The maximum number of results determines how many content fragments from your knowledge base are provided to the artificial intelligence to formulate each response. It is a crucial parameter for balancing response quality and system efficiency.
How the results system works
When a user asks a question, the system:
- Analyzes the question: Identifies keywords and context
- Searches the knowledge base: Finds all potentially relevant content
- Orders by relevance: Ranks results from most to least pertinent
- Limits results: Takes only the first N results (where N is your parameter)
- Provides to AI: The artificial intelligence receives this content to create the response
Impact of number of results
Few results (3-7)
Advantages:
- More focused and direct responses
- Lower token consumption (reduced costs)
- Faster response times
- Less probability of contradictory information
Disadvantages:
- Might miss relevant information
- Less complete responses
- Limitations for complex questions
Ideal for:
- FAQ with direct answers
- Specific and punctual information
- Technical support with standard procedures
Many results (15-25)
Advantages:
- More complete and detailed responses
- Greater coverage of related information
- Better handling of complex questions
- More context for AI
Disadvantages:
- Higher token consumption (higher costs)
- Longer response times
- Possible redundant or contradictory information
- Potentially less focused responses
Ideal for:
- Consulting and advisory
- In-depth questions
- Research and analysis
How to choose the optimal number
Consider business type
- E-commerce: 8-12 results to balance product details and performance
- Technical support: 5-10 results for precise answers
- Professional services: 12-20 results for in-depth consulting
- Education/Training: 15-25 results for complete explanations
Consider question complexity
- Simple questions ("What are the hours?"): 3-5 results
- Moderate questions ("How to configure the product?"): 8-12 results
- Complex questions ("What are best practices for...?"): 15-20 results
Consider available resources
- Limited budget: Lower number to contain costs
- Critical performance: Moderate number for speed
- Quality priority: Higher number for completeness
Testing and optimization
Testing methodology
- Baseline: Start with 10 results (standard value)
- Prepare test questions: 10-15 questions representative of your use case
- Test variations: Try 5, 10, 15, 20 results
- Evaluate quality: Compare completeness and accuracy of responses
- Monitor performance: Check response times and costs
Metrics to evaluate
- Completeness: Does the response cover all aspects of the question?
- Accuracy: Is the information correct and up-to-date?
- Relevance: Is all provided content pertinent?
- Readability: Is the response well-structured and understandable?
Practical examples
Case: Software product support
Configuration: 8 results
Question: "How do I reset my account password?"
Result: Precise response with step-by-step procedure, without superfluous information
Case: Legal consulting
Configuration: 18 results
Question: "What are the tax obligations for a new business?"
Result: Complete response covering multiple areas (registrations, codes, deadlines, etc.)
Seasonal adaptations
- High activity periods: Reduce number to improve performance
- New product launches: Temporarily increase for complete coverage
- Knowledge base updates: Retest optimal after major changes
Best practices
- Monitor constantly: Regularly check effectiveness of chosen parameter
- Adapt to feedback: Modify based on user comments
- Document changes: Keep track of modifications and their reasons
- Periodic testing: Repeat tests when adding new content