Marketing Mix Models (MMM)

Machine learning can help with Business Intelligence by modeling the dynamics of the your specific business.
Marketing involves a lot of decisions with a large number of possible options, should I invest in branded search? On what TV channels should I run my ads? In order to learn from all of the data that is generated from various marketing actions, we use a Bayesian Marketing Mix Model (MMM). It learns to attribute the effects of the various marketing activities on a specific KPI such as sales. This way the model can retrospectively explain what caused that very KPI to rise or fall in certain periods.

Armed with this knowledge the model can be used to run what-if scenarios to see what would have happened if one or several activities were different. As an example; Our sales took a turn for the worse last month. What if we had increased our presence on Meta during that period?

When optimizing budget allocation across any number of media channels and platforms the optimization tools of these platforms cannot be used as they all operate in silos. In order to optimize across the board you need an objective tool that does not have a stake in which media you choose. It is also not a task humans are equipped to do as even small advertisers typically deal with thousands of choices during a campaign.

How a Marketing Model can Help you
Identify Sales Drivers
Understand how media drives sales and how to increase Sales.

By understanding what drives sales, it is then possible to compare various scenarios against each other in order to obtain the highest ROI on media investments.
Optimize Brand Equity
Know what to do to increase your overall brand equity.

Branding is challenging, but with our models, you get actionable insights on how to best grow your brand. Together, connect the dots and link your brand to CX and Sales.
Minimize Churn
Learn how to use media and other levers to most effectively minimize churn.

Understand how to best balance your media campaigns between minimizing churn and acquiring new customers. This is sustainable growth, delivered.
How we do it
Historically, Multiple Linear Regression models have been used to build sales models such as MMM and MTA. We instead rely on a Bayesian formalism which allows us to create robust and consistent models and predictions. Why do we like the Bayesian framework? Because the alternative comes with a lot of problems.

First off, Linear Regression is based on a mathematical principle called Maximum likelihood estimation which stems from a branch of statistics know as Frequentist statistics. The challenge with this approach is that the more variables you add to your model the less you can trust the estimates that come out. In addition to this many variables in your data also tend to be sparse as you typically don't run TV commercials every day on every channel, so media spending and exposures are zero most days.

Second, this is a recipe for a phenomenon called overfitting. It refers to a situation where a model is flexible enough to "remember" each data point present in the training data instead of learning the underlying dynamics, and therefore loses the ability to generalize to data not previously seen. The consequence of this is that an overfitted model will perform badly when predicting the effect of future actions.

depicting under fitting, over fitting and the correct fit of a time series model
Fig. 1: Showcasing how different models fit data.
Model Types we Support
There is a wide range of areas where our models add value, here is an example of some of the models we support.
Marketing Mix Models (MMM)
Measure the effectiveness of your media investments on various KPIs such as new customers, sales, churn, etc. This makes it possible to compare and plan your media campaigns to see how to best allocate your budget across all available media channels. This can then be linked to revenue and profit such that the impact of the media investment is revealed and quantified. The strategic importance of having a MMM available when planning the next months campaign or when negotiating the marketing budget for the coming year cannot be overstated.
Multi-Touch Attribution (MTA)
Multi-touch attribution models focus on how various online touch points interact during a typical customer journey. First, the awareness of the product needs to be built, then the customer needs to consider buying the product before several actions are taken, ultimately, leading to conversion. By allowing complex dynamics between various advertising channels such as display banners, paid search, social media, etc. over time it is possible to see how they all interact with each other to drive sales.

Read More

Customer Cases

How to proceed

Commercial Navigator
See how our commercial navigator can give you the insights you need
Learn More
Alvíss AI
Use our Alvíss AI platfor to build it yourself.
Learn more
Custom Project
Utilize our consulting services to outsource all or some of the work to us.
Learn more