Classification

Supp classifies customer messages into 315 intents across 13 categories using a custom-trained ML model.

How It Works

  1. Customer sends a message
  2. The model returns the most likely intent + confidence score
  3. If confidence exceeds the threshold, the routing rule fires
  4. If confidence is too low, the message is escalated to a human

Classification latency is ~100-200ms.

Categories

CategoryIntentsExamples
No Intent8Off-topic, self-talk, humor, rambling
Account Management31Password reset, 2FA setup, account deletion
Billing & Payment31Refund request, invoice dispute, payment failed
Subscription Management21Upgrade, downgrade, pause, cancel
Product Inquiry31Availability, pricing, feature comparison
Technical Support47Bug report, performance issue, API troubleshooting
Order & Delivery23Order tracking, shipping delay, cancellation
Returns & Refunds24Return request, exchange, defective product
Feature Requests22New feature, UI improvement, integration request
Complaints & Feedback20Service complaint, negative experience
General Inquiry15Business hours, contact info
Verification & Auth19Identity verification, biometric setup
Loyalty & Rewards14Points balance, referral codes, cashback

Use supp_categories to list all categories, and supp_intents to browse intents within a category.

Confidence Thresholds

ThresholdDefaultWhat Happens
Escalation floor0.35Below this → always escalate to human
Auto-success0.70Above this → charge immediately
Routing rule0.80Per-rule minimum to trigger the action

You can customize the routing rule threshold per intent.

MCP Tools

ToolCostDescription
supp_classify$0.20Classify a message
supp_classify_with_priority$0.23Classify + score priority
supp_intentsFreeList all 315 intents
supp_categoriesFreeList the 13 categories

Priority Scoring

Priority scoring runs alongside classification and returns an urgency level:

LevelDescription
lowRoutine questions
mediumStandard support requests
highUrgent issues
criticalOutages, security incidents, data loss

Use supp_priority_score ($0.03) to score a message, or pass include_priority: true with classification.

Batch Classification

For bulk operations through the API, batch mode provides a 50% discount per classification.

Classification Feedback

When customers flag an incorrect classification via the widget, that data is collected and used to improve the model over time. Enable/disable in your dashboard settings.