GuidesAI automation
Tutorial: triaging support tickets with AI
1. The forced schema
const tool = {
name: 'classify_ticket',
description: 'Classifies a support ticket.',
input_schema: {
type: 'object',
properties: {
category: { type: 'string', enum: ['billing', 'delivery', 'technical', 'account', 'other'] },
urgency: { type: 'string', enum: ['low', 'normal', 'high'] },
language: { type: 'string', description: 'ISO 639-1 code, e.g. en' },
summary: { type: 'string', description: 'one sentence' },
},
required: ['category', 'urgency', 'language', 'summary'],
},
};
2. The call, when the ticket is created
const res = await client.messages.create({
model: process.env.CLAUDE_MODEL!,
max_tokens: 300,
system: 'You triage support tickets. High urgency: customer blocked or payment failing.',
tools: [tool],
tool_choice: { type: 'tool', name: 'classify_ticket' },
messages: [{ role: 'user', content: `${ticket.subject}\n\n${ticket.message}` }],
});
const triage = res.content.find((b) => b.type === 'tool_use')?.input;
await helpdesk.tag(ticket.id, triage);
3. Routing
| Result | Queue |
|---|---|
| High urgency | Priority queue, immediate notification |
| Billing | Accounting team |
| Technical | Technical support |
| Foreign language | An agent who speaks it |
4. Measuring
- 1SampleTwo hundred tickets classified by hand.
- 2ComparisonAgreement rate by category and urgency.
- 3AdjustmentInstructions sharpened on frequent mistakes.
The basics: calling the Claude API from Node.js and tool use explained. The principle: what a webhook is.
Frequently asked questions
- Which support tool?
- Zendesk, Freshdesk, Crisp, HubSpot or an in-house tool: most send a webhook when a ticket is created and accept tags through their API.
- What if it picks the wrong category?
- The agent fixes it in one click. Those corrections become the test set for improving the instructions.
- Can it answer directly?
- That is a further step. Start with triage: it carries no risk for the customer and saves time right away.