Automatic Classification

How AI classifies and prioritizes claims

How AI Classifies Claims

Every claim is automatically analyzed and classified. You can review and override classifications.

What Gets Classified

• Category: Quality, Safety, Logistics, Maintenance, Other
• Priority: Critical, High, Medium, Low
• Root Cause: From 9 possible causes
• Sentiment: Positive, Neutral, Negative

Classification Process

1. Input Processing:

   • If text: Analyze directly
   • If voice: Transcribe first, then analyze
   • If image: Extract text via OCR, then analyze

2. Analysis:

   • Identify keywords and patterns
   • Match against historical data
   • Consider context and urgency
   • Predict likely causes

3. Scoring:

   • Calculate confidence level (0-100%)
   • Identify similar past claims
   • Rank possible categories
   • Assign most likely classification

4. Presentation:

   • Show AI suggestion
   • Display confidence score
   • Show alternative options
   • Allow user override

Confidence Scoring

90-100%: Very confident, usually correct

• Example: “Chemical spill” -> Safety, Critical, 98%

70-90%: Confident but verify

• Example: “Equipment noise” -> Maintenance, High, 82%

50-70%: Uncertain, review carefully

• Example: “Strange behavior” -> Could be multiple categories

Below 50%: Very uncertain, likely needs manual review

• Example: Ambiguous description -> Unknown category, 35%

Using AI Classifications

Accept Classification

• Click checkmark to accept
• Claim created with AI suggestion
• System learns from acceptance

Override Classification

• Click to change category/priority
• Select correct classification
• Your correction trains AI
• Next time, AI more likely to get it right

Provide Feedback

• Is classification reasonable? (like/dislike)
• User feedback improves AI
• High dislike rate triggers review

Improving Classifications

Feedback Loop

Each override helps train AI:

• User sees wrong classification
• Corrects it
• AI learns pattern
• Future similar claims classified better

Pattern Recognition

AI identifies patterns:

• “Conveyor” + keywords -> Maintenance
• “Injured” + urgency -> Safety
• “Delivery” + status -> Logistics

Historical Learning

AI improves by seeing:

• What was classified before
• How it was resolved
• Patterns in resolutions
• Seasonal trends

Common Classification Mistakes

Too Many “Critical” Classifications

Problem: AI classifies too many claims as Critical Solution: Adjust confidence threshold higher or review AI keywords for “urgent”

Missing Safety Issues

Problem: Safety claims classified as Quality Solution: Add safety keywords to knowledge base or adjust system prompt

Wrong Department Routing

Problem: Maintenance claims routing to Quality Solution: Review category-to-department mapping

AI Confidence for Different Inputs

Text claims: Usually 75-90% confidence

• Clear descriptions: Higher confidence
• Vague descriptions: Lower confidence

Voice claims: Usually 70-85% confidence

• Clear speech: Better confidence
• Background noise: Lower confidence
• Accents: Varies by model

Image claims: Usually 60-80% confidence

• Clear images: Better confidence
• Blurry images: Lower confidence
• Requires context text

---

Next Steps

• Agent Configuration
• Chat Assistant