The upgraded Plan Performance tabs of your dashboards are powered by a machine learning model, an algorithm that examines large amounts of historical data to find relationships between variables and a target. It can then use those relationships to make predictions about future performance.
Because it's automated, the model can consume extremely large amounts of data very quickly and uncover patterns faster than an entire team of human analysts could.
The model does this by "training" itself - constantly updating and optimizing its understanding of the relationship between variables with each new piece of data it encounters.
Marketing & Media Mix Model and Machine Learning
In the case of ChannelMix's Marketing & Media Mix Model, we use a supervised learning model.
We provide the model with specific inputs (how much was spent on each marketing channel) and a specific output (leads, sales, or revenue generated) and ask the model to find the rules or patterns that link those two sides.
For example, the model might reveal that...
- Spending $10,000 on Channel A leads to $1000,000 in new revenue
- Spending $5,000 on Channel B has a positive effect on revenue for the next 6 days
- Spending on Channel C starts becoming less effective after $20,000 has already been spent
And that's just a tiny, tiny sliver of the patterns that might be uncovered by the model. The model can make predictions for multiple combinations of channels at multiple levels of spending. Or put another way, you can use the model to make predictions.
Machine learning models typically require large amounts of historical data for the model to train itself correctly. For instance, to learn a year-over-year trend we will need at least 2 years' worth of data. While we can model as little as 6 months' worth of data, it's important to keep in mind that the more historical data you have available, the more factors we can model more accurately.
When we are discussing data requirements, we are talking about how many days' worth of valid data we require. In this instance, data being valid means that, for every date in the dataset, there are at least 2 distinct marketing channels with activity and there is activity in the target variable.
ChannelMix will regularly incorporate a client's most recent media data into the model. Generally speaking, updates occur monthly.
As new data is added, the model will update and refine its predictions based on these new results.