Section 6 · Governance and Observability · 10% of exam
6.2 Agent Analytics and Performance Signals
Exam objective
Explain what Agent Analytics measures and how administrators use performance signals to assess service and employee agents.
Aggregate signals into a dashboard
Agent Analytics summarizes performance data across agent sessions so an administrator can compare trends over a selected period, agent, channel, or modality. Salesforce describes outcome areas that can include effectiveness, usage, quality, health, trust, and voice. Dashboards provide a broad view; they help answer whether the agent is being used, whether it is completing work, and whether its quality or operational behavior is changing. A dashboard is not a complete diagnosis. When a chart shows a spike or decline, narrow the filters and open the related sessions to understand the underlying interactions.
Read effectiveness and usage
Effectiveness measures help show whether an agent is meeting its purpose. Depending on the dashboard and agent type, useful signals can include task resolution or success, deflection, escalation, abandonment, engagement, and errors. Usage signals show how many sessions or interactions occur and how they vary by channel or time. These measures need context. A rise in escalations can indicate poor routing, a deliberate policy to involve people in complex cases, or a change in customer needs. A higher session count may reflect growth, a deployment change, or repeated unsuccessful attempts. Interpret signals against the use case and the period being compared. Review quality, trust, and latency
Quality signals help assess whether responses are accurate, relevant, complete, coherent, or aligned with instructions. Trust signals can show instruction adherence, guardrail outcomes, or user feedback, depending on enabled data sources and dashboards. Latency measures help identify slow interactions or processing steps. For voice agents, channel-specific information such as response timing and call conditions can add context. Metric definitions matter: response time to start speaking is different from time until the full response completes. Before acting on a KPI, confirm how it is calculated, what data feeds it, which sessions are included, and whether tracing or audit features are enabled.
Compare, segment, and investigate
Use filters to compare agents, channels, time periods, or user groups that are relevant to the business question. Look for changes over time and differences across segments. Then use Sessions & Intents to inspect transcripts, summaries, actions, and low-quality interactions. Salesforce notes that dashboards show data available in the current org, so understand where data from sandbox and production is stored before comparing environments. Avoid relying on a single average: it can hide a poor-performing intent or channel. Pair aggregate trends with session details, user feedback, and the agent’s release history to decide whether an issue is real and where to investigate next.
When a KPI changes, segment it by agent, channel, timeframe, or intent, then inspect example sessions before changing the configuration.
Available dashboard tabs and metrics can vary by agent type, org configuration, edition, and enabled observability sources.
Extra concept: An average can improve while an important high-risk intent gets worse. Review segments and individual sessions as well as totals.
Official references:
- Get Insights with Agent Analytics https://help.salesforce.com/s/articleView?id=ai.generative_ai_agent_analytics_use.htm&language=en_US&type=5
- Observe Phase Observability Tools https://help.salesforce.com/s/articleView?id=ai.agentforce_observability_observe_phase_tools.htm&language=en_US&type=5
- Understand Agentforce Observability KPIs and Tools https://help.salesforce.com/s/articleView?id=ai.agentforce_understand_observability_kpis_and_tools.htm&language=en_US&type =5
Key takeaways
- Analytics summarizes usage, effectiveness, quality, health, trust, and channel performance.
- Metric definitions and enabled data sources affect what a dashboard represents.
- Use filters and session details to explain trends rather than acting on a single aggregate score.
