Configure the Customer Knowledge Management Agent

  • 7 minutes

The Customer Knowledge Management Agent turns service data from cases, conversations, emails, and notes into knowledge articles. It compares the source material with existing Dynamics 365 knowledge and creates an article when it identifies a knowledge gap. This capability helps an organization capture useful resolutions without depending entirely on authors to recognize and document every recurring issue.

Automation increases the importance of governance. Source records can contain personal information, incomplete troubleshooting, or context that isn't appropriate for a reusable article. Before enabling creation or publication, define which records the agent can process, which data it can use, who reviews its output, and which audiences can receive generated content.

Prepare the environment

Confirm the following prerequisites before configuring the AI agent:

  • Dynamics 365 knowledge management is configured.
  • The organization uses the standard Case or Conversation tables, or a supported custom Dynamics table for service records.
  • Copilot has access to the internal Dynamics 365 knowledge base.
  • Pay-as-you-go billing is configured for the environment.
  • Connection references for Microsoft Dataverse and Copilot Studio are ready.
  • The required Power Automate flows are enabled and the Copilot Studio agent is published.

To verify the connection prerequisites:

  1. In Copilot Service admin center, go to Support experience > Knowledge.
  2. In Customer Knowledge Management Agent, select Manage.
  3. In Prerequisites, select Set up connections and update the connection references to use the appropriate connectors.
  4. Enable the required Power Automate flows.
  5. After the prerequisite tiles show Ready, publish the Copilot Studio agent.

Resolve connection or flow errors before enabling knowledge creation. Test connections with a service account that follows the organization's ownership and support model rather than relying on an individual configurator's account.

Configure real-time knowledge creation

Real-time creation evaluates a case when a representative resolves it or a conversation when a representative closes it. The AI agent compares the content with existing articles and creates an article when the content addresses a gap.

On the Customer Knowledge Management Agent page:

  • On the Case tab, enable Let Copilot use information from current case to create knowledge articles.
  • On the Conversations tab, enable Let Copilot use information from current chat and messaging conversations to create knowledge articles.

Use rules to limit which records the AI agent processes. For example, include cases for supported products and resolution types while excluding test records, low-quality imports, sensitive queues, or cases that lack a verified resolution.

To create a rule, select Manage rules in the real-time creation section, select New, name the rule and condition, and then define the criteria. Test the rule with records that should qualify and records that should remain excluded.

The AI agent uses mapped attributes to produce each article. Standard mappings include information from the case title, description, product, subject, incident resolution, email content, conversation summary, and notes. In Manage attributes, review the available one-to-one and one-to-many mappings and include only data that improves the generated article.

Attribute selection affects both real-time and historical creation. More source data doesn't automatically produce better knowledge. Exclude fields that contain operational noise, internal-only discussion, duplicated text, or information that requires case-specific context.

Configure historical creation

Historical creation evaluates existing cases, conversations, or supported custom records in bulk. It can identify knowledge gaps in resolved service work that predates the AI agent's configuration.

Before starting a historical run:

  1. Enable change tracking for mapped tables and add the tables to the Entity Analytics Config table as required.
  2. Select the Cases, Conversations, or Custom record types tab.
  3. Enable creation from the selected historical source.
  4. Select Manage rules and define a narrow, testable record set.
  5. Optionally, select a custom article template for the rule.
  6. Save the rule and the Customer Knowledge Management Agent settings.
  7. Select Start, and then use View creation process log to monitor the run.

Important

Historical creation is a long-running process that can't be stopped after it starts. A process can evaluate up to 100,000 cases or conversations. Use controlled batches and validate the output before expanding the scope.

By default, historical creation uses an Issue, Cause, and Resolution article structure. A custom template can provide different content sections, section instructions, and styles. Keep the template focused on article content, describe the expected information for each section, and identify required and optional elements.

Add supported custom record types

Organizations that store service work outside the standard Case and Conversation tables can add supported custom record types as knowledge sources. First enable knowledge management for the record type, and then select Add record type from the Customer Knowledge Management Agent page.

Important

Custom record type support is a production-ready preview feature and is subject to supplemental preview terms. Confirm preview-feature policy and regional availability before including it in a production design.

You can add up to five custom record types. Each table needs fields that provide enough problem and resolution context, appropriate attribute mappings, and rules that identify suitable records. Real-time harvesting from a custom table also requires a supported trigger experience on the table form.

Choose review and publication controls

The AI agent can support different levels of automation. Select the model that matches the risk of the content rather than applying one publication policy to every article.

Available controls include:

  • Case-based draft creation lets representatives propose a knowledge draft from a case during or after resolution.
  • Automatic article updates can create minor versions of existing articles. This capability is a production-ready preview feature.
  • Compliance state assigns a default compliance status to generated articles and updates. An article with a Noncompliant status can't be published.
  • Automatic publishing publishes compliant articles to internal audiences or to internal and external audiences, according to the selected setting.
  • Representative review notifies representatives when a resolved case produces an article for review. This experience requires article-edit permissions, in-app notifications, and Timeline and Timeline highlights on the case.

Note

Customer Knowledge Management Agent removes personal data from case data before drafting an article, but it might not detect every instance of sensitive information. Apply review, compliance, and data-loss controls that match the organization's risk.

Begin with reviewed drafts when source quality or compliance risk is uncertain. Consider automatic publication only after rules, mappings, templates, and generated output consistently meet the organization's standards. External publication warrants additional review because customers can act on the content without representative interpretation.

Validate and monitor the AI agent

Use a controlled set of source records to validate the design. Confirm that the AI agent:

  • Processes eligible records and excludes records outside the configured rules.
  • Uses the intended attributes without exposing case-specific or sensitive information.
  • Recognizes existing knowledge instead of producing unnecessary duplicates.
  • Applies the expected template, compliance state, audience, and publication path.
  • Produces content that an authorized reviewer can correct, approve, and maintain.

Knowledge insights provide information about Customer Knowledge Management Agent output and knowledge usage. Detailed report configuration and interpretation belong to the knowledge discovery module. Use those insights with review feedback to refine rules, mappings, templates, and publication controls.

The next unit explains how versions, categories, and translations keep knowledge organized and current after articles enter the managed lifecycle.