Skip to main contentAI Configuration Overview
Configure how AI behaves for your organization.
LLM Provider Selection
Choose which AI model to use:
OpenRouter (Default - Cost-Effective)
- Model: Nemotron Nano 12B V2 VL
- Cost: Free tier available, pay-per-use
- Speed: Fast processing
- Quality: Good for classification
- When to use: Most claims (budget-conscious)
Groq (Audio Processing)
- Model: Whisper Large V3 Turbo
- Cost: Pay-per-use
- Speed: Real-time transcription
- Quality: Excellent audio
- When to use: All voice claims
OpenAI (Premium)
- Model: GPT-4, GPT-4 Vision
- Cost: Premium pricing
- Speed: Varies
- Quality: Best accuracy
- When to use: Critical classifications, complex analysis
Configuration Settings
Classification Confidence Threshold
Set minimum confidence for AI suggestions:
-
High threshold (>80%): Only very confident suggestions
- Pro: Fewer wrong classifications
- Con: More claims need manual review
-
Medium threshold (>60%): Balanced approach
- Pro: Good automation, reasonable accuracy
- Con: Some errors possible
-
Low threshold (>40%): Aggressive automation
- Pro: Minimal manual work
- Con: More corrections needed
Priority Assignment Rules
Configure how AI assigns priority:
- Critical: Safety concerns, urgent keywords
- High: Major issues, equipment down
- Medium: Standard issues
- Low: Routine items
Root Cause Categories
Define the root cause options:
- Equipment failure
- Training gap
- Process error
- Supplier issue
- Material defect
- Environmental
- Maintenance backlog
- Communication breakdown
- Human error
Testing AI Configuration
- Create sample claims with various inputs
- Review AI classifications
- Check confidence scores
- Note any pattern errors
- Adjust configuration if needed
- Retest until satisfied
Track:
- Classification accuracy
- Override rate (how often users change AI suggestion)
- Processing time
- Cost per claim
- User satisfaction
High override rate suggests:
- AI needs retraining
- Thresholds need adjustment
- Categories not clear
- Process needs change
Disabling AI Features
If AI not working well:
- Disable auto-classification
- Use manual categorization
- Revisit AI configuration
- Check sample claims
- Adjust thresholds
Next Steps
