Identity & Access Management · Europe · Retrieval · ranking
Every support ticket arrives with the research done.
Engineers spent most of their response time searching the documentation before they could answer. Now every ticket arrives with the right pages already pinned beside it.
Sign-in fails after certificate rotation
Users see an invalid signature error since this morning's rotation.
- 60–80%
- Of response time was research, now done before the ticket is opened
- 4types
- Guide, configuration, troubleshooting, release note
A software vendor's support team handles deeply technical tickets across a broad product. Every ticket started with the same research: read the question, search the docs, find the right page, check it still applied. On technical tickets that research took 60 to 80% of the time to first response, and it weighed most on the newest engineers.
An AI agent now reads each new ticket, finds the most relevant pages in the product documentation, labels each one as a guide, a configuration reference, a troubleshooting page or a release note, and pins them beside the ticket. It drafts no answers: the engineer reads the references and writes the reply. The index rebuilds whenever the documentation changes, so the references stay current.
- 01 · What we foundAssess
Watched engineers handle real tickets. Research took 60 to 80% of the time to first response, and the documentation, not memory, had to be the source of every answer.
- 02 · What we builtBuild
An agent that reads each ticket, ranks the documentation pages that answer it, labels them by type and shows them inside the support tool. References only, never a generated reply.
- 03 · Where it standsLive
Every ticket opens with the research done. Engineers still write every answer, now with the right page in front of them.
