Can you truly automate Marketing Mix Modeling?

Automation and AI are revolutionizing industries at a breakneck pace, and Marketing Mix Modeling (MMM) is no exception. Automated modeling systems, making use of AI techniques, are becoming table stakes in the marketing world. However, while AI excels in specific parts of the MMM process, it’s not a panacea. Trust remains the critical factor—and trust in MMM models hinges on producing sensible, actionable results that align with real-world business dynamics. 

At ScanmarQED, we’ve been leveraging AI to enhance MMM efficiency for nearly two decades. Our strataQED software, for instance, employs a machine learning algorithm to power its automated ‘Model Search’ function. This genetic algorithm tests tens of thousands of potential model permutations in minutes, dramatically accelerating a traditionally labor-intensive process. 

AI has proven invaluable in automating well-defined and repeatable tasks, but we remain pragmatic about its limitations. As a company we embrace AI for the parts where it can be helpful and then use it in combination with the human touch where needed. In this blog, we explore the areas we feel are important to consider: 

Data  

• Data isn’t perfect: It’s not always clear what each number in a dataset represents or how accurately it was measured. Analysts must dive into the specifics to ensure the data aligns with its intended meaning and context. 

• Addressing missing or faulty data: When or if a data capture goes awry, human intervention is essential to transform or create variables that accurately reflect the effect being studied. Even AI can’t fill gaps it doesn’t recognize, making human judgment critical for identifying and addressing such missing elements. 

• Collaborating for context: Understanding what truly happened often requires direct communication with the teams involved. Analysts must untangle the nuances of data collection and adjust datasets to reflect the reality, ensuring that insights are both accurate and actionable. 

The Nuances of Model-building 

Model-building lies at the heart of MMM, but it is hard to automate. Businesses are complex and no two are alike. AI algorithms can produce statistically sound models based on available data, but what about the complexities of business and industry understanding critical to the model building process? 

Some subtleties: 

MMM is based on inferring relationships probabilistically, so hypotheses need to be constructed and tested. Firstly, this means including all the right variables in the model (price, economy, seasonality, Covid, etc.). For every important variable that is not included in the model, the parameter estimates for the focus variables (typically marketing activities) will be wrong. 

Building a robust model is an iterative journey. Analysts construct models, test variable combinations, and refine or create new variables to better explain sales patterns. This ongoing process helps uncover the true drivers of performance. 

Understanding complex business dynamics often requires detective work. Analysts brainstorm potential explanatory factors, engage with stakeholders across departments, and create or refine variables to reflect the reality of business operations. 

Sense-checking results is crucial to ensure the model reflects the business reality and can credibly answer the business questions. This can include comparing results to norms, previous results, output from other measurement approaches (e.g. incrementality tests), but also the experience of those working in the business day-to-day who are acutely aware of the impact of different levers on performance. 

Theoretically, it is possible to encode assumptions into AI-driven models, but in practice, it is nearly impossible to account for every potential scenario. The range of possibilities is vast, and attempting to pre-empt all of them would be both impractical and highly complex.  

Metadata—encompassing categorizations, hierarchies, expected relationships, and other structural elements—plays a crucial role in providing AI with a meaningful “business context.” Without this context, AI models risk generating insights that may be technically accurate but lack real-world applicability. At ScanmarQED, we are committed to ensuring that our PulseQED platform is built on a smart and flexible data foundation. By structuring data in ways that facilitate AI-driven analysis, we enable both current and future modeling techniques to deliver more precise, actionable insights. 

The Trust Factor 

Many competitors claim AI capabilities but lack transparency, leaving users in the dark about how their models are built and whether they can be trusted.  

A common experience from clients we have spoken with that are trying these automated systems is that they are too simple, the numbers make no sense, no-one believes them, and ultimately the system is abandoned. 

Inevitably automation will improve as more and more of the complex decision-making processes that good modelers follow get thoroughly mapped out and replicated with lines of code, but in our view a better automated MMM workflow incorporates a suite of human-built models, automated data pipelines and automated updates – combining the best of what automation can achieve with the irreplaceable role of human expertise in data analysis. 

At ScanmarQED, we prioritize transparency. Our PulseQED platform combines AI-driven automation with essential human oversight to deliver fast, reliable, and validated models. Users can: 

• Gain confidence in AI-enabled results with full visibility into the modeling process 

• Build models that align with real-world scenarios, delivering clear and actionable insights 

• Run automated model updates on human validated models as often as needed 

• Streamline strategic planning while optimizing marketing performance

A final thought 

AI has undeniably become a powerful ally in MMM and we are now increasingly using it for automating as much of the data collection, data processing and reporting work as possible, but it’s not yet a substitute for human expertise. Trustworthy models are not just statistically sound; they are grounded in reality and tailored to the business’s unique needs. By combining AI’s efficiency with the analyst’s critical judgment, we ensure that MMM delivers not only faster results but also smarter business decisions.