The Generate Insights action uses AI to extract, transform, and structure information from meeting transcripts, Email and Support Case messages, CRM data, and other workflow inputs.
Context
The Context section defines what background information the AI has when generating outputs. This can include meeting details, participants, CRM associations, or organizational metadata.
When your workflow starts from a Call Completed trigger, context is automatically pre-populated with the full transcript, meeting metadata (title/date/time/duration), attendee details, and linked CRM records.
You can insert variables for grounding, e.g.,
{{meeting.name}},{{organization.name}}.Write the context as clear instructions to the AI, for example:
“You are an AI agent tasked with generating deal insights from the meeting transcript. Use the transcript and CRM data to identify MEDDIC fields.”
Best practice: Always include meeting transcript and participant details, and add organizational or team context if available.
Screening Prompt
Screening prompts act as pre-checks to prevent unnecessary field generation.
Always write the screening prompt as a yes/no or true/false question.
Example: “Were next steps discussed in this conversation?”
If the AI determines the answer is no, the field will not be generated.
Useful for optional fields like competitor mentions, next steps, or objections.
Best practice: Use screening prompts on fields that are not always relevant. This reduces noise and conserves credits.
Prompt
The Prompt controls the exact instruction for the AI when generating field content.
Examples:
“What were the next steps discussed? Format as concise bullet points.”
“Who was identified as the economic buyer? Return only the name and title.”
By field type:
Textarea fields - use structured, concise instructions.
Rich Textarea fields - you can request formatting like paragraphs or bold.
Best practice: Keep prompts narrowly focused and reference variables directly, such as the meeting transcript, to reduce hallucinations.
Save / Append Settings
Control how AI-generated values are written to the target field.
Overwrite (default) - replace any existing value in the field.
Combine Intelligently - merge new insights with existing values and deduplicate where possible.
Append to Top - add the new insight at the top of the field.
Append to Bottom - add the new insight at the bottom of the field.
Best practice:
Use Combine Intelligently for running fields like notes or competitor lists.
Use Append to Top for time-sensitive updates like latest next steps.
Split to List
Enable Split to List when you expect multiple discrete items.
Example: if AI generates a bullet list of products discussed, splitting will create individual list entries for each product.
This is essential when you plan to loop over AI outputs later, such as iterating over products, contacts, or action items in a Loop action.
Best practice: Turn on Split to List whenever a downstream step needs to process each item individually, such as using a Loop action to create one CRM record per product discussed.
Test
Use the Test tab to validate before going live.
Select a recent meeting or upload a transcript to simulate generation.
Review outputs in real time to confirm formatting, accuracy, and completeness.
Iterate on context, screening prompt, or prompt until the output is reliable.
Best practice: Test each field with at least 2–3 meetings of different types (discovery, demo, renewal) to ensure robustness.
Example Use Cases
Populate MEDDIC fields like Economic Buyer, Metrics, and Decision Criteria directly from meeting transcripts.
Auto-generate handoff notes for customer success after a closed-won opportunity.
Capture objections, competitor mentions, or risks in structured CRM fields.
Extract product mentions during calls and log them in opportunity records.