Knowledge Base Optimization
Optimize your knowledge base: measure answer quality, find gaps, prioritize sources, and reduce hallucinations.
A well-optimized knowledge base is the difference between an agent that delights callers and one that frustrates them. Here's how to continuously improve your knowledge base quality.
Testing Retrieval Quality
Regularly test your knowledge base by asking questions and reviewing which documents/chunks are retrieved:
- Go to your agent's Knowledge Base tab.
- Use the "Test Query" feature.
- Enter a question a caller might ask.
- Review the retrieved results and their relevance scores.
Common Issues
- Too vague answers — Your documents may lack specific details. Add more focused content.
- Wrong documents retrieved — Content may be ambiguous. Add FAQ pairs for precise control.
- No answer found — The information isn't in your knowledge base. Add the missing content.
- Outdated answers — Documents haven't been updated. Re-upload or re-crawl.
Improving Results
- Better document formatting — Use clear headings, bullet points, and short paragraphs.
- More specific FAQ pairs — Add exact Q&A pairs for frequently asked questions.
- Remove duplicate content — Duplicate information confuses the retrieval system.
- Use natural language — Write content the way your customers would ask about it.
Advanced Tuning
For power users, BHOMY offers advanced settings:
- Chunk size — Adjust how documents are split into searchable chunks.
- Relevance threshold — Set the minimum confidence score for retrieved results.
- Max results — Control how many knowledge base results are provided to the AI.
Knowledge Gap Analysis
Review call transcripts to identify questions your agent struggled with. Look for patterns where the agent said "I don't have that information" or gave generic responses. Add the missing information to your knowledge base to fill these gaps.
Review call transcripts regularly to find questions your agent struggled with, then add that information to your knowledge base. This creates a continuous improvement loop.
Feedback
Was this article helpful?