Set Up Agent Optimization
Set up Agent Optimization to analyze agent sessions, identify knowledge gaps, and dive deeper into unresolved interactions using the Session Tracing Data Model.
Required Editions
| Available in: Enterprise, Performance, and Unlimited Editions with an Einstein for Sales, Einstein for Platform, Einstein for Service, Einstein 1 Service, or Einstein GPT Service add-on. To purchase add-ons, contact your Salesforce account executive. |
| Access and view Agent Optimization | Assign Access Agent Optimization and Data Cloud User permission sets |
- Salesforce Standard Data Model version 1.130 or higher is required.
- Agent Optimization supports Agentforce (Default), Agentforce Employee Agent, and Agentforce Service Agent.
- Agent Optimization supports agent sessions in all languages. However, the analysis and insights are presented in English.
- Agent Optimization supports voice agents.
Einstein Trust Layer Service Support: Feature supports Data Masking, Data Limitation and Billing, and Privacy (PII Masking). Agent Optimization impacts your credit consumption. Data 360 credits are used for storing and querying Agent Optimization data.
- For most organizations, OpenAI is the default model provider for Agent Optimization session analysis.
- In Civilian Government Cloud organizations, Anthropic is the default because OpenAI isn't supported for batch calls in that environment.
- Agent Optimization has a hard dependency on OpenAI (or Anthropic in Civilian Government Cloud organizations). Model provider enablement and disablement for Agentforce is generally available. In Manage Model Provider Access, you can turn model providers on or off org-wide, but disabling OpenAI (or Anthropic in Civilian Government Cloud organizations) breaks Agent Optimization. Keep OpenAI (or Anthropic in Civilian Government Cloud organizations) enabled if you use Agent Optimization.
- In commercial organizations, changing the default optimization behavior from OpenAI to Anthropic requires a special request through Salesforce Support.
- Session analysis doesn't follow your agent's model selection. When Agent Optimization is on, session data is sent to the assigned provider. To stop this processing, turn off Agentforce Session Tracing and Data Model in Einstein Audit, Analytics, and Monitoring Setup.
- Agent Optimization doesn't support geo-aware routing. LLM requests may be processed outside your org's geographic region.
Set Up Agent Optimization for Standard Environments
Enable Agent Optimization in your Salesforce organization to start analyzing agent sessions and optimizing performance.
- Before enabling Agent Optimization, make sure that Data Cloud is provisioned and turned on in your organization, and that you have an active agent.
- Verify that you have the latest version of the Salesforce Standard Data Model (version 1.130 or higher) in your org (sandbox or production). From Setup, go to Apps -> Packaging -> Installed Packages -> Salesforce Standard Data Model. To get the required version, click here and follow the instructions.
- In Setup, in the Quick Find box, enter and select Einstein Audit, Analytics, and Monitoring Setup.
- Turn on Agentforce Session Tracing and Data Model. Agent Optimization is automatically enabled when you turn on Session Tracing.
- In Setup, locate Users.
- Assign users the Access Agent Optimization permission set.
- Save your work.
Set Up Agent Optimization for DC1 Environments
DC1 (Data Cloud One) support refers to a configuration where multiple Salesforce organizations share a single Data 360 instance. This setup allows for efficient data management and collaboration across organizations, with one "home organization" and multiple "companion organizationanizations" or "LOB (Line of Business) organizations".
Companion Organization Enablement: To enable Agent Optimization for companion organizations, Einstein Setup must first be enabled in your home organization. After that, the process for enabling Agent Optimization in each organization remains the same. If you enable Agent Optimization from a companion organization, Data Cloud assets will be provisioned in the home organization and then shared to the companion organization.
The home and companion organizations operate independently, and each organization displays analytics only for its own agents.
You can select different data spaces for each organization where Agent Optimization is enabled.
If more than one agent with the same API name exists in more than one organization and the organizations share the same data space, the agents are aggregated into the same view and appear as one agent.
To avoid this limitation, do one of the following:
- Create different API names for agents across organizations.
- Select different data spaces for each organization.
- Create a data governance rule to control which data groups will have viewing permissions.
- Ensure your home organization has Generative AI enabled.
- Ensure your companion organization has Salesforce Foundations or Agentforce 1 and an active companion connection.
- Verify Data Cloud One appears in your companion organization.
- Assign users the required permission sets in your companion organization: Access Agent Optimization and Data Cloud One User.
Set Up Agent Optimization for Sandbox Environments
Enable Agent Optimization in sandbox environments for testing and development purposes. Sandbox support requires specific configuration steps to reconnect connectors and activate data streams.
- Provision Data Cloud in the sandbox organization.
- Install or upgrade the Salesforce Standard Data Model package to version 1.130 or higher in the sandbox organization. When installation is complete, verify the version on the Installed Packages page in Setup.
-
Reconnect connectors.
Note While connections are replicated to your sandbox, authentication data is not. As a result, all connectors will appear in a "Needs Activation" state.From Setup, go to Data Cloud Setup > Ingestion API > AgentOpt, and click Activate.
-
Retoggle Agentforce Session Tracing and Data Model. Agent Optimization is automatically enabled when you turn on Session Tracing.
All data streams should then activate automatically.

