Tuesday, December 31, 2019

Einstein Prediction Builder: Getting Started

Einstein Prediction Builder allows you to easily build predictions across any Salesforce standard or custom object with clicks, no code! It predicts a field based on a set of historical data.

If you have Einstein Predictions or Einstein Analytics Plus license, then Einstein Prediction Builder is available for you to use, source Einstein Analytics Pricing (as of 31 Dec 2019).

If you org. have Einstein Prediction Builder license, you should see Einstein Prediction Builder in the setup menu.

There is a few blogs have shared on Einstein Prediction Builder use cases and how to configure it:
Einstein Prediction Builder. The Artificial Intelligence by Salesforce
Einstein prediction builder to predict
Einstein Prediction Builder

Einstein Prediction Builder comes with a wizard-style configuration, you need to select an object, a field for prediction, a set of records as "example", and field to store the prediction result.

Currently, it only supports to predict a checkbox field, while the result would be in number between 0 to 100. In summary, here are the steps:

1. Object
Select a Salesforce object you would like to use for prediction.

2. Segment
Do you want to use the whole data of the object for prediction and learning? Select:
- No segment, to use all data, or
- Yes, enter filter of fields to be included
Make sure you have a minimum of 400 records (after filter if you segment the data).

3. Field to Predict 
This is a very important field, Einstein will do the prediction based on historical value of this field, and it needs to be a checkbox or formula that returns checkbox.

In this blog, I have a picklist field called "Status" with value: Sent, No Show, and Attended. No Show and Attended is historical data, while Sent is new records, and Einstein will help us to predict the attendance for all Sent records.

Because it needs to be a checkbox field, let us create a formula field that returns checkbox called "Attend", ISPICKVAL(Status__c,"Attended"), this field will return True if the Status field is Attended.

4. Define a set of data as examples (for learning)
As mentioned that Einstein needs historical data to predict, we need to define historical data for learning. Einstein will not predict records set as examples, but use it to learn the value of "field to predict".

The more samples data (historical) is a better prediction, but make sure to have a minimum of 400 example records, with both Yes and No records should have a minimum of 100 records.

From the above sample screenshot:
- 407 is the historical data for learning, which are records with Status does not equal to Sent
- 188 of them is Yes <-- this means Attend value is True
- 219 of them is No <-- this means Attend value is False
- 3399 records with Status = Sent, this needs to be predicted for the attendance

The below image explains dataset scope:

- Green: whole data in an object
- Yellow: data filtered after segment, data outside yellow will not be used for prediction and learning
- Dark Yellow: historical data set for learning (Status = No Show and Attended)
- Light Yellow: new data set for prediction (Status = Sent)

5. Set a field to store prediction values
Once the predictions enabled, the values will be range from 0 to 100.

Depends on your data size, Einstein may take up to 24 hours to build the prediction, then you can review the scorecard. Once you enable the prediction, Einstein will start to populate the prediction field with a score, including new records.


No comments:

Post a Comment

Page-level ad