A support assistant that reads tickets before your team does
Tickets are sorted by topic and urgency, with a drafted reply ready to approve.
- Client
- Online retailer
- Timeline
- 6 weeks
- Scope
- AI classification, reply drafting
The problem
Refund requests, shipping questions and urgent complaints all landed in one inbox. Replies were slow, and the urgent ones were easy to miss.
What we built
- Each ticket is labeled by topic and priority as soon as it arrives.
- The reply is drafted from the customer's order data and the store's policy documents.
- A person approves or edits every reply, so nothing is sent unseen.
- Low-confidence drafts are flagged, so staff know where to look closely.
6 hrs → 40 minfirst response time
71%drafts sent with little or no edits
100%replies approved by a person
Built with LLM API, Python, helpdesk API.