Creating Metrics & Storyboards for Salary Regression Models

This guide helps you create the metrics and storyboard pages used to interpret a Salary Regression model in One Model. Building these assets typically requires coordination with your CS lead and data engineer to add the necessary code to your processing script so the model output tables and dimensions exist for metric and storyboard creation.  

In most cases, if you are building a new custom regression using the same regression “template shape,” you can reuse the same tables and dimensions and may not need additional data engineering work.

Module Type: Functional

Level: Intermediate-Advanced

Audience: Model & storyboard creators

Before You Get Started

Before your data engineer can add the necessary tables and dimensions to your processing script, you must deploy a Salary Regression model with SHAP enabled in the global settings.

If you have already deployed and built storyboards for other model types (for example classification), but have not deployed a regression model yet, your data engineer may need to add the regression table logic to your processing script after the regression model is deployed with SHAP enabled so the regression and SHAP-based metrics can be built.

1. Deploy a Salary Regression model with SHAP enabled in global settings.
2. Submit a ticket to have your data engineer add the required code to your processing script.
3. Allow your One Model site to reprocess.
4. Meet with your CS advisor to review the templated storyboard pages and metrics and decide what you want to build. You do not need to build every page if it is not relevant.
5. Using the metric guide, create the metrics you need.
6. Work with your CS team to build the storyboard pages you want to publish, and confirm viewers have appropriate access.

If you get stuck, contact your CS team for help.

Salary Regression Metric Guide

Use the Salary Regression metric guide as the source of truth for metric definitions and calculations and as a guide to build metrics for your salary regression model. You are welcome to modify metrics for your organizational needs. Ensure you are only providing metric and storyboard permission to appropriate data access roles. The standard template includes:

• Model-level summaries like predicted average salary, actual average salary, and average over/under prediction (Predicted minus Actual).
• Model performance metrics such as Explained Variance Score and Mean Squared Error.
• SHAP-based driver metrics such as Average Absolute SHAP Value (importance) and positive or negative SHAP impact (directionality).
• Salary distribution metrics at key percentiles (P10, P50, P90) for both actual and predicted salary.

Note: Some metrics are filtered to a specific augmentation. In our guide, this augmentation is called “salary.” You may name your model whatever you like and filter for it in these cases. 

Salary Regression Storyboard Template Pages

This section describes each page in the standard Salary Regression storyboard template and the value it provides. Your CS Advisor can show you these storyboards live upon request. 

Page 1: Regression Model Summary

What it is
An executive-friendly summary of the model’s latest run, including how well it performs and how predicted pay compares to actual pay for the selected population.

What you will typically see

• A plain-language summary of what the model predicts and which population it covers.
• Model details such as model method, feature selection, run status, and population size.
• Actual versus predicted average salary and the average over/under prediction (Predicted minus Actual).
• Actual versus predicted salary distribution at P10, P50, and P90 to show the spread across employees.
• Model performance metrics that summarize how well the model explains variation in salary and how large prediction errors tend to be.

Why it matters

• For stakeholders: A fast health check on whether the model is performing well enough to use and whether it tends to over- or under-predict salary overall.
• For analysts: A starting point to determine whether deeper interpretation is warranted or whether data, features, or population definitions need refinement.

Important interpretation reminders

• Percentiles describe the distribution of salaries across employees, not uncertainty for a specific individual.
• Performance metrics reflect the model’s overall fit and should be considered alongside population scope and data quality.

Page 2: Salary Regression Driver Analysis

What it is
A driver-focused view of what the model relies on most when predicting salary, and whether each driver generally pushes predictions higher or lower.

What you will typically see

• Drivers ranked by overall impact.
• A directionality view showing whether drivers tend to push predicted salary higher or lower in aggregate.
• Interpretation guidance, including the difference between correlation and causation and how category variables should be read relative to a reference group.

Why it matters

• For stakeholders: Increases transparency by explaining what factors are most associated with higher or lower predicted salary.
• For analysts: Helps validate whether the model is learning reasonable signals and flags surprising drivers that may require investigation.

Important interpretation reminder
Drivers explain what the model used to make predictions. They do not prove what causes salary.

Page 3: Salary Regression Filter Comparison

What it is
A comparison of drivers and directionality for the full model population versus a filtered subgroup selected using storyboard filters.

What you will typically see

• A prompt to apply filters to analyze a subgroup.
• An unfiltered driver view that represents the overall model population.
• A filtered driver view that updates when filters are applied.
• Summary blocks for unfiltered and filtered contexts, typically including population count, actual average salary, predicted average salary, and average over/under prediction.
• Performance context to help interpret subgroup results.

Why it matters

• For stakeholders: Supports clearer conversations about how model behavior differs for a specific group, while keeping the overall baseline visible.
• For analysts: Helps detect meaningful shifts in drivers or directionality by segment, which can guide follow-up analysis.

Important interpretation reminder
Filtered driver averages can be higher than unfiltered values even though the filtered group is a subset. This is expected when the metric is an average impact within the subgroup.

Page 4: Salary Regression Group Comparison

What it is
A set of group-based comparisons showing how actual pay, predicted pay, and the average prediction gap differ by key dimensions.

What you will typically see

• Actual average salary by group.
• Predicted average salary by group.
• Over/Under by group, defined as Predicted minus Actual.
• Population size to help interpret stability and representativeness.

Why it matters

• For stakeholders: Offers an intuitive way to see where predicted pay differs most from observed pay across groups, which can prompt targeted review.
• For analysts: Useful for diagnosing where model fit varies by segment and prioritizing follow-up cuts or data quality checks.

Important interpretation reminder
Group-level averages are influenced by group size, job mix, and data quality. Treat this page as a starting point for investigation, not a conclusion.

Using This Template for Other Regression Use Cases

This storyboard shape works for many regression problems where the outcome is a numeric value and you want a consistent interpretability approach:

• A model summary page focused on performance and distribution.
• A driver analysis page explaining importance and directionality.
• A filtered comparison page showing subgroup shifts in drivers.
• A group comparison page highlighting differences by key dimensions.

Examples of other regression use cases that map well to this template:

• Predicting bonus payout amounts
• Predicting sales attainment or revenue for commercial roles
• Predicting overtime hours or labor cost
• Predicting performance ratings when they are captured as a numeric score