Is Marketing Mix Modelling (MMM) for you?

Prelude: Why everyone is talking about it

Hello there, Dr. Wei here! Today, we embark on an exciting journey into the world of Marketing Mix Modelling (MMM). If you’re not in marketing or data science, you might wonder, "What is MMM?" Some of my data science friends say it's gaining popularity because it's resilient against the impending 'cookie apocalypse'. If you're unfamiliar with the term, it refers to the transition to a cookieless world where tracking online behaviour via cookies is becoming obsolete. You know those ads that follow you around after you search for something? That's the magic of cookies at work.

But what happens when cookies crumble? Enter MMM.

The Roots of MMM: A Statistical Foundation

Marketing Mix Modelling has deep roots in statistical analysis, particularly in multivariate regression. Imagine you’re trying to understand how various factors, like TV ads, online banners, and radio spots, impact your sales. Multivariate regression is like a sophisticated detective that helps you unravel these relationships by analysing multiple variables simultaneously. This statistical method allows us to see how different marketing activities contribute to sales, providing a comprehensive view of our marketing efforts.

Understanding Multivariate Regression

In the context of MMM, the dependent variable is typically sales or another key performance indicator (KPI) such as conversion rate, while the independent variables include various marketing activities and external factors.

• Identifying Relationships: By using multivariate regression, we can identify which marketing activities and external factors have the most significant impact on sales. For example, do TV ads generate more sales compared to online banners? How does radio advertising contribute to overall revenue? Additionally, we can understand how external factors like economic conditions or weather influence sales.
• Quantifying Impact: This method not only identifies relationships but also quantifies the impact of each variable. We can determine how much of the sales increase can be attributed to a specific marketing activity or external factor. For instance, if we increase spending on TV ads by a certain amount, multivariate regression helps us estimate the expected increase in sales. Similarly, we can quantify the impact of a public holiday or a change in GDP on sales.
• Controlling for Other Factors: One of the key strengths of multivariate regression is its ability to control for other factors. In a real-world scenario, many variables can affect sales simultaneously. Multivariate regression isolates the effect of each marketing activity and external factor by holding other variables constant. This helps in understanding the true impact of each marketing effort and external influence.
• Dealing with Interactions: Marketing activities often interact with each other and with external factors. For instance, a TV campaign might boost the effectiveness of online banners, and an economic downturn might reduce the impact of promotional activities. Multivariate regression can include interaction terms to capture these synergies and interdependencies, providing a more accurate picture of the overall marketing effectiveness and the broader market context.

Here's a snapshot of the kind of data we work with in MMM:

The Advantages of MMM

One of the standout advantages of MMM is its resilience to the cookie apocalypse. Unlike other methods that rely on tracking individual customer data, MMM uses aggregated data. This means it remains unaffected by the shift to a cookieless world. You won’t need to worry about privacy issues or tracking regulations because MMM doesn't depend on individual cookies at all.

Consider multi-touchpoint attribution analysis, for instance. This method tracks every customer interaction across various touchpoints to attribute conversions accurately. It relies heavily on cookies and individual tracking data to map out the customer's journey, making it highly susceptible to changes in privacy regulations and the cookieless transition.

In contrast, MMM thrives on the granularity of aggregated data, offering valuable insights without the need for customer-level details. By analysing broader trends and patterns across different marketing channels, MMM provides a comprehensive view of overall marketing effectiveness. This makes MMM a robust tool for understanding how various marketing activities contribute to sales, helping you make informed decisions and optimise your marketing strategy efficiently.

The Shortcomings of MMM

However, like any tool, MMM has its shortcomings. While it doesn’t require individual customer data, it does need extensive historical data from various channels and external sources. Typically, for weekly data, a thumb rule is to have at least 2-3 years of data, while for monthly data, a longer period, such as 4-5 years, is preferable to ensure accuracy and reliability.

Gathering and maintaining this extensive historical data can be challenging and time-consuming. Moreover, MMM can be complex to set up and interpret. It requires a solid understanding of statistical modelling, which can be daunting for those new to the field.

