Retail · LATAM
AI Support Agent
- Client
- Confidential
- Year
- 2025
- Services
- AI agents · RAG · Integrations · Evals & monitoring
Overview
A LATAM retail operator was drowning in repetitive support tickets — order status, returns, shipping — with response times stretching into hours. We designed and built an AI support agent that answers in seconds, grounded in their own catalog, policies and order system, and that knows when to hand off to a human.
The challenge
Generic chatbots hallucinate and frustrate customers. The agent had to give accurate, on-policy answers using live order data — and escalate gracefully instead of guessing.
Our approach
We built a RAG pipeline over their documentation, FAQs and order database, wrapped in an agent workflow that classifies intent, retrieves grounded context, drafts a reply with citations, and hands off to a human when confidence is low. Evals and monitoring keep quality measurable. The result: 68% of tier-1 tickets automated, response time down to ~1.2s, CSAT at 4.7/5 and cost per ticket reduced by 54%.
- AI
- Agents
- RAG
- Automation
Have a project like this?