The 5 Pillars of a Best-in-Class Marketing Mix Modeling Capability
Without expertise, it’s easy to do Marketing Mix Modeling wrong.
It’s no surprise that multiple experts on econometrics felt the need to speak up on this in the recent WARC open letter warning of the dangers of platform based MMMs and the pitfalls to avoid. Though we understand the validity of their concerns, we believe that there’s more to the story.
In this article we tackle the same question but we’re looking at the other side of the coin and highlighting what successful Marketing Mix Modeling looks like today. If we sum that up in 5 points, it would be the following:
-
- Transparent models you can easily interpret
- Business nuance through the human touch
- High accuracy through granular data
- Dynamic Optimization and Agile Simulation
- Fast updates through automated data pipeline
We explore what these five elements mean. Whether you’re a marketing data scientist seeking a better way to do MMMs or a media planner looking to maximize your budget allocation, you’ll find useful insights about the power of cutting-edge Marketing Mix Modeling techniques.
TIP
A study by Accenture on advanced MMM methodologies found that: “Best-in-class Marketing Mix Modeling, driven by advanced machine learning, utilizes fast moving and granular optimizations to improve marketing ROI by 14% - 38%” This goes to show how much a business stands to gain by adopting an innovative approach to MMM.Transparent Models
You Can Easily Interpret
There’s a class of machine learning techniques known as black boxes. Algorithms like deep neural networks belong to this category. They have been very useful for solving many problems because of their strong predictive ability. Where the problem lies is in interpreting their results. These algorithms can be so complex that you can’t understand how they came to a certain conclusion.
For many data science challenges, interpretability doesn’t matter as long as you know the results that you’re getting are statistically accurate. However, this is not the case for MMM. In fact, the modeler can generate multiple models that are statistically robust. Then, they need to ensure to use the one that makes sense from both a statistical and business perspective.
For MMMs, the go-to technique for building transparent models is regression. It offers a clear and interpretable structure, making it easier to see how inputs are transformed into robust predictors .
→ Without being able to interpret how the model came to a conclusion, it’s very difficult to know which one represents the business situation most accurately.
A central theme in Marketing Mix Modeling is that it’s both an art and a science. Getting the statistical accuracy right is only one part of the picture. It’s the business story that matters most.
Business Nuance Through
The Human Touch
Having a skilled modeler adjust a Marketing Mix Model is crucial for achieving accurate and reliable results. Their expertise in data processing, model selection, coefficient tuning, and domain knowledge helps create models that reflect the real-world complexities of marketing and provide actionable insights for decision-making.
Business Context: Modelers understand the business context in which the model will be used. They can incorporate domain knowledge and insights from stakeholders into the modeling process. This makes the model more aligned with the business objectives.
Complex Data: Marketing data can be incredibly complex, with various interrelated factors influencing business outcomes. Skilled modelers can identify these complex relationships and ensure the model properly accounts for them.
Data Processing: Modelers can transform variables to capture non-linear relationships, interactions, and seasonality effectively. This process is critical for creating a model that accurately reflects the real-world dynamics of the marketing mix.
Prevent Overfitting: Modelers can apply techniques to prevent overfitting, where a model fits the noise in the data rather than the underlying patterns. This is important to address because overfit models lead to inaccurate predictions.
Model Selection: Skilled modelers can choose or design models that strike the right balance between statistical accuracy and the ability to explain why certain predictions were made. This is especially important because decisions based on the model need to be understood and justified.
Continuous Monitoring and Updating: Markets and consumer behavior evolve over time. Skilled modelers can set up processes for continuous monitoring of the model’s performance and make adjustments as needed to ensure that it remains accurate and relevant.
High Accuracy Through
Granular Data
A Marketing Mix Model’s accuracy is significantly influenced by how much historical data is available, 3 years being the general recommendation. But what happens when you don’t have enough data? One way to solve this is using more granular level.
For instance, instead of examining the relationship between all video impressions and nationwide sales over a given period, modelers can dissect the data into regional, customer segment, or product line categories, subsequently merging these segmented datasets. This process effectively multiplies the total number of data points available for analysis.
To simplify, if you initially have one year’s worth of data collected on a weekly basis, you’d have 52 data points per variable. Yet, by segmenting the same data into three distinct regions, you now have access to 156 data points – a substantial enhancement for your MMM robustness!
In a study comparing the differences, Nielsen found that regional-level models were considerably more robust than national level models. The results showed that models using regional data generated an average R2 of 96%, with an average MAPE of 4%. Comparatively the models that used national data produced an average R2 of 86%, with an average MAPE of 7%.
Dynamic Optimization and
Agile Simulation
For an international company with multiple brands, selling a variety of products across multiple regions, you would need a sophisticated Marketing Mix Modeling capability that can optimize the budgets across the entire portfolio.
In an ideal world, you would model every product at the most granular level ( e.g. SKU), identify diminishing returns curves and saturation levels for each media channel, and then add them all up to get the optimal plan for the portfolio.
Optimizing expenditure for a portfolio raises some tricky questions like:
- Search for Product X is saturated, but it may not be for Product Y. What’s the best overall spend for Search that would optimize for both?
- Most of the budget is spent on a small number of brands. As they start to hit saturation, how do you know which is the next best brand to spend on?
- How do you account for ‘halo effects’ where spending on one brand positively impacts the others?
Dynamic Reporting A Must
Marketing leaders are increasingly seeking real-time insights and actionable data to inform their decisionmaking processes. Traditional marketing mix modeling reports delivered as static PowerPoint presentations no longer cut it. Instead, there is a growing preference for regularly updated results accessible through dynamic dashboards that empower them to dive deep into the data, inspect, analyze, and make informed decisions on the fly.
Marketers can easily share these dynamic dashboards with their teams. With the ability to drill down to the finest details, they can quickly identify trends, correlations, and opportunities relevant to their specific use case. This would usually be overlooked in traditional reports that would’ve been stripped off the technical details to not waste executives’ time.
Fast updates through Automated Data Pipeline
Marketers are calling for faster MMM results that can match the pace of their decision making. Luckily, there have been solutions that allow a faster pace of updates. This is a good thing as oftentimes the main bottleneck to MMM is the data. With this approach modelers can allocate more of their time and expertise to the actual analysis and interpretation of results.
However, it’s important to keep in mind the limitations in an automated setup:
- Some offline channels will still require manual prep. In such cases, there are data preparation solutions that can save time.
- Your rate of updates should keep up with all channels. You will have an inaccurate model if some channels have new data while others are stuck with deprecated data. This is why this approach works best for digital-only marketing strategies.
Fast updates through Automated Data Pipeline
Automated data flows are not limited to providing raw data. They can also include processes for data validation, transformation, and cleaning procedures. Building automated data pipelines ready to be used on new data feeds is crucial as this maintains data quality and ensures that the modeling process is built on clean and accurate data.
Automation can also encompass robust data security measures, such as encryption and access controls, to safeguard sensitive marketing and sales data from unauthorized access or breaches.
Conclusion
Best-in-class MMM, as we’ve explored, entails transparent and interpretable models, the invaluable human touch of skilled modelers, accuracy through granular data, optimization across multiple levels, and agility enabled by automated data pipelines.
