Support
First-line support triage
Classify incoming requests, answer what is already documented through LLM-backed retrieval, and route the rest to the right team.
- Tickets closed at first line
- 61%Tickets closed at first line
- First response
- 8 minFirst response
Context
Three people handled all support for a SaaS product. Half the tickets covered ground the documentation already answered.
Workflow
- 01
Classify
Every ticket gets a topic, an urgency level and a language.
- 02
Retrieve
Vector search over the documentation and resolved tickets, then model reranking to keep only the passages that actually apply.
- 03
Draft
The model drafts a reply grounded in cited sources, in the team's voice. The agent approves or edits before it goes out.
- 04
Route
Anything out of scope goes to the right team with the context already gathered.
Stack
- n8n
- Claude Opus
- Embeddings
- Zendesk
- pgvector