How to Build an AI Ticket Classification System

An ai support ticket classification and routing system eliminates the manual triage step that slows down every support queue. I build these for teams processing 50+ tickets daily where misrouted tickets add hours of delay. The AI reads the ticket, classifies it by type, and routes it to the right team instantly.

The accuracy gets better over time as you feed corrections back into the system.

What You Need Before Starting

• A ticketing system with API access (Zendesk, Freshdesk, or custom)
• Python 3.8+ with the Anthropic SDK
• Historical ticket data for testing accuracy
• Defined categories and team routing rules

Step 1: Define Your Classification Categories

Map every ticket type to a team:

CATEGORIES = {
    "billing": {"team": "finance", "priority": "medium", "sla_hours": 24},
    "technical": {"team": "engineering", "priority": "high", "sla_hours": 4},
    "shipping": {"team": "logistics", "priority": "medium", "sla_hours": 12},
    "returns": {"team": "support", "priority": "medium", "sla_hours": 24},
    "account_access": {"team": "security", "priority": "high", "sla_hours": 2},
    "feature_request": {"team": "product", "priority": "low", "sla_hours": 72},
    "general": {"team": "support", "priority": "low", "sla_hours": 48},
}

Step 2: Build the Classifier

import anthropic
import json

client = anthropic.Anthropic()

def classify_ticket(subject, body):
    category_list = ", ".join(CATEGORIES.keys())

    response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=100,
        messages=[{
            "role": "user",
            "content": f"""Classify this support ticket into ONE of these categories: {category_list}

Subject: {subject}
Body: {body}

Respond with ONLY a JSON object: {{"category": "category_name", "confidence": 0.95}}"""
        }]
    )

    result = json.loads(response.content[0].text)
    return result

Step 3: Route Based on Classification

def route_ticket(ticket_id, subject, body):
    classification = classify_ticket(subject, body)
    category = classification["category"]
    confidence = classification["confidence"]

    routing = CATEGORIES.get(category, CATEGORIES["general"])

    if confidence < 0.7:
        routing["team"] = "triage"
        routing["note"] = f"Low confidence ({confidence}). Manual review needed."

    assign_ticket(ticket_id, routing["team"], routing["priority"], routing["sla_hours"])
    log_classification(ticket_id, category, confidence)

    return routing

Step 4: Process Incoming Tickets via Webhook

from flask import Flask, request

app = Flask(__name__)

@app.route("/ticket-webhook", methods=["POST"])
def new_ticket():
    data = request.json
    ticket_id = data["ticket_id"]
    subject = data["subject"]
    body = data["body"]

    routing = route_ticket(ticket_id, subject, body)

    return {"ticket_id": ticket_id, "routed_to": routing["team"], "category": routing}

Step 5: Build a Feedback Loop

When agents reclassify tickets, log the correction to improve accuracy:

def log_correction(ticket_id, ai_category, correct_category):
    conn = sqlite3.connect("classifications.db")
    conn.execute("""
        INSERT INTO corrections (ticket_id, predicted, actual, corrected_at)
        VALUES (?, ?, ?, datetime('now'))
    """, (ticket_id, ai_category, correct_category))
    conn.commit()

def get_accuracy_report():
    conn = sqlite3.connect("classifications.db")
    total = conn.execute("SELECT COUNT(*) FROM classifications").fetchone()[0]
    correct = conn.execute(
        "SELECT COUNT(*) FROM classifications c LEFT JOIN corrections co ON c.ticket_id = co.ticket_id WHERE co.ticket_id IS NULL"
    ).fetchone()[0]
    return {"total": total, "correct": correct, "accuracy": correct / total if total else 0}

What to Build Next

Add sub-classification. Once a ticket hits the billing team, further classify it into "charge dispute," "refund request," or "invoice question" so agents can prioritize within their queue.

Related Reading

• AI in Customer Service - how classification fits into the support automation stack
• The Pipeline Architecture - the pipeline pattern that powers ticket routing
• The Trust Framework for AI Decisions - building trust in automated classification