AI Agents

1. Go to AI Agents → AI Fields
2. Click "Create Field"
3. Choose field type and configure:
   • Field Name: What you're extracting (e.g., "Budget Range")
   • Field Type: Text, Number, Single Select, Multi Select or Boolean
   • Description: Help AI understand what to look for
   • Extraction Prompt: Instructions for the AI
   • Optional values: For select fields, add the allowed options
4. Set conversation filters and active/inactive status
5. Test on sample conversations, then save

The quality of your extraction prompt directly impacts accuracy:

DO: 'Extract the specific dollar amount or range mentioned for annual budget'
DON'T: 'Get budget info'

DO: 'Extract decision maker titles (e.g., VP, Director, C-level) and their departments'
DON'T: 'Get decision makers'

DO: 'If no budget is mentioned, return "Not Discussed". If they say "flexible", note that.'
DON'T: 'Extract budget'

DO: 'Return dates in YYYY-MM-DD format. If only a month is mentioned, use the last day of that month.'
DON'T: 'Extract dates'

1. Navigate to AI Agents → AI Summaries
2. Click "Create Summary"
3. Configure summary:
   • Summary Name: Descriptive name (e.g., "Enterprise Sales Summary")
   • Summary Type: Label the summary style you want to run
   • Template (required): Define the structure and output format you want
   • Prompt (optional): Add extra AI guidance for generation
   • Status: Set active/inactive to control automatic runs
4. Set conversation filters to control which conversations the summary applies to
5. Use Test Summary to validate output on sample conversations
6. Save the summary
7. Optional: run on past conversations after saving

Create a sales call summary with these sections:

## Deal Status
- Current stage in pipeline
- Changes from last conversation
- Overall deal health (Green/Yellow/Red)

## Key Discussion Points
- Top 3 topics discussed
- Customer sentiment on each

## Stakeholders
- Who participated
- New stakeholders identified
- Decision-making authority

## Objections & Concerns
- Any objections raised
- How they were addressed
- Remaining concerns

## Next Steps
- Agreed action items with owners
- Timeline for follow-up
- Success criteria for next meeting

Keep each section concise (2-3 sentences). Highlight any red flags in bold.

• Balance detail and brevity: More detail isn't always better - focus on actionable insights
• Use consistent formatting: Structured summaries are easier to scan than paragraphs
• Include context clues: Dates, participant names and call type help orientation
• Highlight urgency: Use formatting (bold, caps) to call out time-sensitive items
• Test with diverse calls: Ensure your prompt works for short and long calls, positive and negative outcomes
• Version your prompts: Keep track of what changes improve quality

1. Go to AI Agents → AI Playbooks
2. Select "Create Playbook"
3. Set up basic playbook details:
   • Playbook Name: E.g., "Enterprise Discovery Methodology"
   • Description (optional): Short context for reviewers
   • Status: Active or inactive
4. Add and configure playbook steps:
   • Step Name: What should happen (e.g., "Budget Discovery")
   • Description (optional): Guidance for reviewers
   • Detection Method: AI prompt or keyword matching
   • Sub-items (optional): Add required/optional checklist items per step
   • Order controls: Optionally enforce overall step order and sub-item order
5. Set conversation filters to define where the playbook should run
6. Save the playbook
7. Run Test Playbook and, if needed, Run on Past Data

After each conversation, the playbook generates:

• Start with your best rep: Document what your top performers do consistently
• Keep steps clear and actionable: "Ask about budget" not "Discuss financials"
• Don't overcomplicate: 5-8 steps is ideal; more than 12 becomes unwieldy
• Allow flexibility: Mark optional steps as such; not every call follows the same path
• Update regularly: Review and refine based on what actually works
• Use for onboarding: Playbooks are excellent training tools for new hires

For Keyword Trackers (Level 1):

1. Go to AI Agents → Keyword Trackers
2. Click "Keyword Trackers"
3. Enter keywords/phrases (one per line)
4. Choose case-sensitive matching (optional)
5. Save and activate

For AI Keyword Trackers (Level 2):

1. Click "AI Keyword Trackers"
2. Name your concept (e.g., "Budget Objections")
3. Provide positive examples (phrases that match):
   • "That's outside our budget"
   • "We can't afford that right now"
   • "Price is a concern for us"
4. Provide negative examples (similar phrases that shouldn't match):
   • "Our budget is flexible"
   • "Price isn't an issue"
5. Train on past conversations
6. Review accuracy and adjust examples
7. Activate when satisfied

For Advanced Trackers (Level 3):

1. Click "Advanced Trackers"
2. Define an AI prompt for the concept or pattern you want to detect
3. Add expected categories if you want structured outputs
4. Test the tracker and refine prompts before activation
5. Save and activate

Once activated, trackers provide:

• Statistical Widgets: View mention frequency over time
• Conversation Filtering: Find all conversations where keyword was mentioned
• Trend Analysis: Track how often concepts come up week-over-week
• Correlation Analysis: See which keywords appear together
• Dashboard Integration: Add tracker widgets to Insights dashboard

From the AI Agents page, you can review two management views:

• All Agents view: Name, type, status, category, last updated, created by and runs count
• Quick row actions: Edit, activate/deactivate and delete from the table
• Category management tables: Per-type views (Fields, Summaries, Playbooks, Trackers, Employees, Advanced Agents) with type-specific columns
• Status and scope context: Active/inactive state and configured filters are visible in each automation editor
• Detailed quality checks: Use Test actions inside editors (for example Test Summary, Test Playbook, Test Tracker) before broad rollout

To modify an automation:

1. Open the automation from its table row (or click the Edit action)
2. Update configuration in the editor (rules, prompts, filters and status as applicable)
3. Save changes
4. If you need historical reprocessing, use Run on Past Conversations:
   • Select a date range to control scope
   • Choose mode: Add where missing or Replace existing results
   • Start backfill and monitor progress until completion

For AI Agents, workflow department standards are workflow-owned and sync with AI Agent conversation scope. This keeps assignment behavior deterministic across Workflow, AI Agent editors and runtime processing.

Get more value by using AI agents together:

• Start small: Begin with 2-3 high-impact automations, add more as you see value
• Audit regularly: Review which automations are actually being used quarterly
• Consolidate where possible: Use one well-configured field instead of three similar ones
• Set clear owners: Assign someone to maintain each automation
• Track ROI: Measure time saved vs. effort to maintain