Learn how to strategize the implementation of your Service Assistant subagents for optimal granularity and knowledge grounding.
Required Editions
| View supported editions. |
The example subagent in Grounding Service Assistant with Subagents demonstrates our recommended approach, format, and structure for documenting your business processes into instructions. However, you don’t need to create detailed subagents and instructions for every issue type and its subtype.
- Service Assistant is only supported in the legacy Agentforce Builder. Starting in July 2026, the requirement that new agents can only be created in the new Agentforce Builder doesn’t apply to Service Assistant. You can continue to create and configure a Service Assistant agent in the legacy builder until further notice. See Create Agents in the New Builder Only Starting July 2026.
- Beginning in April 2026, agent topics are now called subagents. There are no changes to functionality. During this transition, you may see a mix of the new and previous terms in our documentation. See Topics Are Now Subagents.
Subagent Granularity
When building your subagents, the detail you provide in your instructions depends on your use cases and if your company intends to use knowledge grounding through Agentforce Data Libraries.
If you aren’t using knowledge grounding or don’t have articles for a specific issue, you must create detailed subagents that specify the processes to resolve the issue. See the example in Grounding Service Assistant with Subagents. You must make sure Service Assistant has all the necessary information to guide the service rep through the resolution process.
Because you must write these subagents manually, we recommend using Service Assistant for your priority issue types. To cover variations of issues efficiently, you can document the variations in a single subagent using conditional language in your instructions, like “If ..., then ...”, “When..., then…”, and “Once you have…”.
When using knowledge grounding, you don't need a granular subagent for every issue type and subtype. Instead, focus on creating subagents centered on high-level issue types, such as “billing dispute”, “transaction declined”, or “payroll change” and use the subagent description to encompass any subtypes. This allows you to use one subagent for a wide range of case types and knowledge articles that cover different processes of the high-level subagent. Here’s an example.
| SUBAGENT PART | EXAMPLE |
|---|---|
| Label | Transaction Decline |
| Description | Assist service reps in helping customers resolve declined credit card transactions. Questions are related to authorization failures, including declining reason codes such as insufficient funds, suspected fraud, incorrect card details, travel blocks, and daily spending limits. |
| Scope | Your job is to assist service reps in identifying the reasons behind declined transactions and providing the necessary steps to either allow the charge or secure the customer's account. You must not handle inquiries outside of declined transactions and account security. |
When you have a high-level subagent that can encompass different variations of an issue, Service Assistant only uses the Subject and Description fields set in your Service AI Grounding configuration to categor the issue and match the case to that subagent.
Once the subagent is matched, Service Assistant then references all fields and objects set in your Service AI Grounding configuration to search for relevant knowledge articles and create plan steps directly from those articles. While your organization may use other fields like picklists to categorize cases into types and subtypes, Service AI Grounding only supports text and string fields. Any text or string fields not set in your Service AI Grounding configuration aren't referenced by Service Assistant when creating a plan.
If you ground on case related objects, Service Assistant only reviews the text in case comments, emails, or the feed. PDFs, images, and other attachments aren't used as grounding sources and don't play a role in creating a service plan or finding related articles.
Structuring Subagents with Knowledge Grounding
When using knowledge grounding, create subagents that have 3 to 5 instructions that outline the general resolution process. Because you have already defined your case subtypes in the subagent's description, you do not need to build multiple, granular instructions for every possible scenario.
Instead, instructions serve as a general framework that describes the overall workflow required for any case in that category. In the “Transaction Decline” example below, the instructions cover standard procedures like authenticating the customer (Instruction 1) and identifying the system response code (Instruction 3).
Instead of writing out the exact resolution steps for every possible subtype, Instruction 4 provides a general framework to guide the agent. By stating “execute the corresponding standard procedure for removing blocks, clearing flags, or advising the customer.”, you tell Service Assistant what the ultimate goal is without detailing exactly how to achieve it. Service Assistant uses knowledge grounding to automatically pull granular procedural information from your articles.
| SUBAGENT PART | EXAMPLE |
|---|---|
| Label | Transaction Decline |
| Description | Assist service reps in helping customers resolve declined credit card transactions. Questions are related to authorization failures, including declining reason codes such as insufficient funds, suspected fraud, incorrect card details, travel blocks, and daily spending limits. |
| Scope | Your job is to assist service reps in identifying the reasons behind declined transactions and providing the necessary steps to either allow the charge or secure the customer's account. You must not handle inquiries outside of declined transactions and account security. |
| Instruction 1 | Make sure the customer has provided the required authentication information: Full Name, Last 4 digits of the card, and Answer to Security Question. Ask the customer to provide the specific details of the declined attempt, including the merchant name, transaction amount, and date of attempt. |
| Instruction 2 | Check the “Available Credit” and “Account Balance” against the transaction amount. Confirm the account is not in arrears or over the credit limit. Verify if the “Card Lock” or “Freeze” feature is currently active in the customer's profile settings. |
| Instruction 3 | Search the Authorization Log in the system using the card details. Locate the specific decline entry to identify the system Response Code. |
| Instruction 4 | Based on the identified response code (such as NSF, Suspected Fraud, International Restriction, Invalid CVV, or Card Status), execute the corresponding standard procedure to remove temporary blocks, clear fraud flags, or advise the customer on the required next steps to allow the transaction. |
| Instruction 5 | After resolving the decline, send the customer a “Transaction Status Update” notification via email confirming that the block has been lifted or outlining the next steps required. |
Additional Guidelines
When using knowledge grounding, Service Assistant automatically retrieves relevant articles to resolve customer inquiries based on the case details and assigned subagent. You don't need to configure this behavior by adding the Answer questions with knowledge action to subagents or creating instructions that state to search the knowledge base.
Here are general guidelines to follow when using knowledge grounding with subagents. Find more details in Knowledge with Subagents in Service Assistant.
| Guidelines | Details |
|---|---|
| Don’t use the Answer Questions with Knowledge action. | Service Assistant doesn’t use the action to retrieve knowledge. Adding the actions to a subagent has no effect. Knowledge retrieval is handled automatically by the Agentforce Data Libraries. |
| Don’t tell the agent to review the knowledge base. |
Don’t write instructions that tell Service Assistant to review your knowledge base. Knowledge retrieval and grounding happens automatically through Agentforce Data Libraries. If referencing a specific knowledge article is a mandatory step in the resolution workflow, you can write an instruction that explicitly states to review the specific article. |
| Expect summarized knowledge steps. |
Service Assistant summarizes the information from your knowledge articles to generate a service plan. This includes the step-by-step procedural information contained in the article. As a result, the exact procedural details found in your knowledge articles might not always be included in the service plan. Service reps may have to look into the article for specific guidance using the articles cited in the Sources section of the component. |