> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/capabilities/agent-insight.md).

# Agent Insight

Agent Insight scores every agent - your people and the AI - on the metrics you choose, with comments, recommendations and trends, by role.

*Last updated: November 6, 2026*

Agent Insight scores how each agent handles real conversations - your human agents and the AI agent alike - on the metrics you choose. Every score comes with a comment and recommendations, so you know not just how someone is doing but what to change.

### Why it matters

* **Quality review at full coverage.** Instead of sampling a few tickets a week, every conversation can be scored.
* **Fair, role-based scoring.** A billing agent is judged on billing, a sales rep on sales.
* **Coaching that is specific.** Each score explains what went well (merit) and what could be better (opportunity).
* **One standard for people and AI.** Compare both on the same measurements.

### How it works

Agent Insight is built on **evaluation sets** that describe a role:

* **The role** - whose work it scores, such as "agents on the billing team, handling invoices, refunds and payment failures".
* **Measurements** - what each conversation is scored on, with guidance and weights.
* **Measurement Instruction (Optional)** - how strictly to score.

You can have several sets. For each agent, the AI applies the set that best matches the agent's profile; a default set covers anyone who matches none.

Conversations are scored in two ways, set in **Agent Insight → Config** under **How Agents Insight Runs**:

* **Automatically** - turn the switch on to score conversations as they come in, against the best-matching set.
* **Manually, with an in-note command** - paste the command shown on the page, `// <command id>`, into the internal note of a conversation to score that one. This works whether or not the automatic run is on. See [In-note Commands](https://doc.aissist.io/capabilities/in-note-command.md).

Scores are rolled up per agent, per metric, per day, with comments and recommendations.

### Score against your own documents

Add your QA guidelines, policies and procedures as [Insight Assets](https://doc.aissist.io/capabilities/insight-assets.md) in **Insight → Insight Assets**. For every conversation it scores, Agent Insight looks up what the measurements need to know that the conversation doesn't say, such as the refund window or the steps of a procedure, and scores against the passages it finds. That way an answer is judged on whether it was right, not only on how it read.

Every evaluation set uses every Insight Asset that is on; there is nothing to set per set. Insight Assets are never used to answer customers.

### The dashboard

Open **Agent Insight → Dashboard**:

* **Overview** - the **Overall Score** with a **Trend** or **Histogram**, filtered by **Evaluation Set**, and **Performance By Agent**, by metric or by date.
* **Agent View** - one agent's **Score Card**, **Overall Score Trend**, recommendations and **Metric Breakdown By Day**.
* **Metric View** - one metric over time, compared across agents.
* **Explore** - every scored conversation, searchable by date or conversation / ticket ID. Open one to read each agent's merit and opportunity and the **References** the score used, or **Re-evaluate** it after changing a measurement.

**Reports** on the Config page lists the generated agent insight reports.

Agent Insight is not available on the free plan.

### Set it up

1. Go to **Agent Insight → Config** and click **Add**.
2. Describe the role and what it should be judged on. Ask Engine drafts the set from conversations that actually happened; review it and create it.
3. Optional: add your QA guidelines and policies in **Insight → Insight Assets**, so answers are checked for accuracy.
4. Turn on the automatic run, or use the in-note command on chosen conversations.
5. Check the dashboard after a few days of traffic.

### With Ask Engine

[Ask Engine](https://doc.aissist.io/capabilities/ask.md) reads the agent insight reports and individual results of your project:

> Look into agent insights. What went well and what can be improved, without naming specific agents?

> Which metric dropped most for the billing team this month, and why?

To add a role-based set, click **Add** on **Agent Insight → Config** and describe it in one sentence. Ask Engine drafts the set for you to review before it is created:

> Score the returns team on whether they check the order before promising a refund.

### Best practice

* Start with one set per team, with three to five measurements that matter most.
* Write measurements as observable behavior: "confirms the order before promising anything", not "is helpful".
* Use the comments in coaching sessions, not just the scores.
* Re-evaluate a few conversations after changing a measurement, to check it scores the way you expect.
* When a measurement depends on a policy ("only offers refunds the policy allows"), add that policy as an Insight Asset.

### Related

* [Insight Assets](https://doc.aissist.io/capabilities/insight-assets.md)
* [Session Insight](https://doc.aissist.io/capabilities/session-insight.md)
* [Evaluation](https://doc.aissist.io/capabilities/evaluation.md)
* [Weekly Reports and Alerts](https://doc.aissist.io/capabilities/weekly-reports.md)
* [In-note Commands](https://doc.aissist.io/capabilities/in-note-command.md)
