Define Your Target Outcome

Preliminary Considerations

To begin exploring data with Einstein Discovery, consider some preliminary questions:

• What outcome (such as a KPI) do you want to explore? Einstein Discovery detects patterns related to three types of use cases: numeric (measures), binary classification (two-value text outcomes), and multiclass classification (from 3 through 10 possible outcomes). Broadly speaking, promising solution candidates often involve KPIs associated with large volumes of data and many business decisions.
• What explanatory variables do you want to include in analysis? These factors can influence your outcome.
• Where can you find this information? What are possible data sources? Salesforce objects? Data that is external to Salesforce?
• Is there enough data for Einstein to analyze? For details, see Einstein Discovery Capacities and Requirements.

For more ideas, see Everything You Need for Einstein Discovery.

Identify the Outcome Variable You Want To Analyze and Improve

Decide which outcome variable you want to explore, and at what granularity. The outcome variable could be a KPI value (such as revenue, discount, cost measure, or duration) or other quantifiable outcome. You can also use categories (text fields) with two values (binary) or 3–10 values (multiclass) as an outcome variable. In general, binary outcomes are less accurate to predict than continuous value outcomes. Occasionally, a new metric is created, such as customer revenue by month. It’s possible to create outcome variable metrics.

In addition, clarify your goal. For numeric and binary classification outcomes, Einstein Discovery orients its analysis based on maximizing or minimizing the outcome variable. For example, your goal can be to maximize net margin or minimize customer churn. For multiclass classification outcomes, your goal is to predict the most likely outcome among 3–10 possible values.

Identify Explanatory Variables to Examine

Think about which variables can possibly describe or influence the outcome. For example, to investigate sales, potential influencer variables can include Discount, Days between Lead Received and Last Contacted, Lead Source, Region, Vertical, Competitor, and Promotion. When selecting predictor variables, you want to gather a maximum amount of information from a minimum number of variables. Einstein Discovery helps this process by eliminating variables that do not have good explanatory power from the model it generates.

Note For proof-of-concept projects, keep your input data from 10 through 25 variables. It’s faster to learn and improve your data preparation skills with a less complex model.

Select any fields (predictor variables) that directly affect the outcome. Ensure that the variable data is clean and consistent. The order and meaning of input predictor variables must remain the same from record to record. Inconsistent data formats, “dirty data,” and outliers can undermine the quality of analytical findings.

Then you shape the data into analytical fields with derived variables that describe or influence the outcome variable. Shaping data in a CRM Analytics dataset involves subject-matter expertise and data literacy to successfully select, create, and transform variables for maximum influence.

Identify Data Sources and Fields

Based on your outcome and explanatory variables, determine which data sources can best represent the variables for business processes associated with the outcome variable. Potential sources include Salesforce objects (including custom objects) and data that is external to Salesforce.