Classification
Supp classifies customer messages into 315 intents across 13 categories using a custom-trained ML model.
How It Works
- Customer sends a message
- The model returns the most likely intent + confidence score
- If confidence exceeds the threshold, the routing rule fires
- If confidence is too low, the message is escalated to a human
Classification latency is ~100-200ms.
Categories
| Category | Intents | Examples |
|---|---|---|
| No Intent | 8 | Off-topic, self-talk, humor, rambling |
| Account Management | 31 | Password reset, 2FA setup, account deletion |
| Billing & Payment | 31 | Refund request, invoice dispute, payment failed |
| Subscription Management | 21 | Upgrade, downgrade, pause, cancel |
| Product Inquiry | 31 | Availability, pricing, feature comparison |
| Technical Support | 47 | Bug report, performance issue, API troubleshooting |
| Order & Delivery | 23 | Order tracking, shipping delay, cancellation |
| Returns & Refunds | 24 | Return request, exchange, defective product |
| Feature Requests | 22 | New feature, UI improvement, integration request |
| Complaints & Feedback | 20 | Service complaint, negative experience |
| General Inquiry | 15 | Business hours, contact info |
| Verification & Auth | 19 | Identity verification, biometric setup |
| Loyalty & Rewards | 14 | Points balance, referral codes, cashback |
Use supp_categories to list all categories, and supp_intents to browse intents within a category.
Confidence Thresholds
| Threshold | Default | What Happens |
|---|---|---|
| Escalation floor | 0.35 | Below this → always escalate to human |
| Auto-success | 0.70 | Above this → charge immediately |
| Routing rule | 0.80 | Per-rule minimum to trigger the action |
You can customize the routing rule threshold per intent.
MCP Tools
| Tool | Cost | Description |
|---|---|---|
supp_classify | $0.20 | Classify a message |
supp_classify_with_priority | $0.23 | Classify + score priority |
supp_intents | Free | List all 315 intents |
supp_categories | Free | List the 13 categories |
Priority Scoring
Priority scoring runs alongside classification and returns an urgency level:
| Level | Description |
|---|---|
low | Routine questions |
medium | Standard support requests |
high | Urgent issues |
critical | Outages, 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.