From Marketing Mix Modelling to Econometrics: Top 10 FAQs answered

Or, everything you wanted to know about MMM and econometrics, but didn’t dare to ask.  

MMM and econometrics are powerful strategic tools, but many of the old views of MMM persist. It's just for sales modelling. It's only for media. But those views mean MMM is often misunderstood, misused and under-valued.

In this episode, we put the top 10 most searched for questions about MMM to our UK GM & Partner, Matt Andrew, to help bring to life the breadth of MMM’s capabilities and bust a few myths.  

SPOLIER ALERT: it’s not just for media… Or sales.

Dive into the episode on YouTube, Spotify, Apple or search for Ekimetrics on your favourite podcast platform.

Find out:  

1. How does Marketing Mix Modelling work?  

2. What are the benefits of Marketing Mix Modelling?  

3. What data is needed for Marketing Mix Modelling?  

4. How do you implement Marketing Mix Modelling?  

5. What are the limitations of Marketing Mix Modelling?  

6. How do you interpret the results of Marketing Mix Modelling? What are some of the most useful outputs of Marketing Mix Modelling?  

7. How does Marketing Mix Modelling differ from other marketing analytics?  

8. Can Marketing Mix Modelling predict future sales? Or other KPIs?  

9. Should all brands invest in MMM? How do you know if it’s right for you?  

10. If you’re starting out with MMM for the first time, what advice would you give?

‍

Whether you're just starting out or a complex, global, multi-category business, you can expect to hear straightforward answers to these questions with further discussion on how to use MMM as a strategic value driver.  

The conversation ranges across a broad set of applications, from forecasting and scenario planning to pricing and driving your sustainability agenda. And an equally broad set of integrations. From how to use MMM in conjunction with other techniques, such as multi-touch attribution to experiments to develop a unified marketing measurement capability to how to nest real-time data using Bayesian Priors. Plus we consider how to embed customer lifetime value and capture creative effectiveness. And much more...    

And for those who listened in, here’s the Bayesian modelling explainer:  

‍Real Time Data and Bayesian Hierarchical Modelling  

Classical MMM models use linear regression, where the more data is available, the more robust the model and its conclusions. Typically they require between 3 and 5 years of historical data, representing between 36 to 160 monthly observations, or 150 to 250 weekly observations to produce accurate results.  

Which is why using shorter outcome periods – or real time data – can be a challenge; small sample sizes can lead to reduced model accuracy.  

To counter this, using Bayesian statistics in MMMs accommodates time-varying coefficients to reflect dynamic market conditions. That’s because Bayesian models use probabilities to deal with statistical problems.

The main difference compared to classical, frequentist, methods is that Bayesian estimation allows you to incorporate previous knowledge or beliefs into the model. The prior beliefs are then combined with the existing data to improve decision making as more data becomes available. This approach allows for more nuanced understanding and prediction of sales, accounting for the evolving influence of marketing channels over time.

However, there are some limitations. Bayesian estimation, while flexible, is not inherently designed to enhance the precision of model estimates. It introduces adaptability by allowing model coefficients to change over time, which can decrease potential bias within the model. That means Bayesian estimation is most effective when applied to panel data that vary across segments, such as regions or stores, since it relies on fewer assumptions compared to time series data.  

If you would like to discuss modelling techniques or any of the questions raised and answered in the podcast, we’d love to hear from you.