Prompting Guide

Effective prompting is the foundation for success with Brightsy AI. This guide provides practical examples and best practices for crafting prompts that leverage the full potential of Brightsy's tools and capabilities.

Prompting Fundamentals

Clear Directives

Always begin with clear, specific instructions that define the goal:

You are a customer support specialist for Acme Corp. Your task is to analyze the following customer email and categorize it as "Billing Question", "Technical Support", or "Product Inquiry". Then draft a professional response that addresses their concern.

Customer Email: {{email_content}}

Structured Prompts

Use structured formats to guide the agent through complex reasoning:

To analyze this sales data, please follow these steps:
1. Identify the top 3 performing products by revenue
2. Calculate month-over-month growth percentages
3. Highlight any concerning trends in the data
4. Recommend 2-3 specific actions based on your findings

Sales Data: {{sales_data}}

Brightsy-Specific Techniques

Agent Instructions

When configuring agents in Brightsy, use detailed system instructions to define behavior:

You are a data analysis expert specializing in financial data. Your role is to help users understand complex financial information, identify trends, and make data-driven recommendations.

When analyzing data:
- Always begin by understanding what type of analysis the user needs
- Use appropriate quantitative methods based on the data type
- Present findings in a clear, concise format with actionable insights
- When unclear about what method to use, explain pros and cons of different approaches
- Format currency values with appropriate symbols and decimal places

You have access to the following tools:
- Data Management tools for querying financial records
- Chart generation tools for visualizing trends
- Web tools for retrieving market comparison data

Scenario Design

When creating scenarios, provide context and clear execution instructions:

# Monthly Sales Report Generator

## Context
This scenario runs on the 1st of each month to analyze the previous month's sales data and generate a comprehensive report.

## Instructions
1. Query the sales records for the previous month using the CMS Data Management tools
2. Calculate key metrics: total revenue, units sold, average order value, top 5 products
3. Compare results to the same month last year and the previous month
4. Generate visualizations for month-over-month and year-over-year trends
5. Identify any anomalies or significant changes in purchasing patterns
6. Create a formatted report with executive summary and detailed analysis
7. Email the report to the sales director and executive team

Tool-Specific Prompting Examples

Data Management Tools

When working with CMS and record types:

Retrieve all customer support tickets created in the last 7 days that have a status of "Unresolved" and contain keywords related to our checkout system. Analyze these tickets to identify common issues and suggest potential solutions.

This prompt effectively utilizes Brightsy's Data Management capabilities by:

• Specifying clear criteria for record filtering
• Defining a time range for relevance
• Including status and content filters
• Requesting specific analytical outcomes

Example implementation using Brightsy tools:

get_records(
  account_id="{{account_id}}",
  type_id="support_ticket",
  data_filter=[
    {"field": "created_at", "operator": ">=", "value": "{{seven_days_ago}}"},
    {"field": "status", "operator": "=", "value": "Unresolved"},
    {"field": "description", "operator": "contains", "value": "checkout"}
  ],
  limit=100
)

Natural Language Tool Instructions

Models can often understand how to use tools without explicit code examples. Here are effective ways to instruct models to use tools through natural language:

Find all high-priority customer issues reported in the last 24 hours and create a summary report categorized by department.

The model understands to:

1. Use data management tools to query recent customer issues with high priority
2. Filter by timestamp and priority fields
3. Group results by department
4. Generate a structured summary

Generate a visualization showing our monthly revenue trends compared to expenses for the past year, with a forecast for the next quarter.

The model knows to:

1. Query financial records for revenue and expense data
2. Structure the data in a time series format
3. Use chart generation tools to create appropriate visualizations
4. Apply forecasting calculations for future projections

When a user submits the contact form, validate their email format, check if they're an existing customer in our database, and route their inquiry to the appropriate department based on the subject line.

The model understands this workflow requires:

1. Form input validation
2. Database lookup for customer verification
3. Text analysis of subject line
4. Conditional logic for routing
5. Record creation for the inquiry

UI Tools

For collecting information through forms:

Gather information from the user about their marketing campaign needs. Ask for the campaign name, target audience, budget range, timeline, and primary goals. Use this information to recommend appropriate marketing channels and strategies.

Example implementation:

user_info_request({
  "props": {
    "title": "Marketing Campaign Information",
    "description": "Please provide details about your upcoming marketing campaign",
    "schema": {
      "type": "object",
      "properties": {
        "campaign_name": {"type": "string", "title": "Campaign Name"},
        "audience": {"type": "string", "title": "Target Audience Description"},
        "budget": {
          "type": "string", 
          "title": "Budget Range",
          "enum": ["Under $5,000", "$5,000-$15,000", "$15,000-$50,000", "Over $50,000"]
        },
        "timeline": {"type": "string", "title": "Expected Timeline (in weeks)"},
        "goals": {
          "type": "array",
          "title": "Primary Campaign Goals",
          "items": {
            "type": "string",
            "enum": ["Brand Awareness", "Lead Generation", "Direct Sales", "Customer Retention", "Other"]
          }
        }
      },
      "required": ["campaign_name", "audience", "budget", "timeline", "goals"]
    }
  }
})

Agent Collaboration

For scenarios involving multiple agents:

You are the Project Manager agent coordinating a content creation workflow. When a new blog post request arrives:
1. Extract topic details and target audience from the request
2. Delegate initial research to the Research Agent
3. Based on research findings, instruct the Content Writer Agent to create a draft
4. Send the draft to the Editor Agent for review and refinement
5. Once approved, use the CMS tools to publish the content
6. Notify the requester that their content has been published

For each step, incorporate feedback loops to ensure quality and alignment with requirements.

