Agent Analytics: How to Monitor Agentforce Agents

Hey everyone,

Your agent is live. People are talking to it. Now the real question: is it actually helping them?

To monitor Agentforce agents, use three tools. Agent Analytics gives you dashboards. Sessions & Intents in Agentforce Studio lets you trace one conversation step by step. Enhanced event logs show every message in a session.

Quick terms. A session is one conversation between a user and an agent. An intent is what the user was trying to do. A trace is a step-by-step record of what the agent did.

And one warning before anything else: the old Agentforce Analytics is being retired. More on that at the end.

“Is my agent working overall?” Use Agent Analytics

Agent Analytics is your dashboard view of agent performance and conversation metrics.

Per Set Up Agent Analytics, it’s available in Enterprise, Performance and Unlimited Editions with Salesforce Foundations or Agentforce 1 Edition. You install the Agent Analytics apps from Setup, and admins share the workspaces so users can see the dashboards.

One thing to know up front: Help says data only appears for new conversations after you set up the Session Tracing data model, and it can take a while. So don’t panic when last month’s conversations aren’t there.

“Why did this one conversation go wrong?” Use Sessions & Intents

This is where you play detective. From the Trailhead unit: open the App Launcher, select Agentforce Studio, then under Observe & Optimize select Sessions & Intents.

You’ll see two tabs:

• Processed Sessions have intents, cluster tags and quality scores. Intents usually take 4 to 5 hours after a session closes. Clustering (grouping similar intents) runs weekly.
• Unprocessed Sessions are active, just closed or still processing. You can still open their Trace tab right away.

So if you can’t find a conversation from ten minutes ago, check Unprocessed. It’s there.

Click an intent to open its Interaction Summary. Go straight to Quality Score Reasoning, an AI-written explanation of the score. Trailhead calls it the most actionable field for knowing what to fix.

Then open the Trace tab. It lists each step in order with timing: subagent routing, actions, retrievals and variable updates. This is how you find out the agent picked the wrong subagent, or an action was slow.

“What did the user actually say?” Use enhanced event logs

Per Help’s enhanced event logs article, event logs capture the events in a session. With enhanced event logs on, you also see the user’s messages, next to agent messages and events.

To turn them on: in Setup Quick Find, select Agentforce Agents, pick your agent, click the edit icon next to Keep a record of conversations with enhanced event logs to review agent behavior, select it and save.

Two things to know. They need Data 360 (Salesforce’s data platform) and Einstein data collection and storage turned on. And they keep data for only seven days and use Data 360 credits. Help recommends them as part of your testing plan.

What do I need to set up first?

Session Tracing. Per Help, it provides the data model Agent Analytics runs on. Before you start, confirm Data 360 is provisioned, Einstein is on and you have an active agent.

1. In Setup, go to Apps, Packaging, Installed Packages, and check the Salesforce Standard Data Model package is version 1.130 or higher.
2. In Quick Find, open Einstein Audit, Analytics, and Monitoring Setup.
3. Turn on Agentforce Session Tracing.

The article lists a few more tasks, like assigning the Platform Integration User the Data Cloud Salesforce Connector permission set. In a sandbox, you also have to activate the Salesforce CRM Connector and its data streams. They copy over from production but stay inactive.

Feedback metrics (thumbs up or down, guardrails) and Knowledge/RAG metrics are optional. RAG means retrieval-augmented generation: the agent pulls in your content to answer, like a Data Library. Help says to turn them on only if you’ll use them, since they add extra data objects.

What happened to Agentforce Analytics?

The Utterance Analysis article says Agentforce Analytics retirement was scheduled to start in May 2026, and it recommends Agent Analytics instead.

Utterance Analysis grouped user requests into clusters for the Agentforce (Default) agent, with reports running on weekends. If you still have it, clone a report before changing it, or you can break the analytics process. The same article says Agentforce (Default) gets no new features after June 17, 2025.

Short version: don’t build anything new on it.

Where do I go next?

Monitoring is one piece of running an agent. The Agentforce guide covers the rest, and the Agentforce tutorial walks through building one. The certification study guides compare every cert.

Which of these tools are you using today? Let me know!

Hope this helps!

Best,

Nick