Part 3: What data you need for MMM
What external data sources you can leverage in Marketing Mix Modeling
What data you need for MMM
Lesson 3.3 - What External data to use in MMM
In this lesson (4:37)
- What external data is relevant in MMM
- How to use this data in MMM
Just because you can doesn't mean you should
While it’s important to include all relevant data sources, we have to keep in mind at the same time how the data is leveraged in MMM.
MMM is based on timeseries analysis, which means that the model is trying to find a signal from the outcome variable data and link this signal back to one of the factors.
The signal has to be strong enough, so there has to be a statistical significance in the observed phenomenon in order for the model to form a link between the two.
What is a bad external data source?
Let’s take weather as an example.
Weather can have a substantial impact on the baseline, but this signal can become elusive if we’re trying to use for example changes in temperature as an input variable.
This is because at the end of the day, moving from 16 Celsius to 18 Celsius isn’t likely going to have a noticeable effect by itself, and so the model won’t see variation between gradual changes and sales.
What is a good external source?
Instead, if we flag certain weather condition, such as hot days (let’s say +25 Celsius), as an on/off variable, the model has a better chance spot the effect of “hot days” by comparing the difference with “non-hot days”, as there could be a clear decline in demand just because people venture outside to enjoy the weather and shop less online.
Similar weather flag could be dry/rainy days, which the model could use in explaining a higher baseline on rainy days as people stay indoors and use their time for shopping online.