AI Ticket Triage

Intelligent support ticket routing

Support teams lose real time just reading and categorizing tickets before anyone can act on them. unoblox extracts urgency, category, and the right team from raw ticket text — through one OpenAI-compatible endpoint, billed in rupees.

Why AI triage

Tickets land in plain text from email, chat widgets, and web forms. A language model reads that text once and returns structured routing data in milliseconds — no manual reading queue before work can start.

Top models for triage

Model	Input ₹/1M	Output ₹/1M	Advantage
Qwen3 1.7B	₹0	₹0	Free, fast, sufficient for most categories
GPT-4.1	₹201.6	₹806.4	High accuracy on ambiguous or urgent tickets
Claude Sonnet	₹201.6	₹1008	Reads tone as well as content

Prices per 1M tokens.

Triage implementation

from openai import OpenAI

client = OpenAI(
    api_key="ub-gw-...",
    base_url="https://api.unoblox.ai/v1"
)

ticket = {"subject": "Website down, urgent", "body": "Our site is showing 500 errors for all users."}

prompt = "Triage this support ticket: " + str(ticket) + "\n\nReturn JSON: {category: technical|billing|feature-request|account, priority: critical|high|medium|low, team: engineering|finance|product|success, summary: one-line summary}"

response = client.chat.completions.create(
    model="openai/gpt-4-1",
    messages=[{"role": "user", "content": prompt}],
    response_format={"type": "json_object"}
)

print(response.choices[0].message.content)

A routing table that maps priority to response time

Priority	Example	Target first response
Critical	Full outage, payment failure	Immediate, page on-call
High	Feature broken for many users	Within 1 hour
Medium	Feature broken for one user	Within 1 business day
Low	Question, minor cosmetic issue	Within 2–3 business days

Route "critical" straight to a paging system, not just a ticket queue — triage only helps if the output actually changes what happens next.

Handling low-confidence calls

Ask the model to return a confidence score alongside its classification. Anything under your chosen threshold (many teams start around 0.7–0.8) goes to a human for a first look instead of auto-routing — this catches ambiguous tickets before they land on the wrong team.

Common use cases

• SaaS customer support routing
• E-commerce helpdesk triage
• Banking and fintech inquiry categorization
• Insurance claim intake processing
• Internal HR/IT ticketing

Frequently asked questions

Q: Can I use the free Qwen3 1.7B for triage? A: Yes — for most category and priority labels it performs well and costs nothing up to the freemium cap. Reserve GPT-4.1 or Claude Sonnet for tickets needing finer judgment.

Q: How do I prevent misrouting? A: Start with a confidence threshold — anything below it goes to a human reviewer instead of auto-routing. Refine your category taxonomy as you collect more labelled examples.

Q: Does triage work with multilingual tickets? A: Yes — Claude Sonnet and GPT-4.1 both handle Hindi, Marathi, Tamil, Telugu, and other regional languages without a separate translation step.

Q: Can I detect urgency from tone, not just keywords? A: Yes — ask the model to assess emotional tone alongside category and priority, and route highly frustrated tickets to senior agents regardless of stated urgency.

Q: How do I integrate with Jira, Zendesk, or Freshdesk? A: Most ticketing tools fire a webhook on ticket creation. Call unoblox from that webhook, then use the response to set tags, priority, and assignee automatically.

Q: What happens if the model gets the category wrong? A: Log corrections your team makes and periodically review them — patterns in what gets corrected tell you where to tighten the prompt or add a missing category.

Get started in rupees → https://unoblox.ai/sign-in