How to Create and Manage Knowledge Base from Conversations Using AI Agent
Knowledge from Conversations helps organizations transform valuable support interactions into knowledge base content by identifying recurring customer issues from resolved tickets. The feature analyzes validated conversations, detects common patterns, and generates AI-powered article suggestions to help teams build and maintain an effective knowledge base.
After a ticket is closed, eligible conversations are analyzed and grouped based on similar issues. As recurrence is tracked over time, the system automatically generates a Review Suggestion when the configured occurrence threshold is reached, enabling teams to document commonly reported issues more efficiently.
The feature includes the following sections:
- Learning Rules: Choose which tickets resolved by human agents the AI Agent can learn from and set how many suggestions it creates.
- Awaiting Recurrence: Questions that haven’t repeated often enough yet.
- Review Suggestions: Approve, edit, merge, or reject articles the AI Agent wrote from resolved tickets.
How it Works
The Knowledge from Conversations workflow follows these steps:
- A ticket is closed.
- The system waits 48 hours after the ticket is closed.
- Basic ticket validation is performed.
- Learning rules validation is performed.
- Eligible conversations are summarized and grouped under Awaiting Recurrence.
- Recurrence is tracked for similar issues.
- When the configured recurrence threshold is reached, a Review Suggestion is generated.
- The Review Suggestions become available for review.
Closed Tickets (48 hours) │ ▼ Basic Ticket Validation │ ▼ Learning Rule Validation │ ▼ Awaiting Recurrence (AI Summary + Recurrence Tracking) │ ▼ Recurrence Threshold Reached │ ▼ Review Suggestions Generated
Ticket Validation
Before a conversation is considered for knowledge creation, the system validates the associated ticket.
Validation may include checks such as:
- The ticket is in the Closed status.
- The ticket has been closed for at least 48 hours.
- The ticket contains sufficient conversation content for analysis.
- The ticket meets internal processing requirements.
Tickets that do not meet the validation criteria are excluded from further processing.
Learning Rules
The Learning Rules section allows administrators to define which resolved conversations are eligible for article generation. Let the AI agent learn from tickets your human agents resolve. Tickets closed before you turn this on aren’t used.
Use this section to:
- Select the ticket conditions that can contribute to knowledge creation.
- Configure article generation settings.
- Control which conversations are analyzed for recurring issues.
- Enable or disable individual Learning Rules.
Only tickets from active and configured Learning Rules are processed for article suggestions.
Configure Learning Rules
- Navigate to AI Center > Knowledge from Conversations > Learning Rules.
- Toggle on Learn from resolved conversations.
- Click Add Rule.
- Configure the source criteria by defining which tickets are used to generate knowledge for AI Agents. The following have to be defined here;
Field Description Name Specifies a unique name for the ticket source. This name helps identify the source configuration used to select tickets for AI knowledge generation. Ticket Conditions Defines the criteria that tickets must meet to be included as a knowledge source. Only tickets matching all configured conditions are considered for knowledge extraction. Quality Filters Keeps abandoned and unresolved threads from being used in agent responses. The following quality filter is available:
• Resolved by a human agent, not automatically closed - Includes only conversations that were manually resolved by an agent. Conversations closed automatically due to SLA timeouts, inactivity, or no response are excluded from knowledge generation. This option is selected by default and cannot be changed. - Click on Add Rule.
Explore the GIF below for more insights.
- If you turn off learning from resolved conversations, new conversations will no longer be used to generate suggestions. Existing suggestions and knowledge articles will remain available and will not be removed.
- You may add up to 15 rules for your AI Agent to learn.
Managing Learning Rules
You can perform the following actions on Learn from resolved conversations page:
| Action | Description |
|---|---|
| Enable or Disable a Rule | Use the toggle switch in the Status column to control whether a learning rule actively participates in knowledge generation. |
| Edit an Existing Rule | Select the Edit (✏️) icon to modify the rule name, source criteria, or other rule settings. |
| Delete a Rule | Select the Delete (🗑️) icon to permanently remove a learning rule that is no longer required. |
| Deactivate a Rule | Turn off the rule using the Status toggle to temporarily pause knowledge generation from that source without deleting the rule. |
| Filter by Status | Use the Status filter dropdown to view rules based on their current state: All Statuses, Active, or Paused. |
Only conversations that match an active Learning Rule are considered for knowledge generation.
Suggestion Controls
Suggestion Controls allow administrators to control how review suggestions are generated and managed. To configure suggestion controls, follow the steps below:
-
Navigate to AI Center > Knowledge from Conversations > Learning Rules.
-
Toggle on Learn from resolved conversations.
-
Click Suggestion Controls.
-
Configure the required settings.
-
Click Save. Please check the GIF file below for more insights.