One of the trickiest issues we encounter in real-world MMM applications is multicollinearity. This occurs when independent variables in a regression model are highly correlated, making it difficult to determine their individual effects on the dependent variable. Imagine trying to measure the impact of TV ads and online banners on sales, but these two activities often occur simultaneously. It becomes challenging to untangle their individual contributions, leading to less reliable results. Multicollinearity is somewhat inevitable in the real world where marketing activities overlap and interact in complex ways. In practice, when building models, we typically want to drop or combine highly correlated variables. This can be done through techniques like Variance Inflation Factor (VIF) analysis to identify and remove problematic predictors or by aggregating similar variables to reduce redundancy and improve the model's robustness. Sometimes, Principal Component Analysis (PCA) is also used to reduce dimensions, transforming correlated variables into a smaller set of uncorrelated components, which helps in simplifying the model without losing significant information.

Why MMM is So Popular

Despite its traditional roots and complexity, MMM is enjoying a resurgence in popularity. The impending cookieless future has marketers scrambling for alternatives, and MMM stands out as a reliable method. Its ability to provide aggregated insights without infringing on privacy makes it particularly attractive. By utilising MMM, businesses can continue to optimise their marketing strategies without relying on invasive tracking methods.

MMM reveals the impact of various marketing activities on sales and other key performance indicators, allowing you to:

• Make Data-Driven Decisions: Understand the impact of different marketing activities to enhance ROI.
• Optimise Budget Allocation: Identify the most efficient channels and allocate your budget effectively.
• Gain Predictive Insights: Anticipate future performance and adjust your strategies proactively.
• Achieve Competitive Advantage: Stay ahead by optimising your marketing efforts and responding swiftly to market changes.

A Niche Gap to Fill

Now, let's talk about the talent gap. Traditional marketing analytics training for bachelor and master levels has often focused on classic software such as SPSS (Haha, I used to teach this for my master-level students). While SPSS is a powerful tool, it doesn’t provide an automated pipeline to generate data-driven insights seamlessly. Marketing professionals need to acquire data science skills and statistical knowledge to enhance their model-building capabilities. On the other hand, data scientists might need to bolster their understanding of marketing principles to apply their technical expertise effectively.

Would you consider this a niche for you to fill? If you have a background in either field, bridging this gap could be a fantastic opportunity. By combining expertise in data science with deep marketing insights, you can drive powerful, data-driven marketing strategies. This blend of skills is not only valuable but increasingly essential in today’s data-driven marketing landscape.

The Future of MMM in an AI-Driven World

With the rapid advancements in AI and machine learning, one might wonder if MMM will continue to thrive. The answer is a resounding yes. AI can enhance MMM by automating complex data analysis and providing deeper insights through advanced algorithms.

Real-time data is becoming increasingly crucial in the realm of MMM. For instance, Facebook’s Robyn, an open-source marketing mix modelling platform, continually evolves to integrate real-time data capabilities. The Robyn API allows marketers to automate the ingestion and processing of fresh data, ensuring that models are always up-to-date with the latest information. This ability to handle real-time data means you can adjust your marketing strategies on the fly, responding to market changes as they happen. Real-time insights enable more agile and dynamic marketing approaches, making your campaigns more responsive and effective.

Join Me on This Journey

Are you excited about the potential of MMM and the blend of marketing and data science? Follow me for more insights as I explore this fascinating intersection. Over the next few weeks, I will dive deeper into the MMM theme, uncovering more knowledge and practical applications. In my next newsletter, I'll introduce Chapter 1: how an MMM dataset looks and how you can simulate your own dataset for an MMM portfolio project. I'm eager to take you on this journey, showcasing real case studies and practical examples. Stay tuned for an insightful exploration of Marketing Mix Modelling!

#MarketingMixModelling #MMM #DataScience #Analytics #MarketingStrategy #DataDriven #AI #DigitalMarketing #BusinessInsights #Datadrivenmarketing #Innovativemarketers