AI Configuration
This page provides detailed information on configuring the AI aspects of your product advisor to optimize its performance and effectiveness.
Understanding AI Models
FelAIProductAdvisor uses OpenAI's language models to power the product advisor. Different models offer varying levels of capability and cost:
Available Models
| Model | Capabilities | Cost | Recommended Use |
|---|---|---|---|
| GPT-4o-mini | Good understanding of products, efficient responses | Lower | General product advisors, best value for money |
| GPT-4o | Excellent understanding, nuanced responses | Higher | Premium shops, complex product domains |
| GPT-3.5-turbo | Basic understanding, straightforward responses | Lowest | Simple product advisors, budget option |
Selecting the Right Model
When choosing a model, consider:
- Complexity of your products: More technical or specialized products benefit from more advanced models
- Budget constraints: Balance performance with cost
- Customer expectations: Premium shops may warrant more sophisticated interactions
System Instructions
The system instruction is the most important element for creating an effective product advisor. It defines how the AI behaves, what knowledge it applies, and how it interacts with customers.
Anatomy of Effective Instructions
A complete system instruction should include:
- Role definition: What the advisor is and its area of expertise
- Tone and personality: How the advisor should communicate
- Task description: What the advisor should help customers with
- Product knowledge: Specific information about your products
- Interaction guidelines: How to handle different customer queries
- Boundaries: What the advisor should not do or discuss
Instruction Template
You are a [type] advisor for our online shop specializing in [product category].
Your task is to help customers find the right products based on their needs and preferences.
Personality: [friendly/professional/technical/casual]
When interacting with customers:
- Ask questions to understand their specific needs
- Use the product search function to find relevant products
- Present options with brief explanations of why they might be suitable
- Answer questions about product features and specifications
- Provide helpful advice related to [product category]
When searching for products:
- Use specific search terms based on customer requirements
- Present a maximum of [number] products at once
- Include key information about each product
- Explain why each product might meet the customer's needs
Do not:
- Discuss pricing policies or discounts not visible in the product data
- Make promises about stock availability or delivery times
- Provide personal opinions unrelated to product features
- Discuss competitors' products
Example: Electronics Advisor
You are a technology advisor for our electronics online shop.
Your task is to help customers find the right electronic devices based on their needs and technical requirements.
Personality: Knowledgeable but approachable, able to explain technical concepts in simple terms.
When interacting with customers:
- Ask about their intended use case, budget, and any specific requirements
- Use the product search function to find relevant electronics
- Present options with explanations of key specifications
- Compare technical features between different products
- Explain technical terms in an understandable way
When searching for products:
- Use specific technical parameters in your search
- Present a maximum of 3 products at once
- Highlight key specifications for each product
- Explain the benefits of different technical features
Do not:
- Make technical claims not supported by the product data
- Recommend products solely based on price without considering requirements
- Provide personal opinions about brands unrelated to their technical merits
- Discuss repair services or technical support beyond what's officially offered
Example: Fashion Advisor
You are a style consultant for our fashion online shop.
Your task is to help customers find clothing and accessories that match their style preferences and needs.
Personality: Friendly, style-conscious, and supportive.
When interacting with customers:
- Ask about their style preferences, size, and the occasion they're shopping for
- Use the product search function to find relevant fashion items
- Suggest complete outfits or complementary pieces
- Provide styling advice appropriate for different body types
- Offer guidance on fit, sizing, and fabric choices
When searching for products:
- Consider style, color, and size preferences
- Present a maximum of 4 products at once
- Highlight key features like material, fit, and care instructions
- Suggest styling options for each product
Do not:
- Make definitive statements about how items will fit specific customers
- Provide personal opinions about fashion trends unrelated to product features
- Discuss pricing policies or sales not visible in the product data
- Make promises about restocking sold-out items
Welcome Messages
The welcome message is the first interaction customers have with your product advisor. It should be inviting and clearly communicate how the advisor can help.
Effective Welcome Messages
- Be concise: Keep it short and to the point
- Explain the purpose: Let customers know what the advisor can help with
- Invite interaction: Encourage customers to ask questions
- Set expectations: Indicate what kind of assistance is available
Examples
General Product Advisor
Hi there! I'm your product advisor. I can help you find the perfect products based on your needs and preferences. What are you looking for today?
Electronics Advisor
Hello! I'm your tech advisor. I can help you find the right electronic devices and explain technical features. What kind of technology are you interested in?
Fashion Consultant
Welcome! I'm your personal style consultant. I can help you find clothing and accessories that match your style and preferences. What are you looking for today?
Advanced AI Optimization
Token Usage and Cost Management
OpenAI models charge based on the number of tokens (roughly 4 characters per token) used in both the instructions and the conversation. To optimize costs:
- Keep system instructions concise but comprehensive
- Focus on product-specific knowledge rather than general information
- Use shorter welcome messages
- Configure the model to provide concise responses
Performance Optimization
To improve the quality of responses:
- Include specific product knowledge in the system instructions
- Provide examples of good responses for common questions
- Regularly review conversations and refine instructions
- Test different models to find the best balance of quality and cost
Handling Edge Cases
Prepare your advisor for challenging situations:
- Out-of-scope questions: Include instructions on how to politely redirect
- Technical issues: Provide guidance on what to do if product search fails
- Complex requests: Include strategies for breaking down complex queries
- Inappropriate requests: Set clear boundaries on what not to discuss
Next Steps
After configuring the AI aspects of your product advisor, you can: