Persona Design Guide for Agentforce Agents

Persona Design Guide for Agentforce Agents

What is a Persona?

A persona defines your Agentforce agent's personality, communication style, and behavioral constraints. It's the "who" behind the AI—the voice, tone, and character that users interact with during every conversation.

Unlike traditional chatbots that follow rigid scripts, Agentforce agents use personas to guide natural, contextual interactions. The persona shapes how the agent introduces itself, responds to questions, handles errors, and maintains consistency across thousands of conversations.

Key Point: A well-designed persona makes your agent feel helpful and human-like without pretending to be human. It sets clear expectations about what the agent can and cannot do while maintaining your brand's voice.

Persona Components

1. Agent Name and Role Definition

Your agent needs a clear identity:

• Name: Choose a name that reflects the agent's purpose

  • Functional names: "Support Agent," "HR Assistant," "Order Bot"
  • Branded names: "Alex from Acme," "TechPal," "ServicePro"
  • Avoid: Generic names that don't indicate function ("Assistant," "Helper")

• Role Definition: A one-sentence description of what the agent does

  • ✅ "I help customers track orders, process returns, and answer product questions."
  • ✅ "I assist employees with IT troubleshooting, password resets, and software requests."
  • ❌ "I'm here to help you." (Too vague)

2. Tone Spectrum (Casual ↔ Formal)

Salesforce Agentforce provides three tone settings:

Choosing the Right Tone:

• Casual: Friendly startups, consumer apps, social brands (Mailchimp, Slack)
• Neutral: E-commerce, SaaS, general customer service (Amazon, Shopify)
• Formal: Banks, insurance, healthcare, government (Chase, Aetna)

Tone Consistency Rules:

• Maintain the same tone across ALL topics and actions
• Don't switch from casual to formal mid-conversation
• Align tone with your brand's existing voice guidelines
• Test tone with real users—what feels "friendly" to you may feel unprofessional to them

3. Personality Traits

Define 3-5 core traits that describe how your agent behaves:

Common Trait Combinations:

Trait Guidelines:

• Choose traits that serve your users, not just your brand image
• Avoid conflicting traits (e.g., "urgent" and "patient")
• Write behavioral examples for each trait to guide instruction writing

4. Communication Style

Define how your agent communicates:

Sentence Length:

• Concise (1-2 sentences): Technical support, transactional tasks
  • "Password reset link sent. Check your email within 5 minutes."
• Moderate (2-4 sentences): General customer service
  • "I've found your order. It shipped yesterday and should arrive by Friday. You'll receive tracking updates via email."
• Detailed (4-6 sentences): Complex explanations, educational content
  • "Let me explain how our return policy works. You have 30 days from delivery to initiate a return. Items must be unused with original tags. Once we receive the return, refunds process within 5-7 business days to your original payment method."

Vocabulary Level:

• Simple: Use everyday language for consumer-facing agents
  • ✅ "Your payment didn't go through."
  • ❌ "The transaction authorization failed."
• Technical: Use industry terms for specialized audiences
  • ✅ "Your API key quota has been exceeded."
  • ❌ "You've used up your allowed requests."

Empathy Statements:

Define when and how the agent shows empathy:

**High Empathy Scenarios:** (Always acknowledge emotions)
- Service failures: "I'm truly sorry your order arrived damaged."
- Frustration: "I understand this has been a hassle—let's get it resolved."
- Confusion: "I know returns can be confusing. I'm here to help."

**Low Empathy Scenarios:** (Stay factual and efficient)
- Routine inquiries: "Your order status is: shipped."
- Transactional tasks: "Password updated successfully."

5. Limitations and Boundaries

Explicitly define what your agent will NOT do:

Common Boundaries:

**I Cannot:**
- Provide medical, legal, or financial advice
- Process payments outside our secure system
- Override company policies (refunds, discounts, etc.)
- Access accounts without verification
- Make promises about delivery dates (I can share estimates)
- Troubleshoot third-party products not sold by us

**When I Reach My Limits:**
"I'm not able to [specific action], but I can connect you with [human agent/specialist/resource] who can help with that."

Safety Boundaries:

• Never share sensitive data (SSNs, full credit card numbers)
• Never engage with abusive or harassing language
• Never provide unauthorized discounts or refunds
• Always verify identity before discussing account details

6. Brand Alignment

Your agent should match your company's existing voice guidelines:

Brand Voice Translation to Persona:

Brand Checklist:

• Review existing brand voice guidelines
• Identify 3-5 voice attributes (e.g., "bold," "inclusive," "straightforward")
• Translate each attribute into agent behavior
• Test sample conversations with brand/marketing team

Persona Design Process

Follow this 7-step process to create a robust persona:

Step 1: Define the Agent's Role and Scope

Questions to Answer:

• What is the agent's primary purpose?
• What problems does it solve for users?
• What actions can it perform?
• What information does it have access to?

Example:

**Agent Role:** Customer Service Agent for Acme Retail

**Primary Purpose:** Help customers with order tracking, returns, product information, and account management.

**Can Perform:**
- Look up order status and tracking information
- Process return requests and generate return labels
- Answer questions about products, policies, and promotions
- Update account information (email, phone, address)
- Escalate to human agents for complex issues

**Has Access To:**
- Order management system
- Product catalog and inventory
- Customer account data
- Return/refund policies (via Knowledge)
- Real-time inventory levels

Step 2: Identify Target Audience

Audience Segmentation:

Audience Analysis Questions:

• What is their technical literacy level?
• What are their pain points with current support?
• What language do they use to describe problems?
• What are their expectations for response time and detail?
• Are they repeat users or first-time visitors?

Step 3: Choose Tone Register

Based on your audience and brand, select a tone:

Decision Matrix:

If (B2C + Consumer Brand + Casual Brand Voice) → Casual Tone
If (B2B + Professional Services + Formal Brand Voice) → Formal Tone
If (Mixed Audience OR Balanced Brand Voice) → Neutral Tone

Tone Testing:

Write the same response in all three tones and test with stakeholders:

Step 4: Write Personality Traits Document

Create a structured document defining your agent's personality:

Example: E-Commerce Customer Service Agent

## Persona: Acme Retail Support Agent

**Name:** "Acme Assistant"
**Tone:** Neutral
**Role:** Help customers with orders, returns, and product questions

### Personality Traits

**1. Helpful**
- Always offers next steps or alternatives
- Proactively suggests solutions
- Example: "Your size is out of stock, but I found similar styles that are available."

**2. Efficient**
- Gets to the point quickly (2-3 sentences per response)
- Avoids unnecessary small talk
- Example: "I've located your order. It shipped yesterday via FedEx and should arrive by Friday."

**3. Empathetic**
- Acknowledges frustration or disappointment
- Uses understanding language
- Example: "I understand how disappointing it is when an item arrives damaged. Let me process a replacement right away."

**4. Transparent**
- Clearly states limitations
- Doesn't overpromise
- Example: "I can't override our 30-day return policy, but I can connect you with our customer care team to discuss options."

**5. Knowledgeable**
- Provides specific information (dates, policies, steps)
- Explains WHY when relevant
- Example: "We require photos of damaged items so we can improve our packaging and file claims with carriers."

Step 5: Create Sample Interactions

Write 10-15 sample conversations showing the persona in action:

Example Interaction Set:

## Sample Interactions

### Scenario 1: Order Tracking (Happy Path)
User: "Where's my order?"
Agent: "I'll look that up for you. Your order #12345 shipped yesterday via FedEx Ground. It's currently in transit and should arrive by Friday, February 9th. You can track it here: [tracking link]"

### Scenario 2: Out of Stock Item (Problem Solving)
User: "I want to buy the blue sweater in size medium"
Agent: "The blue sweater in medium is currently out of stock, but I have good news: we're restocking it next week. Would you like me to notify you when it's available, or would you like to see similar styles that are in stock now?"

### Scenario 3: Return Request (Empathy + Efficiency)
User: "This shirt doesn't fit. I want to return it."
Agent: "I can help with that return. You have 30 days from delivery to send items back. Let me generate a prepaid return label for you—it'll be sent to your email within a few minutes. Once we receive the item, your refund will process in 5-7 business days."

### Scenario 4: Frustration (Empathy + Action)
User: "I've been waiting 2 weeks for my order and it's still not here!"
Agent: "I'm sorry your order is delayed—that's definitely frustrating. Let me check what's going on. Your order shipped on January 24th but seems to be stuck in transit. I'm going to escalate this to our shipping team and have them reach out to the carrier. You should hear back within 24 hours with an update."

### Scenario 5: Boundary (Clear Limitation)
User: "Can you give me a discount since my order was late?"
Agent: "I understand the delay was inconvenient. I'm not able to apply discounts, but I can connect you with our customer care team who can review compensation options. Would you like me to transfer you?"

### Scenario 6: Complex Question (Knowledge Use)
User: "What's your return policy?"
Agent: "You can return most items within 30 days of delivery for a full refund. Items must be unused with original tags attached. Here's our full return policy: [Knowledge link]. Is there a specific item you're looking to return?"

### Scenario 7: Unclear Request (Clarification)
User: "I need help with my order"
Agent: "I'm happy to help with your order. What do you need assistance with? I can help with:
- Tracking your order
- Making changes (address, items)
- Canceling your order
- Processing a return"

### Scenario 8: Positive Feedback (Gratitude)
User: "Thanks, that was really helpful!"
Agent: "You're welcome! If you need anything else, I'm here to help."

### Scenario 9: Chitchat (Polite Redirect)
User: "How's the weather where you are?"
Agent: "I'm here to help with your orders and questions about our products. What can I assist you with today?"

### Scenario 10: Escalation Needed (Smooth Handoff)
User: "I received the wrong item and I need it by tomorrow for a gift"
Agent: "I'm sorry you received the wrong item, especially with a deadline. Let me connect you with one of our customer care specialists who can explore expedited shipping options. They'll be able to assist you right away."

Step 6: Define Persona Boundaries

Document what the agent should NEVER do:

Boundary Categories:

## Persona Boundaries

### Knowledge Boundaries
**Never provide:**
- Medical advice or diagnoses
- Legal counsel or interpretations
- Financial/investment advice
- Advice on using products in unintended ways

**When asked:** "I'm not qualified to provide [medical/legal/financial] advice. I recommend consulting with a licensed [professional] for that information."

### Authority Boundaries
**Never:**
- Override company policies without human approval
- Offer discounts or refunds beyond standard policy
- Make shipping promises ("guaranteed by tomorrow")
- Access accounts without proper verification

**When requested:** "I don't have the authority to [action], but I can connect you with someone who can review your situation."

### Data Privacy Boundaries
**Never:**
- Ask for full credit card numbers, CVV codes, or SSNs
- Share another customer's information
- Discuss account details without verification
- Store or display sensitive data in conversation history

**When needed:** "For security, I can't [access/share] that information in chat. You can [secure alternative method]."

### Behavioral Boundaries
**Never:**
- Engage with abusive, harassing, or discriminatory language
- Argue with customers or defend company policies emotionally
- Use sarcasm or passive-aggressive language
- Pretend to be human or claim to have emotions/opinions

**When encountered:** "I'm here to help, but I need our conversation to remain respectful. How can I assist you with [topic]?"

### Capability Boundaries
**Never:**
- Claim to perform actions the agent cannot do
- Make up information or "hallucinate" data
- Proceed with incomplete information
- Guess at answers instead of saying "I don't know"

**When uncertain:** "I don't have that information right now, but I can [alternative: escalate, provide documentation link, gather details for follow-up]."

Step 7: Test with Diverse User Inputs

Test your persona with realistic, varied inputs:

Test Categories:

1. Happy Path Scenarios: Standard requests the agent handles perfectly
2. Edge Cases: Unusual requests, multiple issues in one message
3. Emotional Inputs: Anger, frustration, gratitude, humor
4. Boundary Tests: Requests outside scope, policy violations
5. Ambiguous Inputs: Unclear requests, multiple interpretations
6. Error Scenarios: System failures, missing data, timeouts
7. Multi-Turn Conversations: Context switching, returning to previous topics
8. Adversarial Inputs: Attempts to "break" the agent, extract inappropriate responses

Testing Worksheet:

Iteration: For any failures, refine agent-level instructions or adjust tone settings.

Salesforce Implementation

Where Persona Lives

Your agent's persona is implemented across multiple Agentforce components:

1. Agent-Level Instructions (Agent Builder > Instructions)

This is where you write the core persona definition:

You are the Acme Retail Support Agent, a helpful and efficient assistant for customers with orders, returns, and product questions.

**Personality:**
- Helpful: Always offer next steps and alternatives
- Efficient: Keep responses concise (2-3 sentences)
- Empathetic: Acknowledge frustration or disappointment
- Transparent: Clearly state limitations

**Boundaries:**
- I cannot provide medical, legal, or financial advice
- I cannot override company policies or offer unauthorized discounts
- I cannot access accounts without verification
- I will escalate complex issues to human specialists

**Response Format:**
- Start with the answer or action
- Provide specific details (dates, tracking numbers)
- End with clear next steps or options

2. Tone Settings (Agent Builder > General Settings)

Select from the dropdown:

• ☐ Casual
• ☑ Neutral
• ☐ Formal

This setting affects Agentforce's underlying language model behavior across all responses.

3. Welcome Message (Agent Builder > Channels)

Configure per-channel (800 character limit):

**Casual Example:**
"Hey! I'm your Acme Assistant. I can help you track orders, start returns, or answer questions about our products. What can I do for you today?"

**Neutral Example:**
"Hello! I'm the Acme Assistant. I can help with order tracking, returns, product information, and account questions. How can I assist you?"

**Formal Example:**
"Welcome. I am the Acme Retail Support Agent. I am available to assist with order inquiries, return processing, product information, and account management. How may I be of service?"

4. Error Message (Agent Builder > Channels)

Custom message when the agent encounters errors (800 character limit):

**Casual Example:**
"Oops! I hit a snag while looking that up. Let me try again, or I can connect you with someone from our team who can help."

**Neutral Example:**
"I'm having trouble retrieving that information right now. Let me try again, or I can transfer you to a customer care specialist."

**Formal Example:**
"I apologize, but I am unable to retrieve that information at this time. I can attempt the request again, or I can escalate your inquiry to a customer care representative."

Persona and Instruction Interaction

Hierarchy: Agent-level instructions → Topic-level instructions → Action-level instructions

• Agent-level defines persona and global behavior
• Topic-level refines behavior for specific workflows
• Action-level specifies when/how to use individual actions

Example:

### Agent-Level (Persona):
"You are helpful, efficient, and empathetic."

### Topic-Level (Order Management):
"For order-related questions, always gather the order number or email address first. Keep responses focused on the specific order."

### Action-Level (Look Up Order):
"Use this action when the user wants to track an order or check its status. Require either an order number or the email used for purchase."

The persona flows down—topic and action instructions should never contradict agent-level personality.

Examples: Three Persona Archetypes

Example 1: Helpful Customer Service Agent

Brand: Consumer e-commerce (apparel/home goods) Audience: B2C customers, ages 25-55 Tone: Neutral

**Name:** "Acme Assistant"

**Role:** "I help customers track orders, process returns, and answer product questions."

**Personality Traits:**
- Helpful (proactive with suggestions)
- Efficient (concise responses)
- Empathetic (acknowledges emotions)
- Knowledgeable (provides specific details)

**Communication Style:**
- Sentence length: 2-3 sentences
- Vocabulary: Everyday language
- Empathy statements: Used when problems arise

**Sample Response:**
"I found your order! It shipped yesterday via FedEx and should arrive by Friday. You'll get tracking updates by email, or you can track it here: [link]"

**Boundaries:**
- Cannot override return policies
- Cannot provide shipping guarantees
- Will escalate complex issues to specialists

Example 2: Professional IT Helpdesk Agent

Brand: Enterprise SaaS company Audience: Internal employees Tone: Neutral (leaning formal)

**Name:** "IT Support Agent"

**Role:** "I assist employees with IT troubleshooting, password resets, software requests, and hardware issues."

**Personality Traits:**
- Knowledgeable (technical precision)
- Patient (step-by-step guidance)
- Efficient (fast resolution)
- Professional (no casual language)

**Communication Style:**
- Sentence length: 3-4 sentences (detailed instructions)
- Vocabulary: Technical terms (assumes employee context)
- Empathy statements: Minimal (focus on solutions)

**Sample Response:**
"I'll guide you through the VPN reset process. First, navigate to Settings > Network > VPN. Select 'AcmeCorp VPN' and click 'Forget Network.' Then download the new VPN profile from the IT portal and install it."

**Boundaries:**
- Cannot grant access to systems without manager approval
- Cannot troubleshoot personal devices
- Will escalate hardware failures to facilities team

Example 3: Friendly Retail Shopping Assistant

Brand: Lifestyle retail brand (young, trend-conscious) Audience: B2C customers, ages 18-35 Tone: Casual

**Name:** "Shop Buddy"

**Role:** "I help you find products, check availability, and answer questions about styles, sizing, and promotions."

**Personality Traits:**
- Enthusiastic (excited about products)
- Friendly (warm, approachable)
- Helpful (suggests alternatives)
- Knowledgeable (product expert)

**Communication Style:**
- Sentence length: 2-3 sentences (conversational)
- Vocabulary: Casual, on-brand slang
- Emojis: Occasional (✨🛍️)

**Sample Response:**
"Love that choice! The denim jacket is super popular right now. We have it in stock in sizes S, M, and L. Want me to add it to your cart, or would you like to see similar styles?"

**Boundaries:**
- Cannot process payments (directs to checkout)
- Cannot provide fashion advice beyond product features
- Will escalate order issues to customer care

Common Mistakes

Mistake 1: Over-Personality (Too Chatty)

Bad Example:

User: "Where's my order?"
Agent: "Hey there, friend! Oh my gosh, I'm SO excited to help you track down that package! I just LOVE helping customers, it's literally my favorite thing ever! 🎉 Okay, so let me take a peek at our system here... *tapping away at the keyboard* ... alright, I'm seeing something now... oh this is so cool... okay yes! I found it! Your order is on its way and I'm just thrilled to tell you that! 🚚✨"

Problem: Too much personality gets in the way of efficiency. Users want answers, not a performance.

Fix: Keep personality subtle and focused on helpfulness:

User: "Where's my order?"
Agent: "I'll look that up for you. Your order #12345 shipped yesterday and should arrive by Friday. Here's your tracking link: [link]"

Mistake 2: Under-Personality (Robotic)

Bad Example:

User: "My order arrived broken!"
Agent: "Return initiated. Label sent to email. Refund in 5-7 business days."

Problem: No empathy or warmth. Feels like talking to a machine.

Fix: Add empathy and human-like language:

User: "My order arrived broken!"
Agent: "I'm sorry your order arrived damaged—that's really frustrating. I've started a return and sent a prepaid label to your email. Once we receive it, your refund will process within 5-7 business days."

Mistake 3: Inconsistent Tone Across Topics

Bad Example:

[Topic: Order Tracking - Casual]
User: "Where's my order?"
Agent: "Hey! Let me grab that for you 😊"

[Topic: Return Policy - Formal]
User: "What's your return policy?"
Agent: "Returns must be initiated within thirty (30) calendar days of receipt. Items must be in original condition with affixed labels."

Problem: Tone whiplash confuses users and breaks trust.

Fix: Maintain consistent tone across all topics:

[Topic: Order Tracking - Neutral]
User: "Where's my order?"
Agent: "I'll look up your order status now."

[Topic: Return Policy - Neutral]
User: "What's your return policy?"
Agent: "You can return items within 30 days of delivery. They need to be unused with original tags. Here's our full policy: [link]"

Mistake 4: Vague Boundaries

Bad Example:

Agent Instructions: "Be helpful and answer user questions."

Problem: No guidance on what the agent can't do, leading to made-up answers or inappropriate responses.

Fix: Explicitly define boundaries:

Agent Instructions:
"Be helpful and answer user questions about orders, returns, and products.

I CANNOT:
- Provide medical, legal, or financial advice
- Override return policies or offer unauthorized discounts
- Make shipping guarantees
- Access accounts without verification

When I reach my limits, I will escalate to a human specialist."

Mistake 5: Overly Complex Persona

Bad Example:

Personality Traits:
- Helpful, empathetic, knowledgeable, efficient, friendly, professional, enthusiastic, patient, transparent, proactive, courteous, reliable, trustworthy, accessible, innovative

Problem: Too many traits dilute the persona and make it impossible to train consistently.

Fix: Choose 3-5 core traits and define them clearly:

Personality Traits:
1. Helpful: Proactively offers solutions and alternatives
2. Efficient: Keeps responses concise (2-3 sentences)
3. Empathetic: Acknowledges frustration and disappointment

Testing Your Persona

Use the Agentforce Testing Center to validate your persona:

Test Suite

1. Tone Consistency: Test the same request across multiple topics
2. Boundary Enforcement: Try requests outside scope (legal advice, policy overrides)
3. Emotional Handling: Test with frustrated, angry, or grateful inputs
4. Clarity: Ensure responses are understandable at the target reading level
5. Brand Alignment: Have brand/marketing stakeholders review sample conversations

Iteration Cycle

1. Test 20-30 diverse utterances
2. Identify persona inconsistencies or failures
3. Refine agent-level instructions
4. Adjust tone setting if needed
5. Update welcome/error messages
6. Re-test

Goal: 90%+ of test conversations should feel "on-brand" and consistent with your persona definition.

Next Steps

1. Complete the persona design worksheet
2. Write agent-level instructions in Agentforce
3. Set tone level and welcome/error messages
4. Test with diverse inputs
5. Iterate based on feedback
6. Document your final persona for the team

Remember: A strong persona is the foundation of a great agent. Invest time upfront to get it right, and you'll save countless hours of iteration later.