Advanced Techniques

Conditional Logic

Use if-then structures to handle different scenarios:

Analyze the customer's purchase history and provide personalized recommendations.

If the customer has purchased technology products in the last 30 days:
  - Recommend accessories and complementary products
  - Include information about extended warranty options
Else if the customer primarily purchases home goods:
  - Focus on seasonal home decor and new arrivals
  - Include upcoming sales in relevant categories
Else:
  - Provide recommendations based on current bestsellers
  - Include a first-time purchase discount

Chained Reasoning

For complex analytical tasks, guide the agent through sequential thinking:

Examine the attached quarterly financial report and provide insights:

1. First, identify the key revenue drivers and how they changed from last quarter
2. Then, analyze the expense categories that showed the most significant variations
3. Next, calculate the impact of these changes on overall profitability
4. Based on these findings, suggest potential areas for improvement
5. Finally, summarize the 2-3 most important action items for leadership

Best Practices

1. Be Specific: Clearly articulate expected outputs and formats
2. Provide Context: Include relevant background information
3. Use Examples: When appropriate, provide examples of desired outputs
4. Layer Complexity: Start with basic requests before adding complexity
5. Test & Refine: Iteratively improve prompts based on performance
6. Documentation: Maintain a library of effective prompts for reuse

Common Pitfalls

1. Vague Instructions: Avoid ambiguous language that could be interpreted in multiple ways
2. Contradictory Directives: Ensure all parts of your prompt align with the same goal
3. Over-Constraining: Provide guidance without unnecessarily limiting creative solutions
4. Ignoring Tool Capabilities: Always craft prompts with awareness of available tools
5. Missing Validation: Include criteria for what constitutes a successful response

Advanced Prompting Techniques

Zero-shot Prompting

Zero-shot prompting instructs the AI without providing examples, relying on the model's pre-trained knowledge:

Classify the following customer feedback into one of these categories: Product Quality, User Experience, Customer Service, or Price.

Feedback: "I found the checkout process confusing and gave up halfway through my purchase."

This prompt works because it:

• Clearly defines the task (classification)
• Provides specific category options
• Presents clear input content
• Doesn't require examples for the model to understand

Few-shot Prompting

Few-shot prompting provides examples to guide the model's responses:

Classify the following customer emails by department:

Email: "My recent order #45789 hasn't arrived yet. It's been 7 days."
Department: Shipping

Email: "I need to change my subscription plan from Basic to Premium."
Department: Billing

Email: "The mobile app keeps crashing when I try to upload photos."
Department: Technical Support

Email: "I received the wrong product in my order #78123."
Department: 

This approach:

• Demonstrates the expected pattern through examples
• Allows the model to recognize patterns across similar inputs
• Reduces ambiguity in how to format responses
• Works well for classification, tagging, and formatting tasks

Chain-of-Thought Prompting

Chain-of-Thought prompting guides the model through explicit reasoning steps:

Calculate the profit margin for product X based on the following information:
- Manufacturing cost: $45 per unit
- Packaging cost: $5 per unit
- Marketing allocation: $10 per unit
- Shipping cost: $15 per unit
- Retail price: $120 per unit

Think through this step-by-step:
1. First, calculate the total cost per unit by adding all expenses
2. Then, determine the gross profit by subtracting total cost from retail price
3. Finally, calculate the profit margin percentage by dividing gross profit by retail price and multiplying by 100

Benefits include:

• Breaking complex problems into manageable steps
• Improving accuracy on mathematical or logical tasks
• Creating transparent reasoning paths
• Allowing intervention at specific reasoning stages

Retrieval Augmented Generation (RAG)

RAG combines the model's capabilities with retrieved information:

Using the customer data profile and previous purchase history provided below, generate personalized product recommendations:

Customer Profile:
Name: Alex Thompson
Age: 34
Interests: Outdoor activities, photography, technology
Previous Purchases:
- Waterproof camera case (3 months ago)
- Hiking boots (1 month ago)
- 128GB SD card (2 weeks ago)

Based on this specific customer information, provide 3 personalized product recommendations with brief explanations of why each product would appeal to this customer.

This technique:

• Enhances responses with specific retrieved information
• Produces more relevant and personalized outputs
• Grounds recommendations in actual data
• Combines factual information with generative capabilities

By applying these prompting techniques, you can maximize the effectiveness of Brightsy AI's tools and create powerful, automated workflows that deliver consistent, high-quality results.

Additional Resources

For comprehensive information on prompt engineering techniques and best practices, visit promptingguide.ai - an excellent resource covering everything from basic prompting to advanced methods like Chain-of-Thought, Tree of Thoughts, and ReAct.