AI agent reports | Hams AI Docs

The AI agent report shows what your AI agents did in a period, and how the conversations ended. For each conversation cycle, with or without the AI agent, see the conversation log.

The AI agent is off by default. Hams support turns on the Workflows + Agent feature for your account. Administrators and users with the Manage reports permission can open the report.

Open the report

Open Reports

Click Reports in the sidebar.

Open the tab

Click the AI Agent tab. The report opens at /app/accounts/{account_id}/reports/ai-agent.

Set the filters

Select a date range. To see one AI agent, select it in the AI Agents filter.

Summary cards

Card	What it shows
Conversations handled	The conversations in which the AI agent had at least 1 run. Test messages in the playground do not count.
Messages sent	The messages that the AI agent sent to customers.
Resolutions	The conversations that the AI agent resolved with its Resolve Conversation tool.
Handoffs	The conversations that the AI agent transferred to a human agent. A handoff can come from the Handoff to Human tool, from no AI credit, or from a closed reply window.
Resolution rate	The percentage of the conversations that the AI agent handled that it resolved on its own.
Handoff rate	The percentage of the conversations that the AI agent handled that it handed off to a human agent.
Tool calls	The action tools that the AI agent called, for example tags, notes, assignment, and HTTP tools. The report does not count system tools and knowledge searches.
Avg turns / conversation	The average number of model turns that the AI agent took in each conversation.
Knowledge lookups	How many times the AI agent searched its knowledge.
Knowledge availability	The percentage of knowledge searches that found relevant content.

The conversation, outcome, and tool values count only the activity after the start of AI agent telemetry. The report shows a note about this.

Charts

Activity over time: the Conversations handled, the Resolutions, and the Handoffs of each day.
Tool usage: each action tool, with its Calls and its Success rate.
Outcome breakdown: the conversations that the AI agent resolved, handed off, or hit an error in: Resolved, Handed off, or Error. A conversation can appear under more than one outcome.

Export the report

Click Export. Then click Export CSV. The system downloads hams_agent_report.csv. The file has the period, the summary values, and the tool usage table.

CSAT by AI agent

The AI agent collects ratings with its Collect CSAT tool. These ratings show in Reports > CSAT with the other ratings. The AI agent shows as the assigned agent.

To see the ratings of one AI agent, select it in the AI Agents filter. You can compare the ratings of an AI agent with the ratings of your human agents. In the CSV file of the CSAT report, the agent column shows the name of the AI agent.

The CSAT report with the AI Agents filter: the summary cards and the ratings in the period.

Find the cause of a bad answer

Open the reply

Open the conversation. Under the reply of the AI agent, click Generated by AI. The panel How was this reply generated? opens.

Read the steps

Read Generation steps, Knowledge used, and Reasoning. They show which tools ran and which documents and FAQs the AI agent used.

Read the tool results

To see what each tool returned, open the conversation log. Click the row of the cycle. Then click the Breakdown tab.

Fix the knowledge

If the correct answer is not in the results, add it to the knowledge of the AI agent. You can also add it to the FAQ.

Fix the instructions

The correct answer can be in the results, but the AI agent said something different. In that case, change the instructions of the AI agent.

Fix the tool

A tool can run at the wrong time, or not run. In that case, change the When to use text of the tool.

Only administrators see Generated by AI. See Assistants.

The Used in conversations column on the Documents and FAQ tabs shows how often the AI agent uses each source and each FAQ. A source with a count of 0 can be a sign of a problem in the content.