Data insights are discoveries derived from analyzing your data. With each model iteration, Einstein Discovery meticulously examines your data, unveiling valuable insights. These insights are rapid, unbiased, and statistically significant, presented through charts and detailed explanations for easy comprehension. They serve as a launchpad for exploring the connections between your model’s input variables and its objectives. In this article, we will explore data insights in Einstein Discovery.
To access data insights, navigate to the left panel and click on “Data Insights.” This action prompts Einstein Discovery to showcase the data insights screen.
The model summary (1) provides essential details including the model goal, total rows analyzed, average outcome, and any deviations from the previous version. You can toggle the variables panel by clicking the “Explore a Variable” button.
The variables panel (2) displays the explanatory variables within your model along with their correlation to the model outcome, arranged in descending order of correlation strength. Correlation is expressed as a percentage, indicating the degree of statistical relationship. For instance, in our model, Loan Amount exhibits the highest correlation with maximizing Credit score, followed by Married.
The insight summary panels (3) highlight the most impactful variables positively and negatively influencing the outcome variable.
Einstein Discovery generates various types of insights:
Descriptive:
Derived from historical data through descriptive analytics and statistical analysis, descriptive insights serve as the foundational insights. They offer an overview of the contributing factors to the outcome, informed by a statistical examination of your dataset. Einstein Discovery employs bar charts to illustrate disparities, patterns, and significance in the values.
Insights are organized by their statistical significance, with the top insights representing the most influential factors on the model outcome.
The goal set here is to maximize credit score. Each insight indicates whether something “was better” or “was worse” in relation to that goal. A green circle with an arrow indicates a condition that takes you closer to your goal. A red circle with an arrow indicates a condition that takes you further from your goal.
This insight shows you that, among various Ethnicities, borrowers of Cuban Ethnicity have a credit score higher than the average credit score, while those of Puerto Rican Ethnicity have credit score lower than the average.
In the graph, hover over a bar to view a popup that shows underlying details.
Drill Down to a Single Variable (First-Order Insights)
The insights list shows insights that are associated with Race only. The filter selector above the insight shows the variable you’re investigating.
The first insight in the list shows how the Race is associated with the outcome. It represents a summary of all values associated with the variable. For descriptive insights, this is also known as a first-order insight, because it examines how one variable (Race) explains variation in the outcome variable (maximize credit score).For eg,From the graph it is visible that the race ‘American Indian or Alaska Native’ has the maximum value from the average credit score which is 605.45.
Explore Subgroups (Second-Order Insights)
Scroll down to the next insight in the list.
This insight shows how a combination of variables (subgroups) are associated with the outcome variable.
In this article we explore data insights in Einstein Discovery focusing on descriptive insights which are the foundational data insights in Einstein Discovery. We will cover the diagnostic and comparative insights in another article.
To learn more, click Introduction to Einstein Discovery (salesforce.com).
Take Five Consulting is a technology company, based in Virginia U.S., that specializes in the Mortgage Banking vertical especially LOS implementation and application development. Take Five Consulting creates and implement mortgage technology and software specifically for Mortgage Industry.
