Just like its name, it uses AI to let Einstein learn from historical data to make predictions about almost any field. You can then use the predictions to alert the reps on what to do next.

Let’s say, XYZ company’s restaurants are famous in town and has online reservation system built via salesforce. Since XYZ has many customers, it is very important for the general manager to allocate proper staffing to all the tables in restaurants. But, there is been a regular issue that customers aren’t always show interest about canceling reservations they can’t make. This has severe impact on staff allocations. Since there are too many reservations and too few employees, calling customers to confirm about reservation is been headache.

Einstein Prediction Builder helps to predict the likelihood that customers show up for a reservation, so you can prioritize which customers to call for confirmation.

ZYZ has an SF org with a custom object named Reservation. Each Reservation record represents a customer’s reservation at one of XYZ’s restaurants.

Reservation has Status picklist field with following options:

  • Completed
  • No Show
  • Upcoming

Now, the prediction builder’s goal is to predict how likely the status = “No Show”.

Below are the prerequisites/steps to prediction builder creation:

  • Prediction Name

Any sensible name for the prediction.

  • Prediction Object

This is the object to predict. In our case, Reservation.

  • Prediction Field

Einstein can predict number fields, checkboxes, and picklists. Though status is a picklist field, we want particularly to predict “No Show” value. Let’s create a checkbox-formula field on Reservation named “No Show” with the formula:

ISPICKVAL(Status__c,"No Show")

Since this is of type checkbox, we can take this as prediction field.

  • Save-to field

For checkbox fields, Einstein gives you the probability that the value is True.

For picklists, Einstein gives you the most likely value plus its confidence in this field.

For number fields, Einstein gives you the predicted value.

Create a custom field of type number to store prediction results in our case. Let’s say, this field is named “Predicted No Show”.

  • Fields Used to Make a Prediction

By default, Einstein analyzes all fields on the predicted object. If you want Einstein to skip a field, deselect it. Most of the time, it’s best to leave Einstein’s default selections as they are.

  • Data segment

Segmenting your data lets you focus your prediction on groups, such as a specific group of clients, a sales region, or a division in your company.

This comes with 2 options:

  • No segment (use all data selected on previous page)
  • Yes, focus on a segment (Advanced)

Leave default (first one, in our case).

  • Example records

Some records have a known value for a field you want to predict. Use those records as examples.

Einstein needs to know which records have the right values. Then it can analyze these records to make predictions for records with unknown values.

Always this must be historical data. Select past reservations, so it’s known whether they were no-shows or not. Select the “Status” field, the “Does not equal” operator, and the “Upcoming” value

Review your selections and click on “Build Prediction” button to trigger the AI to run at the back end.

Happy Coding!