Considerations for Complaint Management Assistance
Before creating an agent by using the Complaint Management Assistance template, consider supported functionality, usage, limitations and allowances, limits, and other issues.
Required Editions
| Available in: Lightning Experience |
| Available in: Professional, Enterprise, and Unlimited Editions with the Agentforce for Financial Services add-on license or included in Agentforce 1 Financial Services Edition. Requires each user to have the Agentforce for Financial Services add-on to access the action. |
Language and Locale Support
Agentforce Complaint Management Assistance supports English in these locales.
| Language | Locale | Code |
|---|---|---|
| English | English (United States) | en_US |
| English | English (United Kingdom) | en_GB |
Large Language Model (LLM) Support
Agentforce Complaint Management Assistance supports these models. Agent actions can make calls to other predefined LLMs. Bringing your own model isn’t supported, but custom actions that run prompt templates can use any Salesforce-managed model. See Large Language Model Support.
| Model |
|---|
| OpenAI GPT-4o |
Einstein Trust Layer Service Support
Agentforce Complaint Management Assistance supports the Trust Layer services supported by the Agentforce Employee Agent type.
Complaint Management Assistance Guidelines and Considerations
- The subagent and actions are context-specific to public complaint or case records. Initiate the actions from a valid record page, and the agent verifies if the record ID begins with the appropriate prefix before processing. You can't initiate actions from unrelated object pages.
- When retrieving similar historical complaints, the agent uses the Find Similar Complaints action, which calls the findSimilarComplaints invocable action through an autolaunched flow. The action searches historical records and returns complaints that semantically match the current complaint’s details.
- The agent excludes the current, active complaint record from the historical search results to prevent returning the same record as a historical match.
- When drafting customer communications, the agent appends mandatory regulatory disclaimers at the end of the communication. However, users should always review these disclaimers before sending to make sure that they align with the most current regional compliance and audit-ready standards.
- The Update Record action depends on the user clearly defining the target field and the required value. If a user requests a field update without sufficient context, the agent executes the Generate Resolution Note action to derive the value.
- Make sure the user profiles with Agentforce Employee Agent access have read permissions on the public complaint fields.
- Make sure you've enabled Data 360 and uploaded the Data Kit bundle before you use the Find Similar Complaints action. Without a configured data library, the action can’t retrieve historical complaint records.
Prompt Template Considerations
- The Generate Resolution Note action relies on data from related action plans and tasks to determine the progress of a complaint. If your Salesforce instance doesn’t actively use action plans or tasks, or if complaint handlers don’t keep these related records updated, the generated resolution summaries lack detailed progress information and reflect only minimal data.
- The agent never states that a complaint is resolved in its summary until the record’s status is explicitly marked as a final state, for example Resolved, Closed, or Completed. If the status is ambiguous or in progress, the agent lists only the completed and ongoing actions.
- The agent doesn’t forecast future actions. The resolution note avoids language regarding expected or planned steps, focusing only on actions taken or currently in progress.
- When the agent generates a resolution note, it automatically appends the selected cause type and cause subtype to the end of the summary.
Other Considerations
- If the request contains combined utterances from multiple distinct intents, then the agent can’t process a request accurately. For example, the request "Update the complaint status to resolved and draft a customer response" can't process sequentially as intended. For complex requests, break your instructions into individual prompts. For example, first ask the agent to "Update the status", and after confirmation, ask it to "Draft the response”.
- The agent’s search actions can’t run complex aggregation queries. For example, Give me the average resolution time for all debt collection complaints this year.
- The agent input confirmation screens may show the input values entered by a user without their corresponding labels, depending on UI configurations.
- If a custom action or a prompt template doesn’t execute, the agent shows a generic error message with a session ID for troubleshooting.
Did this article solve your issue?
Let us know so we can improve!
