der Lans, R. (2017) Handbook of marketing decision models. International Series in Operations Research & Management Science 254. NY: Springer.
Franco, G. (2022, February 18). How to Marketing Mix Modeling (MMM) with Excel — Intro. Cassandra.
Taylor, M. (2023). Marketing Mix Modeling in 7-18 Weeks. VexPower.
Walentosky, M. (2022, September 27). Marketing Mix Modeling (MMM) Explained. Bounteous.
Leave a Reply