Overview
Governance and Observability carries 10% of the exam across two objectives. The first covers managing and monitoring agents once they are live, and the second covers agent analytics and agent optimization. Together they describe the operating discipline that keeps agents useful and safe after deployment.
Monitoring includes watching agent health and usage in Agentforce observability tools such as the Agentforce Command Center, reviewing escalations to humans, and using session tracing to replay a conversation step by step: the topic selected, the actions called, their inputs and outputs, and the final response. Analytics, built on Data 360, report on session volumes, outcomes such as resolution and escalation rates, and topic usage. Optimization features group real utterances to show where requests fall outside existing topics, where the agent struggles and which changes would help. Governance also draws on the Einstein Trust Layer audit trail and feedback data for review.
How it's tested
Questions start after go-live. A manager reports rising escalations, a customer complaint names a bad answer, or leadership asks whether the agent is worth the investment. You choose the monitoring view, trace, analytics report or optimization insight that answers the question, then the change it suggests.
-
One conversation went wrong
A specific complaint is raised, and you choose session tracing to see topic, actions, inputs and outputs for that session.
-
Trends and outcomes
Leaders want resolution, escalation or adoption trends, and you choose agent analytics dashboards built on Data 360.
-
Find what to improve
Many requests go unresolved, and you use optimization insights on utterances to spot missing topics or weak actions.
Escalations from a Service agent doubled after a product launch. Session traces of a few escalated conversations show the agent selecting a general topic and finding no relevant action. Optimization insights group hundreds of recent utterances about the new product's installation, none matching an existing topic. The fix is a new topic with a knowledge action grounded in installation guides, followed by tests and continued monitoring.
Objectives 2
Official objectives from the Salesforce outline, in outline order.
5.1 Explain the process for managing and monitoring agents.Health, usage and session-level monitoring.
- What it means
- Managing agents means overseeing their status, versions, usage and behavior in production. Monitoring uses Agentforce observability features, including dashboards for health and usage and session tracing that records each step of a conversation, alongside escalation queues in Omni-Channel and Trust Layer audit data.
- What to understand
- Distinguish aggregate monitoring from investigation. Dashboards show whether the agent is healthy and used, while session tracing explains a single conversation: which topic was chosen, which actions ran with what inputs and outputs, any errors, and the final response. Monitoring data lives in Data 360, so it must be enabled. Governance adds controls such as version management, deactivating an agent when needed, reviewing feedback and limiting who can change agents.
- How it shows up
- A complaint about one conversation arrives, and you choose the monitoring tool that shows exactly what the agent did.
- Common confusion
- Using an aggregate dashboard to explain why a single conversation failed.
5.2 Explain agent analytics and agent optimizationOutcome metrics and improvement insights.
- What it means
- Agent analytics report on sessions, outcomes such as resolved, escalated or abandoned conversations, topic and action usage, and trends over time. Agent optimization analyzes real utterances and sessions to cluster requests, highlight unhandled intents and failure points, and suggest where topics, instructions or actions need work.
- What to understand
- Analytics answer how the agent is doing; optimization answers what to change. Use analytics for adoption and value questions, such as containment and escalation trends for leadership. Use optimization when the agent needs improvement and you want evidence of which requests it misses. Feed findings back into the lifecycle: change topics or actions, add tests to Testing Center, deploy and keep monitoring.
- How it shows up
- Leadership asks about value, or the team asks what to fix next, and you choose analytics or optimization.
- Common confusion
- Treating optimization suggestions as automatic changes; they still require builder review, testing and deployment.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Agent observability
Set of tools for monitoring, tracing and analyzing live agents.
- Remember
- Built on Data 360 data.
- Tested as
- Every monitoring question.
- Trap
- Expecting it without Data 360.
-
Agentforce Command Center
Central view of agent health, usage and performance.
- Remember
- Use for aggregate monitoring across agents.
- Tested as
- Operations oversight scenarios.
- Trap
- Using it to explain one session.
-
Session tracing
Step-by-step record of a single conversation.
- Remember
- Shows topic, actions, inputs, outputs and errors.
- Tested as
- Specific complaint investigations.
- Trap
- Choosing dashboards instead.
-
Agent analytics
Reports on session volumes, outcomes and topic usage.
- Remember
- Answers how the agent performs over time.
- Tested as
- Value and adoption questions.
- Trap
- Using raw message counts.
-
Agent optimization
Insights from clustered utterances and failure points.
- Remember
- Shows what to fix next.
- Tested as
- Improvement planning scenarios.
- Trap
- Assuming it applies changes itself.
-
Escalation rate
Share of sessions handed to a human.
- Remember
- Rising rates signal coverage or quality gaps.
- Tested as
- Service agent health questions.
- Trap
- Treating all escalations as failures.
-
Containment or resolution
Sessions completed by the agent without handoff.
- Remember
- Common value metric for Service agents.
- Tested as
- Leadership reporting.
- Trap
- Ignoring customer satisfaction alongside it.
-
Trust Layer audit trail
Stored prompts, responses and toxicity data.
- Remember
- Supports governance reviews.
