PredictionBuilderSettings in SFDX

Architectural Overview#

Einstein Prediction Builder delivers custom machine learning capabilities across the Salesforce platform without requiring dedicated data science pipelines. By analyzing historical records on standard and custom objects, it automatically trains predictive models to calculate outcomes such as customer churn risk, invoice late-payment probability, and opportunity win likelihood.

The PredictionBuilderSettings metadata type represents the administrative settings governing Einstein Prediction Builder. Enabling this metadata type activates the AI inference engine, establishes data processing connections, and permits models to write calculated scores directly to custom fields.

In automated CI/CD environments, deploying PredictionBuilderSettings ensures that developer sandboxes and scratch orgs match production predictive capabilities, preventing regression in intelligent automation.

Directory and File Conventions#

PredictionBuilderSettings is stored as a singleton settings file within the settings/ directory:

Key Metadata Fields#

The XML schema of PredictionBuilderSettings provides top-level enablement switches:

Declarative XML and Package Manifest Examples#

Below is a production PredictionBuilder.settings-meta.xml configuration:

<?xml version="1.0" encoding="UTF-8"?>
<PredictionBuilderSettings xmlns="http://soap.sforce.com/2006/04/metadata">
    <enablePredictionBuilder>true</enablePredictionBuilder>
    <isPredictionBuilderTrial>false</isPredictionBuilderTrial>
</PredictionBuilderSettings>

Scratch Org Configuration (config/project-scratch-def.json)#

{
  "orgName": "Einstein Prediction Scratch Org",
  "edition": "Enterprise",
  "features": ["EinsteinPredictionBuilder"],
  "settings": {
    "predictionBuilderSettings": {
      "enablePredictionBuilder": true
    }
  }
}

Package Manifest (package.xml)#

<?xml version="1.0" encoding="UTF-8"?>
<Package xmlns="http://soap.sforce.com/2006/04/metadata">
    <types>
        <members>PredictionBuilder</members>
        <name>Settings</name>
    </types>
    <version>68.0</version>
</Package>

CLI Retrieval and Deployment Commands#

Manage PredictionBuilderSettings using Salesforce CLI commands:

# Retrieve PredictionBuilder settings
sf project retrieve start --metadata Settings:PredictionBuilder

# Deploy PredictionBuilder settings to the target org
sf project deploy start --metadata Settings:PredictionBuilder

# Deploy and validate in a single transaction
sf project deploy validate --metadata Settings:PredictionBuilder

Architect Best Practices and Gotchas#

• Feature License Dependencies: Attempting to deploy enablePredictionBuilder: true into an org lacking Einstein Prediction Builder licenses will trigger deployment errors such as Feature not supported.
• Target Custom Field Provisioning: Before deploying or activating models created in Prediction Builder, ensure the receiving CustomField (typically a Number(3, 0) for probabilities or Text for categories) exists in the target environment.
• Trigger Recursion Awareness: Predictions update records via background automated processes. If target records have heavy record-triggered Flow or Apex triggers, ensure recursion guards are implemented to prevent governor limit exhaustion during bulk scoring cycles.

Frequently asked questions

What does PredictionBuilderSettings control in Salesforce?

It enables Einstein Prediction Builder, allowing declarative machine learning models to score records and populate predictive custom fields.

Where is PredictionBuilderSettings stored in an SFDX project?

It is located at force-app/main/default/settings/PredictionBuilder.settings-meta.xml and retrieved as Settings:PredictionBuilder.

Can PredictionBuilderSettings be deployed to scratch orgs?

Yes, provided the scratch org definition file includes the EinsteinPredictionBuilder feature license.

How does Einstein Prediction Builder output its predictions?

Models write numeric probabilities or categorical predictions directly to designated custom fields on the scored object.