One AI Generative Attributes with Video

Generative Attributes create data that can be richer in predictive value than what's available on individual employee records.

What's a Generative Attribute?

Generative attributes are additional data points that can be added to the data frames for your predictive models that are contenders for being selected as features. They exist because the way we model data for analytic purposes is often different from how we want our data structured for machine learning.

A standard non-generative attribute is something about a person (or job or anything else you're making predictions for) at a point in time, such as their salary. A generative attribute can be a count of events or unique attributes over a specified period of time, a flag if a person meets a set of criteria, or a higher level aggregation such as team.

Generative attributes enable the creation of data that is often richer in predictive value than the things that are stored on individual employee records in your HRIS.

Flavors & Examples of Generative Attributes


At a high level, the “flavors” of generative attributes are metrics, customized metrics, and aggregations. By aggregations we mean data grouped by something other than the unique identifier in the base query. Team or Location are examples. Generative attributes can be created for a specified period of time or for all time up through the population date. They can also be binary (Has Been Promoted for example).


For the examples listed below, the following apply:

  • Time selection is configurable independent of the base query

  • Multiple versions of a generative attribute can be created for different time selections. For example:

    • Promotions Past Year

    • Promotions All Time

  • Other attributes such as "Location" can be substituted for "Team"

  • The following is a list intended to generate ideas. You may not have the data available to create some of these examples


Direct Reports

Team Average Age



Team Average Tenure

Manager Changes

Position Tenure

Team Headcount

Assignment Changes

Level Tenure

Team Average Salary

Job Changes

Pay Grade Tenure

Team Promotions

Location Changes

Compa Ratio

Team Hires

Org Changes

Level of Education

Team Transfers

Pay Grade Changes

Application Source

Team Terminations

Position Changes


Team Avg Review Score

Corrective Actions

Safety Incidents

Team Avg Engagement

Navigating to Generative Attributes in One Model

Generative attributes are created and managed in the One AI query builder. “Query builder” is what we call the data framing component of One AI. For reference, "Recipes" are part of the query builder. Generative attributes are available in both the Recipes configuration and the Custom / Advanced Model setup. For more information about One AI Recipes, please see One AI recipes overview.  

  1. Select One AI from the top navigation bar

  2. Select the + Add Machine Learning Model button

  3. Give the model a name. Please note that models cannot be re-named once they are saved

  4. Under Using Data From select One AI Recipe

  5. Select Configure One AI Recipe

  6. Select an option from the What are you interested in predicting? dropdown

  7. Make the initial required selections to enable Attribute selection

    1. For a recipe, make all required selections for the first two questions. This will enable the "Which generative attributes do you want to use in your prediction?" question to be expanded, which is the last question in the series.

    2. For a Custom / Advanced Model, make all required selections in the "Define Prediction" and "Define Population" sections. Then proceed to the Select Attributes section and scroll down to Generative Attributes below the Column Selections.

Managing Generative Attributes

The list of existing generative attributes is searchable and can be included in or excluded from your query by selecting the checkboxes next to them. They are not selected by default when you create a new one, so be sure to select the ones you want included. Existing generative attributes can be edited or deleted by selecting the corresponding icons next to each.

Once you create generative attributes, they can be used in multiple Augmentations. They’re saved to your site, not to the specific augmentation you’re creating. That said, only generative attributes that can be joined to the fact table from which the population metric for your augmentation was created will be displayed. If you create a generative attribute for a recruiting augmentation and then go back and create a core workforce augmentation, the recruiting generative attribute may not be listed.

Creating New Generative Attributes

As you create generative attributes, note that there are information icons that display helpful tips when you roll over them. Hopefully these are useful the first few times you use the tool.


Starting simple, a generative attribute can be just an existing metric on your One Model site. Promotions or Transfers are examples. This is how you would configure a generative attribute for Transfers in the past year:

Customized Metrics

Metrics can be filtered independent of the base query and can also be tied to periods of time that differ from the core query, including All Time. Binary attributes can also be created. These capabilities can be combined in some cool ways. You could create a generative attribute that denotes whether or not each person has ever had a Promotion. You could even get more fine grained and limit it to Promotions that were coded as a Progression.

The builder automatically generates a name for the Generative Attribute but it can be customized. This one is a bit long so I’m just going to call it “Has Been Promoted as Progression”.



Probably the most exciting thing that Generative Attributes enable is aggregations that differ from the base query. Your base query is usually as granular as possible, meaning a row per person. Attributes about each person individually don’t tell the whole story about what they experience at your company as an employee though. Things like team or office location dynamics often correlate with job satisfaction and also attrition. Generative attributes allow you to add data rolled up to the team or location level, among others, to the query. Using Group By is how aggregations are configured. Be sure that the column you select to Group By is contained in your selected core attributes. Failure to do so will result in all null values for that attribute.

Here's a generative attribute for Team Headcount:

Selecting & Validating Generative Attributes

Once your generative attributes are created, remember to select the ones you want included in your data set since they are not selected by default.

The final step before saving your recipe is to validate your selected Generative Attributes to ensure data integrity. This can be done by expanding the "Would you like to verify that all of the selections you have made are valid?" section of the recipe and Generate Data Statistics. A report containing statistics about all of the data the recipe generates will be displayed. Generative Attributes should be listed near the top of this report.

In the image below, the highlighted rows are Generative Attributes. Note the "Non-null Count" values for each. In cases where the non-null count is significantly less than the overall count, null filling may need to be applied. This can be done on the Augmentation configuration screen in the "Per Column Interventions" section.

Next Steps

Once you're happy with the Generative Attributes you have selected, you can finish configuring your recipe if you haven't already and save it.

After running your augmentation, you can observe the fruits of your labors. Selecting a pending run opened the EDA report. Scrolling down to Variable Status will tell you what One AI did with your generative attributes. Hopefully you have some success with them being selected as features.

We provided examples of three generative attributes in this article but there are many possibilities for what you can create. See the list above for inspiration.

Video - Generative Attribute Guide

Watch this video to learn how to create and incorporate generative attributes in your machine learning models to improve performance and robustness. 

You will learn:

  • The concept of generative attributes and their importance in providing additional, meaningful information beyond the original dataset
  • How to create effective generative attributes in One AI by selecting appropriate metrics, time selections, grouping, & filters 
  • The importance of validating generative attributes to ensure data integrity & model readiness before running machine learning models
  • How generative attributes can improve model performance by enriching the feature set & enabling the discovery of more complex patterns in the data

Audience: Intermediate I-Plane.svg

Before watching this video, you should know how to:

  • Create a machine learning model with a recipe
  • Use metrics in Explore
  • Interpret the EDA Report

Run Time: >19 minutes

Was this article helpful?

0 out of 0 found this helpful



Please sign in to leave a comment.