When a teammate finds an AI-generated message that is incorrect, incomplete, confusing, or not aligned with the expected behavior, they can report it directly from Inbox. The report is then sent to the Optimization feature under SleekFlow AI, where users with SleekFlow AI access can review the issue, understand what went wrong, test suggested changes, and update the related Playbook.

Optimization is useful for teams that want to continuously improve AI agent quality while keeping review and deployment controlled.

Before you begin

Before using Optimization, note the following:

• You will need the required access to AgentFlow
• Only AI-generated messages can currently be reported from Inbox.
• Reported messages are sent to Optimization for review.
• Changes deployed from Optimization are applied immediately to the live AI agent.
• There is currently no undo option after deploying an update.

When to use the Optimization feature

Use Optimization when your team needs a structured way to review and improve AI-generated responses.

Common use cases include:

• An AI agent misses an important SOP step from a Playbook.
• An AI agent answers when it should have escalated or followed another workflow.

Optimization helps your team:

• Report problematic AI-generated messages directly from Inbox.
• Review reported issues in one central place.
• Group similar or repeated reports into a single suggestion.
• Inspect conversation context and execution trace before deciding what to change.
• Test proposed changes before applying them.
• Apply Playbook updates from Optimization.

Suggest an adjustment from Inbox

When you find an AI-generated message that needs improvement, you can suggest an adjustment directly from the Inbox conversation.

This lets teammates share what they think the AI agent should change, while keeping admins or AI managers in control of reviewing and applying updates in Optimization.

Suggest an adjustment

You can follow the steps below to suggest an adjustment for an AI-generated message:

1. Click on the icon on the left navigation bar to go to Inbox
2. Open the conversation that contains the AI-generated message.
3. Find the AI-generated message you want to adjust.
4. Click “Adjust” on the message.
   

A side window will open. This window shows:

• The original AI-generated message
• A free-text comment box
• A submit action
• An option to view the message’s execution trace

Use the comment box to explain what should be improved. You can describe what was incorrect, incomplete, unclear, or not aligned with the expected behavior.

After the suggestion is submitted, it is sent to “Optimization”  for review. An admin or AI manager can then review the suggestion, check the context, test the proposed change, and decide whether to apply it.

Tip: Write a clear suggestion suggestion
 When suggesting an adjustment, include:

• What should be improved in the AI-generated message
• What the expected answer or behavior should be, if known
• Any relevant customer, product, channel, or workflow context

Review the execution trace before submitting an adjustment

If you are not sure why the AI agent generated a specific response, you can review the message’s execution trace before submitting your adjustment.

From the adjustment side window, click “Advanced: View execution trace”.

The side window will switch from the adjustment form to the execution trace view. You can review how the AI agent processed the customer’s message, then go back to the adjustment form to continue writing your feedback.

Tip: Review the trace if you are unsure what went wrong
 The execution trace can help you understand what information the AI agent used before you submit your suggestion. To learn more, refer to View AI message’s execution trace in Inbox.

Submit the suggestion

After entering your feedback, submit the suggest.

Once submitted:

• A success message appears.
• The report is sent to the “Optimization” section under the AgentFlow page
• Users with SleekFlow AI access can review the report in the “Optimization” page

Review reported issues in the “Optimization” page

Optimization is the central workspace for reviewing reported AI-generated messages, grouped issue patterns, and suggested improvements.

Accessing the “Optimization” page

You can follow the steps below to open the “Optimization” page:

1. Click on the icon on the left navigation bar to go to the SleekFlow AI page
2. Click on the “Optimization” tab in the top menu
   

Understanding the “Optimization” page

The page is organized into three main areas:

Issues to review

The left panel shows the list of suggestions waiting for review.

Each item represents an issue that may need an update to the AI agent’s Playbook. Similar issues may be grouped into the same suggestion, so your team can review repeated patterns together instead of handling each report separately.

The list shows a short issue title, the related AI agent, and when the issue was created or last updated.

Select an issue from this list to review its details.

Issue details

The middle panel shows the context behind the selected issue.

