On the fifth day of Agentforce, Salesforce gave to me...five prompt patterns, four channel strategy, three action types, two data sources, and a chatbot in a web tree!
Sarah, a financial advisor at a boutique wealth management firm, starts her Monday morning with 47 unread emails. Three are urgent client questions about portfolio performance during last week's market volatility. She needs to check Salesforce for account details, consult with her operations team via Slack about transactions in progress, review portfolio positions in her financial planning software, and craft personalized responses—all while preparing for a 9:00 AM client meeting.
The Art of Speaking to AI
If the Atlas Reasoning Engine is your agent's brain, prompts shape its thinking. Output quality directly correlates to prompt quality.
Research Insight: Prompt engineering can improve LLM task performance by 50-300% compared to naive prompting approaches.
Source: Stanford HAI Research
Pattern 1: Role-Task-Format (RTF)
The RTF pattern provides a clear structure for every prompt by defining three essential components:
| Component | Purpose | Example |
|---|---|---|
| Role | Define persona | "You are an expert customer service rep..." |
| Task | Specify objectives | "Help customers troubleshoot issues..." |
| Format | Structure output | "Conversational tone, under 150 words..." |
Pattern 2: Few-Shot Prompting
Show, don't just tell. Provide examples of desired input-output pairs to demonstrate tone, structure, and edge case handling.
Few-shot prompting teaches your AI agent by example, ensuring it understands not just what to do, but how to do it in practice.
Pattern 3: Data Grounding
Anchor responses in verified CRM data to prevent hallucinations:
Use ONLY the following verified information:
Order Number:
Status:
Do NOT provide information not listed above.Data grounding ensures your agent stays factual and trustworthy by limiting responses to information that exists in your Salesforce system.
Pattern 4: Positive Guidance
Frame instructions positively to create clearer action paths:
| ❌ Negative Framing | ✅ Positive Framing |
|---|---|
| "Don't forget to verify identity" | "Always verify identity by asking for email" |
| "Never reveal internal policies" | "Focus on what we CAN do for the customer" |
Positive guidance tells your agent what to do, not just what to avoid, creating more confident and helpful responses.
Pattern 5: Scope Boundaries
Define clear operational limits for your agent:
You CAN:
- Check order status
- Process returns under $500
- Apply standard promos
You CANNOT:
- Issue refunds over $500
- Access payment card details
- Make policy exceptions
When outside scope: "I'll connect you with a specialist who can help."
Explicit boundaries prevent your agent from overstepping while maintaining a helpful tone when escalation is needed.
Key Takeaways: Day 5
✓ RTF provides clear structure for every prompt
✓ Examples demonstrate better than explanations
✓ Grounding prevents hallucinations
✓ Positive guidance creates clearer action paths
✓ Explicit boundaries prevent overstepping
Series Navigation
Day 5 of 12 | ← Day 4: Omnichannel | → Tomorrow: Six metrics that prove your AI ROI!
Ready to Start Your Agentforce Journey?
Vantage Point helps at every stage—from strategy and design to implementation and optimization.
📧 info@vantagepoint.io
🌐 vantagepoint.io/services/technology/salesforce/agentforce
About the Author
David Cockrum founded Vantage Point after serving as Chief Operating Officer in the financial services industry. His unique blend of operational leadership and technology expertise has enabled Vantage Point's distinctive business-process-first implementation methodology, delivering successful transformations for 150+ financial services firms across 400+ engagements with a 4.71/5.0 client satisfaction rating and 95%+ client retention rate.
-
-
- Email: david@vantagepoint.io
- Phone: (469) 652-7923
- Website: vantagepoint.io
-
