Models
The model functionality in Mix Modeler allows you to configure, train, and score models specific to your business objectives. The training and scoring supports AI-driven transfer learning between multitouch attribution and marketing mix modeling.
The models are based on the harmonized data that you create as part of the Mix Modeler application workflow.
A model in Mix Modeler is a machine learning model employed to measure and predict a specified outcome based on a marketer鈥檚 investments. Marketing touchpoints and summary-level data can be used as an input. Mix Modeler allows you to create variants of models based on different sets of variables, dimensions, and outcomes, such as revenues, units sold, leads.
A model requires:
- One conversion.
- One or more marketing touchpoints (channels) comprised of summary-level data, marketing touchpoint data (event data) or both.
- A configurable lookback window.
- A configurable training window.
A model can optionally include:
- External factors.
- Internal factors.
- Prior knowledge of marketing contributions from other sources such as past stakeholder experience, incrementally testing, other models.
- Spend share, which uses relative spend share as a proxy when marketing data is sparse.
Create a model
To create a model, use the Mix Modeler step-by-step guided model configuration flow available when you select Open model canvas. See Create a model for more details.
Manage models
To view a table of your current models, in the Mix Modeler interface:
-
Select
-
You see a table of the current models.
The table columns specify details about the model.
table 0-row-2 1-row-2 2-row-2 3-row-2 4-row-2 5-row-2 6-row-2 layout-auto Column name Details Name Name of the model Description Description of the model Conversion event The conversion you have selected for the model. Run frequency The running frequency of training the model. Last run The date and time of the last training of the model. Status The status of the last run of the training of the model.
-
To change the columns displayed for the list, select
You can take the following actions on a specific model.
Model insights
The model insights functionality is only available on successfully trained and scored models.
To view the insights of a model:
-
Select
-
Select the model name.
You are redirected to Model Insights.
View details
To view more details of a model:
-
Select
-
Select
Duplicate
You can quickly duplicate a model.
-
Select
-
Select
Edit
You can edit the name, description and the scheduling of training and scoring of a model.
-
Select
-
Select
In the Edit model dialog:
-
Enter a new Name and Description.
-
To enable scheduling, enable Status. You can only enable scheduling for models that are trained and scored.
-
Select a Scoring frequency:
- Daily: Enter a valid time (for example
05:22 pm) or use
- Weekly: Select a day of the week and enter a valid time (for example
05:22 pm) or use - Monthly: Select a day of the month from the Run on every dropdown menu and enter a valid time (for example
05:22 pm) or use
- Daily: Enter a valid time (for example
-
Select a Training frequency from the dropdown menu: Monthly, Quarterly, Yearly, or None.
-
-
-
Select Save.
Re-train
Re-train a model is only available on successfully trained models.
Consider to re-train a model when you want to:
- Include new incremental marketing and factor data. For example, over the last quarter, market dynamics have changed or your marketing data distribution has changed significantly.
To re-train a model:
-
Select
-
Select
In the Train model dialog, select the option to:
- Train model with last 2 years of marketing data, or
- Train model using specific date range of data.
Specify the date range. You can use the
-
Select Train to re-train the model.
Score or re-score
You can incrementally score a model based on new marketing data or re-score a model for a specific date range.
Consider to re-score a model when you want to:
- Correct incorrect marketing data. For example, the recent paid search data you included in the training and scoring of the model missed a week of data.
- Use new incremental marketing data that has become available through updates in the datasets you have configured as part of your harmonized data.
To score or re-score a model:
-
Select
-
Select
In the Score marketing data dialog, select the option to:
- Score new marketing data from mm/dd/yyyy, to score your model incrementally using new marketing data, or
- Score specific date range of marketing data to re-score for a specific date range.
Specify the date range. You can use the
-
Select Score. When re-scoring a model using a specific data range, you see an Existing model is replaced dialog, prompting you to confirm to replace the model with new scores for the selected date range. Select Replace model to confirm.
Delete a model
To delete a model:
-
Select
-
Select
To delete multiple models:
-
Select multiple models.
-
From the blue action bar, select
note warning WARNING The model is deleted immediately.