- Tested as
- Compliance questions after go-live.
- Trap
- Expecting it to show action execution steps.
-
Feedback data
User ratings and comments on AI responses.
- Remember
- Stored in Data 360 for analysis.
- Tested as
- Quality improvement scenarios.
- Trap
- Ignoring it in reviews.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Which tool answers which question
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Session tracing | One conversation step by step | Explains why | Using it for trends |
| Agent analytics | Aggregated outcomes and usage | Shows how much and how well | Using it to debug one session |
| Agent optimization | Clustered utterances and gaps | Shows what to change | Expecting automatic fixes |
| Trust Layer audit trail | Prompt and response records | Supports compliance review | Expecting action-level detail |
Monitoring signals
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Rising escalations | More handoffs to humans | Coverage or quality gap | Adding instructions without data |
| Action errors | Failed action executions | Permissions or dependency issue | Blaming the model |
| Unmatched utterances | Requests outside topics | Missing topic | Broadening an existing topic too far |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Dashboard for a complaint
- What candidates pick
- Open the analytics dashboard to explain one bad answer.
- Why it looks right
- Dashboards are the first monitoring tool people think of.
- Why it's wrong
- Aggregates cannot show which topic, actions and inputs a single session used; session tracing can.
Clue to watch for A specific customer conversation is named.
-
Instructions fix everything
- What candidates pick
- Add instructions to handle new request types.
- Why it looks right
- It is the fastest edit.
- Why it's wrong
- If optimization shows requests outside every topic, the agent needs a new topic or action, not more text in an unrelated topic.
Clue to watch for Clusters of unmatched utterances.
-
Volume equals value
- What candidates pick
- Report total sessions as proof of success.
- Why it looks right
- It is a large, rising number.
- Why it's wrong
- Value depends on outcomes such as resolution, escalation and satisfaction, which analytics reports separately.
Clue to watch for Leadership asks whether the agent helps the business.
-
Auto-apply suggestions
- What candidates pick
- Accept optimization suggestions directly in production.
- Why it looks right
- They come from real data.
- Why it's wrong
- Changes still need review, testing in Testing Center and controlled deployment to avoid regressions.
Clue to watch for The scenario mentions production stability.
-
Audit trail as trace
- What candidates pick
- Use the Trust Layer audit trail to see which actions ran.
- Why it looks right
- It records AI activity.
- Why it's wrong
- The audit trail focuses on prompts, responses, masking and toxicity; session tracing shows the agent's reasoning steps and action calls.
Clue to watch for The question is about action execution.
Answer framework
Use these steps when a scenario concerns an agent already running in production.
- Clarify the question: is it about one conversation, an overall trend, or what to improve next? Each question points to a different tool.
- For one conversation, use session tracing to see the selected topic, actions, inputs, outputs and errors, which shows exactly where the reasoning went wrong.
- For trends and business value, use agent analytics on outcomes such as resolution, escalation and topic usage, filtered by channel and date range.
- For improvement planning, use optimization insights on clustered utterances to find missing topics or weak actions that users repeatedly request.
- Turn findings into changes, add matching Testing Center cases, deploy through the normal process and keep monitoring the same signals.
Practice question
Want more like this? The AI-201 practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
This section is short, so focus on mapping questions to tools. After building and testing an agent, run twenty or thirty conversations against it, including some it cannot handle. Then open the session tracing view and walk through two sessions step by step, noting how topic selection, action inputs and outputs and errors are displayed. Look at the analytics and optimization views in the same org to see what each offers, even if sample data is small.
Write a simple table with three columns: question, tool and likely fix. Fill it with cases such as a single complaint, rising escalations, an executive value review, a new type of request and a failing action. Trailhead modules on Agentforce analytics and observability, along with recent release notes, cover the official material; this area has changed more than most since 2025, so prefer current sources.
High-priority ideas are session tracing versus analytics and the improvement loop. Underestimated is the dependence on Data 360 for storing observability and audit data.
FAQ
How many questions come from Governance and Observability?
Governance and Observability is 10% of the exam per third-party copies of the current guide, so plan for about six scored questions across its two objectives.
Where did adoption monitoring go?
In the 2025 outline it sat under Development Lifecycle. The Spring '26 guide moves managing and monitoring agents, analytics and optimization into this section.
Is session tracing the same as debug logs?
No. Session tracing is an Agentforce observability feature that shows an agent's reasoning steps and action calls for a conversation, while debug logs capture Apex and platform execution details.
Do I need Data 360 for agent analytics?
Yes. Agent analytics, session tracing data, Trust Layer audit records and feedback are stored in Data 360, so it must be provisioned.
Does agent optimization change the agent automatically?
No. It surfaces patterns and suggestions. A builder reviews them, makes changes, adds tests and deploys through the normal lifecycle.
What metric best shows Service agent value?
Outcome metrics such as resolution or containment and escalation rates, ideally alongside customer satisfaction, show value better than session totals.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 5 knowledge
You have covered monitoring, session tracing, agent analytics and optimization. Check whether you can pick the right observability tool for each post-launch question.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the AI-201 AI tutor Get any concept explained your way