MIM Control Datasets | All about ChannelMix Control Datasets
What is a control dataset?
If you believe an external factor is impacting your data, then you should consider using a control dataset. MIM Global Control Datasets provide an easy way to account for variability in your data.
Control datasets are a vital part of any good MIM. The more context the model has about a specific marketing environment, the better it can understand the impact of marketing spend leading to higher quality recommendations on how to spend in the future.
There are constraints on what kind of data can be utilized as a control dataset. Check out the article AI Explained | Requirements for Control Datasets for more information. Continue reading to learn more about control datasets!
Example Case
A State Park runs promotions in the off-season. They want to use that promotional calendar to help the model learn more intelligently about any potential impact their promotional calendars might have had in the off-season. If there is a spike in visitors in the winter, the model is going to just interpret that as normal seasonal behavior. But when we tell the model we are running promotions, we are telling it "Hey! I had this promotion going on that might have influenced that spike." Then the model can look at trends that I can't see with my naked eye to see if maybe it did have some "marketing impact".
Why use a control dataset?
Control datasets are used to provide more context to your historical performance. By providing a more holistic view of your performance to the model, it is better able to tease apart the effect of your marketing spend versus the effect of other factors in your business ecosystem. This allows the model to better understand the diminishing returns of different marketing channels and the return from the channels as well as provide better recommendations on how to distribute your budget.
How do I set up a control dataset?
These datasets use a standard ChannelMix Google Sheet template. Learn more about our Google Sheet templates in ChannelMix Google Sheet Templates. If Google Sheets is not an option for your organization, you can work with your CSM for alternative solutions.
Is my data before my control dataset used in the model?
By default, if you choose to include a control dataset in your model, any data before the earliest start date in your control data will not be included in the model. Here's why:
We are using this data for a reason - to make our data smarter. Therefore, anything before this dataset where we can't give the model this information will still be trying to find trends without the context of your control dataset, making it less smart than the portion of data with the control data, thus weakening your model.
What happens to my model if I stop filling out my Google Sheet with control data?
If you stop filling out a promotional calendar, the default going forward is that you are not running any promotions.
if you stop filling out your brand awareness sheet, the default going forward will be the median of your existing data variable.
How do promotions play a role in predictions?
By default, the model will give you predictions as if you are not running any promotions.
If you include future dates in your control data, the model will include these when producing recommendations.
How can I see my control data in my dashboard?
The model summary tab of your ChannelMix dashboards with MIM capability provides an estimate of the "return" on each control variable. In the context of ChannelMix's MIM, return is defined as the predicted change in a given target variable that results from a 1% change in the control variable.
For example, in an MIM with a target variable of new customers, the return on your promotional ads is estimated as the the predicted number of new customers generated from a 1% increase in promotional ads. In this way, you can evaluate the contribution of different control variables against one another.