Setting Description Suggestions Generated per Day Specifies the maximum number of article suggestions that can be generated each day. Use this setting to control the volume of AI-generated article suggestions. A maximum of 25 suggestions and a minimum of 1 suggestion can be generated per day. Pause Generation at Pending Count Specifies the maximum number of pending suggestions allowed before the system temporarily stops generating new suggestions. Generation resumes when the pending count falls below the configured threshold. A maximum of 50 and a minimum of 20 pending suggestions can be set. Auto-Reject Unreviewed Suggestions After Specifies the number of days a Review Suggestion can remain unreviewed before it is automatically rejected. This helps keep the Review Suggestions list current by removing stale suggestions. A maximum of 60 days and a minimum of 30 days can be set. Minimum Occurrences Before Review Specifies the minimum number of times a similar issue must occur before it is promoted from Awaiting Recurrence to Review Suggestions. When the configured occurrence count is reached, a Review Suggestion is automatically generated for review. A minimum of 4 occurrences and a maximum of 10 occurrences can be set.
Example
The following configuration demonstrates how generation limits work:
- Suggestions generated per day: 25
- Pause generation at pending count: 50
- Auto-reject unreviewed suggestions after: 30 days
- Minimum occurrences before review: 4 occurrences
With this configuration:
- A maximum of 25 article suggestions can be generated each day.
- Article generation pauses when 50 suggestions are awaiting review.
- Suggestions that remain unreviewed for 30 days are automatically rejected.
- An issue must occur at least 4 times before a Review Suggestion is generated.
- Changes to Suggestion Controls affect future article generation and do not modify existing Review Suggestions or recurrence counts.
- Generation may be temporarily paused when the pending suggestion count reaches the configured limit.
Awaiting Recurrence
Awaiting Recurrence is a tracking stage for validated issues identified from resolved conversations that have not yet reached the configured recurrence threshold required to generate a Review Suggestion knowledge base article.
The system uses AI to summarize the issue and resolution, groups similar conversations, and tracks how often the issue occurs. Once the required occurrence count is reached, the issue is automatically promoted to Review Suggestions. Administrators can also search, manually promote, or delete tracked issues, while issues that do not reach the threshold before their expiration date are automatically removed from tracking. Explore How to Manage Awaiting Recurrence from Conversations.
Review Suggestions
Review Suggestions are AI-generated knowledge base article recommendations created from recurring customer issues identified in ticket conversations. Once the configured recurrence threshold is met, BoldDesk generates a Review Suggestion, allowing agents to validate, edit, merge, convert it into a draft, or reject it before publication. Explore How to Manage Review Suggestions from Conversations.
Troubleshooting
Closed ticket does not appear in Awaiting Recurrence
Common causes:
- Ticket did not pass basic ticket validation.
- Ticket does not match any active Learning Rule.
- Ticket was closed less than 48 hours ago.
Review Suggestions are not being generated
Common causes:
- Recurrence has not reached Minimum occurrences before review.
- Generation is paused because Pause Generation at Pending Count has been reached — check and remove.
- Daily generation has reached Suggestions Generated per Day.
- Suggestions may be removed from the pending list due to Auto-Reject Unreviewed Suggestions After, affecting what reviewers see over time.
Frequently Asked Questions
-
Why is a closed ticket not appearing in Awaiting Recurrence?
The ticket may have failed basic ticket validation, may not match an active configured Learning Rules, or may have been closed for less than 48 hours. -
When does Knowledge from Conversations generate a Review Suggestion?
Knowledge from Conversations generates a Review Suggestion when a grouped issue’s recurrence count reaches the configured Minimum occurrences before review value. -
What happens when “Pause generation at pending count” is reached?
The system pauses new Review Suggestions generation automatically and resumes generation when the pending suggestion count falls below the configured threshold. -
What happens to Review Suggestions that are never reviewed?
Review Suggestions that exceed the configured Auto-reject unreviewed suggestions after timeframe are automatically rejected. -
Can an agent edit a Review Suggestion before publishing?
Yes. Agents can review, edit, merge, convert the suggestion into a draft article, or reject the Review Suggestions before publication. -
Do Generation Limits changes affect existing recurrence tracking or existing suggestions?
No. Generation Limits changes apply to future generation and do not modify existing Review Suggestions or existing recurrence counts. -
Can Knowledge from Conversations exclude auto-closed tickets?
Yes. The Resolved by an agent, not auto-closed quality filter is enabled by default and cannot be unchecked, so auto-closed tickets are excluded from knowledge generation. -
What happens if a ticket is reopened after contributing to knowledge?
If Remove from knowledge once a ticket is reopened is enabled, knowledge extracted from that ticket is removed when the ticket is reopened and is added back only after the ticket is resolved again.