Grounding with Voice Call Record Data

Record Processing Initiation

Record monitoring begins after the service rep accepts the call and the call meets your eligibility criteria. Service Assistant then reviews the transcript to identify the customer's intent and issue and assign a subagent. There's no fixed threshold for how much transcript is needed for Service Assistant to identify the issue and start the plan. Plan initiation depends on the telephony model, transcript processing time, and clarity of the transcript. After Service Assistant identifies the issue, it assigns a subagent and immediately presents the first step. Unlike case and messaging processing, voice processing doesn't show a plan summary or a Start Plan button.

Along with the context requirements, meet these requirements before record processing starts.

• Salesforce admins and service reps require direct access to your voice call eligibility criteria flow. Because the flow doesn't run on the ServicePlanner User's Run Flows app permission, grant access through a custom permission set. See Eligibility Criteria and Set Up Service Assistant for Voice.
• The ServicePlanner User requires app and object permissions for the Voice Call object. See Service Assistant for Voice Permissions.

Transcript Quality

Service Assistant's ability to assign a subagent and create steps depends on the quality and speed of the transcription.

• Plan generation can be delayed at the start of a call. Before Service Assistant generates the first step of the plan, it requires enough messages in the transcript to understand the context of the call and the issue to resolve. If the transcript hasn't caught up to the conversation, the first step can take longer to appear. Service Assistant doesn't show a status indicator while it monitors the transcript, so the component stays blank until it identifies the issue and shows the first step. You can see a message such as, “There's not enough information to start the plan. Add more details, then check back here.” Plan processing starts when enough transcript text is available. What counts as enough information varies and depends on the clarity and speed of the transcription and how quickly the customer surfaces the reason for the call.
• Plan quality depends on transcript quality. If your transcription provider drops words, introduces errors, or delivers text in large delayed batches rather than incrementally, the plan can be less accurate or less timely. Transcript quality also affects agent actions. If the transcript doesn't clearly capture the context that triggers a specific action, that action doesn't always surface as a step in the plan. For example, a customer requests a refund, but the transcription renders it as an unrelated phrase. Reps can manually start agent actions or quick actions in the agent chat when the plan doesn't reflect what they're hearing on the call.

Record Processing and Updates

Grounding in a voice call is an active, fluid process. As the conversation unfolds, Service Assistant monitors each new transcript segment and continuously adapts the plan, so the steps always reflect where the conversation stands. For each step, Service Assistant runs a real-time search of all your grounding sources. As soon as the context shifts, Service Assistant adjusts the plan and creates a new step to match. Because a voice call is a live conversation, plan steps are more fluid than in case or messaging plans and can change frequently to stay current with the call.

By default, Service Assistant for Voice grounds on the conversation transcript and standard sources, such as subagents and knowledge articles. To add Voice Call field data and related-record data, choose one of these options.

• Related record grounding: Turn on Include Context from Related Records on the Voice tab of the Service Assistant Setup page. A configurable prompt template pulls standard and custom Voice Call field data into the service plan, and an overridable flow retrieves the related record details, such as Case and Account fields. Service Assistant captures updates to the grounding fields, prompt template, or flow and uses them to ground the next plan step. It doesn't regenerate the current step. Related record grounding is supported only in dynamic plans. See Grounding Voice Service Plans with Related Records (Optional) and Set Up Related Record Grounding for Voice (Optional).
• Agent action: Build an agent action that takes VoiceCall.Id as input and queries the RelatedRecordId field. Because RelatedRecordId is a polymorphic lookup field, your action must handle multiple object types.

For the agent action option, Voice Call records also include RecipientId and EndUserId fields that identify the participant who received the call. Depending on your configuration, you can query those fields instead of or in addition to the related record. For details, see the VoiceCall Object reference.

Considerations

• Service Assistant is active only when the Voice Call record is owned by a human user. If a bot or queue owns the record, Service Assistant doesn't process the record or generate a plan.
• Voice processing doesn't show a plan summary or a Start Plan button. The first step appears automatically after the issue is identified.
• Service Assistant can reference the transcript only from the Voice Call record page. Service Assistant can't access the transcript when it's installed on other record pages, such as Case or Account.
• Service Assistant and the chat box stay available for about 24 hours after the call ends, so reps can complete wrap-up tasks such as requesting a call summary or drafting a follow-up email.
• Service Assistant also supports context variables for passing record data to your actions. See Set Up Context Variables.