Case study 06 · Customer support
Inquiry Response AI — Grounded drafts with review
Drafts inquiry replies and screens banned phrases
In operation · Knowledge base integration
- Industry
- Customer support
- Scope
- AI system build · operations & monitoring
- Technology
- Admin console · knowledge base · review agent · prohibited-term checks
- Status
- In operation · Knowledge base integration
- Client
- Undisclosed — contracts and acceptance records can be shown under NDA
- The repeated work
- Similar answers were rewritten from scratch each time, and replies were slow whenever the team was busy. Rushed answers sometimes contained things that should not have been said.
- How it works now
- Draft replies are produced from a knowledge base and go through a review stage that shows its reasoning. Prohibited wording is caught before anything is sent.
- How it runs
- A response simulator checks replies before anything actually goes out, and the reasoning behind each judgment is laid out on one screen, so you can see why it answered that way.
How the flow changed
- Before
- Inquiry→write from scratch→send
- After
- Draft→evidence check→send
What we built
- Admin console
- knowledge base
- review agent
- prohibited-term checks
Have similar work?
Tell us about one task you do today and we start with a diagnosis. The diagnosis report is free.
If you need to verify the contracts and the acceptance, we put an NDA in place and show you the contract list and the acceptance records directly in a consultation.