Monitor and Improve Agent Performance

Learn how to use analytics, evaluation, and feedback tools to continuously improve your agents

Deploying an agent is just the beginning. AGO’s Dashboard, Thread Evaluator, and feedback tools help you spot quality issues and fix them before they affect more users.

Prerequisites

• At least one agent deployed and handling conversations
• Admin or Account Manager access

1. Check the Performance Dashboard

   The Dashboard gives you a high-level view of how your agents are doing.

   1. Navigate to Dashboard in the sidebar — the Overview tab opens by default
   2. Select a time range (e.g., Last 7 days) from the header filters
   3. Review the four key metrics at the top:

   Metric	What to Watch For
   Number of Threads	Is conversation volume growing or declining?
   Escalation Rate	Rising rates mean the agent can’t handle some questions
   Positive Evaluation Rate	Dropping rates signal quality issues
   Number of Users	Are new users engaging with the agent?

   4. Compare to the previous period using the +/- percentages. A sudden change is worth investigating.

   See Performance Dashboard for all available metrics and tabs.

2. Browse Conversations

   Numbers tell you what is happening. Reading conversations tells you why.

   1. Navigate to Chat History in the sidebar
   2. Click More filters and apply filters to focus your review:
      • CX Score Range = Poor or Needs Improvement — find the worst conversations
      • Support Asked = Yes — see what the agent couldn’t handle
      • Needs Review = Yes — conversations flagged by the LLM evaluator
   3. Read through 5-10 conversations to understand patterns

   Look for patterns, not individual failures. If multiple conversations fail on the same topic, that’s a knowledge gap. If the agent misunderstands instructions, that’s a prompt issue.

   See Conversations for all filtering options.

3. Enable the Thread Evaluator

   The Thread Evaluator automatically scores every conversation across five quality dimensions.

   1. Navigate to Settings > Thread Evaluation > Thread Evaluator
   2. Toggle the evaluator on
   3. Review the five scoring dimensions and their default weights:

   Dimension	Default Weight	What It Measures
   Resolution Quality	25%	Did the agent address the request?
   Response Accuracy	20%	Were responses consistent and contradiction-free?
   Tone & Empathy	20%	Did the tone match your agent’s guidelines?
   Efficiency	20%	Was the conversation handled without unnecessary back-and-forth?
   Customer Effort	15%	How easy was it for the user to get their answer?

   4. Optionally add custom instructions to align scoring with your quality standards
   5. Click Save

   Once enabled, conversations are scored daily. Conversations below the minimum thresholds are flagged for review.

   See Thread Evaluator for configuration details.

4. Analyze CX Scores

   After the evaluator has run for a day or two, check the Evaluation tab.

   1. Navigate to Dashboard and select the Evaluation tab
   2. Review the CX Score gauge — this is your overall quality score (0-100)
   3. Check dimension averages — which dimension is pulling the score down?
   4. Review the CX Score distribution — what share of conversations fall into Poor (0-49), Needs Improvement (50-79), and Good (80-100)?
   5. If you have multiple agents, check CX Score by Agent to compare performance

   CX Score Range	Interpretation
   80-100	Good — minor refinements only
   50-79	Needs improvement — investigate the lowest dimensions
   0-49	Poor — review conversations in this range to find root causes

   See CX Score for how the score is calculated.

5. Review User Feedback

   Direct feedback from users gives you specific signals that automated scoring may miss.

   1. Navigate to Dashboard and select the Feedback tab
   2. Review the issue type distribution — which category dominates?
   3. Click through to individual conversations to understand the context
   4. Navigate to Feedbacks in the sidebar for the full list with filtering and reviewer assignment

   Common patterns and fixes:

   Feedback Type	Likely Cause	Fix
   Inaccurate	Outdated or conflicting content	Update the knowledge base
   Incomplete	Documents missing detail	Expand existing docs or add new ones
   Information Not Found	Topic not covered	Add new documentation
   Technical Issue	System error	Check integration configurations

   See Feedback Analytics and Feedback List.

6. Set Up Topic Analysis

   Topic analysis helps you categorize conversations automatically and spot trending subjects.

   1. Navigate to Settings > Conversation Labels and create your topics (see Conversation Labels)
   2. Navigate to Settings > Topic Analyzer and enable it (see Topic Analyzer)
   3. After conversations are analyzed, check the Topics tab on the Dashboard (see Topics Analytics)

   The Topics tab shows a treemap of topic distribution, daily trends, and per-agent breakdowns — including CX Score and escalation rate per topic.

7. Create an Improvement Loop

   Use what you’ve learned to fix issues systematically:

   1. Identify — use the Dashboard to find the biggest issue (low CX dimension, high escalation, common feedback type)
   2. Diagnose — read conversations in Chat History to understand the root cause
   3. Fix — update the knowledge base, agent instructions, or tool configuration
   4. Test — run Simulation Testing to validate the fix
   5. Deploy — push the change to production
   6. Monitor — watch the Dashboard to confirm the metric improves

   Repeat this cycle weekly.

Checklist

• Performance Dashboard reviewed
• Low-scoring conversations investigated in Chat History
• Thread Evaluator configured and enabled
• CX Score dimensions analyzed in the Evaluation tab
• User feedback reviewed
• Topics configured and Topic Analyzer enabled
• First improvement cycle completed

What’s Next

• Performance Dashboard - All dashboard tabs and metrics
• Thread Evaluator - Evaluation configuration
• CX Score - Scoring methodology and interpretation
• Simulation Testing - Validate changes before deploying
• Agent Best Practices - Design patterns for better agents