This section helps you understand what was reported before deciding whether to apply the suggested update. It may include:

• Likely cause: A summary of what Optimization identified as the possible reason for the issue.
• Reported instances: One or more reports related to the same issue.
• Agent note: Feedback submitted by the teammate from Inbox.
• Conversation example: The conversation related to the reported AI-generated message.
• Conversation and Logic trace views: Toggle between the customer conversation and the AI agent’s logic trace.

Use this section to verify whether the suggestion matches the actual conversation context.

Suggested fix

The right panel shows the recommended update for the selected issue.

Depending on the issue, this section may include:

• A summary of the suggested update
• The Playbook area that may be updated
• An explanation of how the update addresses the reported issue
• An action to review the suggested update, such as Review playbook update
• Test results, such as the number of scenarios passed or requiring review
• Actions to Review failed scenarios, Apply update, or Discard suggestion

Use this section to review, test, apply, or discard the suggested update.

Note: Suggestions require human review
 Optimization can suggest updates based on submitted suggestions, but the update is not applied automatically. Review the issue details, check the suggested fix, and test the change before applying it to the live AI agent.

Review the suggested update

If the suggested fix requires a Playbook change, click “Review playbook update” to open the detailed update view.

This view lets you review exactly what Optimization is suggesting before you test or apply the change.

The suggested update page includes:

• The Playbook section that will be updated
• The full surrounding Playbook content for context
• The number of added and removed lines
• Highlighted changes
• An “Use AI to edit suggestion” panel

Added content is highlighted in green. Removed content is highlighted in red with strikethrough formatting.

Use this view to compare the original Playbook instruction with the suggested update. This helps you confirm whether the change addresses the reported issue without changing unrelated instructions.

In the “Use AI to edit suggestion” panel, you can ask AIto refine the suggested update before applying it. For example, you can ask AI to make the instruction clearer, adjust the structure, or suggest another version of the fix.

Once you have completed, click “Close” at the top right corner to close this panel.

Review and manage the “Suggested fix” panel

The “Suggested fix” panel shows the recommended update for the selected issue. It explains what Optimization suggests changing and how the update may address the reported issue.

Depending on the test result and suggestion status, you can take the following actions from this panel:

Review failed scenarios

If the impact test shows that one or more scenarios need review, click “Review failed scenarios”.

Use this to review scenarios that did not pass or may need human judgment before you apply the update. After reviewing the failed scenarios, you can decide whether to adjust the suggestion, test again, apply the update, or discard it.

Apply the suggested update

Click “Apply changes” or “Apply anyway” to apply the suggested update.

Use “Apply anyway” only when you have reviewed the flagged scenarios and are comfortable deploying the update despite the warning.

Once applied, the related Playbook update is deployed to the live AI agent.

Discard the suggestion

Click “Discard” when you do not want to apply the suggested update.

Use this when the suggestion is not needed, does not match the issue, or does not require a  Playbook change.

Once discarded, the suggestion is removed from the open review list. No changes are applied to the AI agent.

Handle suggestions without a suggested fix

Some suggestions may not include a suggested Playbook update.

This can happen when Optimization needs more context, or when the issue requires human judgment before a fix can be suggested.

If no suggested fix is shown, review the issue details and conversation context. You can then decide whether to:

• Use AI to explore a possible fix
• Update the related  Playbook manually
• Discard the suggestion if no action is needed

Best practices

• Write clear adjustment suggestions
  When suggesting an adjustment from Inbox, explain what should be improved and what the expected answer or behavior should be. Add customer, product, channel, or workflow context when relevant.
• Review grouped suggestions carefully
  Similar reports may be grouped into one suggestion. Review the reported instances to confirm that they share the same likely cause.
• Use the logic trace when the cause is unclear
  If you are not sure why the AI agent responded in a certain way, review the **Logic trace** to understand the intent, referenced knowledge, actions, and integrations involved.
• Review failed scenarios before applying changes
  If the test result shows scenarios that need review, check them before applying the update. Use “Apply anyway”only when you understand the impact and are comfortable deploying the change.
• Review deployment details carefully
  Updates are applied immediately to the live AI agent and cannot currently be undone. Review the affected agent, test result, and change summary before deployment.