Orbit Support
A production AI agent embedded in Orbit's Zendesk. It reads the ticket, searches the docs and past resolutions, drafts a reply, and either sends it or hands off to a human with the context pre-written.
- Client
- Orbit
- Timeline
- 9 weeks
- Services
- AI AgentsIntegrations
- Stack
- AI Agents· LangChain· Zendesk· RAG
What we walked into.
Orbit's support team was drowning: 1,400 tickets a week, 18-hour first response, and three agents on the verge of quitting. They had tried an off-the-shelf AI plugin that made things worse by confidently answering wrong.
The approach.
- 01
Built a RAG layer over their docs, past tickets, and product changelog — cited sources in every draft reply.
- 02
Wrote a confidence router: high-confidence answers auto-send, medium-confidence go to human draft, low-confidence never touches the customer.
- 03
Set up a golden-set eval suite that runs every deploy so regressions get caught before production.
- 04
Added a cost dashboard so Orbit knows per-ticket spend and can cap it.
What actually changed.
The difference is the confidence router. It will not pretend to know. That is why our team trusts it — and why CSAT went up, not down.