◎ CASE / 001◎ AI AGENT◎ KR◎ 2024● LIVE
Korean SaaS Co.
CS Support AI Agent
CS Support AI Agent
24/7 automated response for 200+ daily inquiries. Humans handle escalation only.
ROLE
AI Agent Design · Build
PERIOD
2024.06 — 2024.08 (8 weeks)
MARKET
KR
HEADLINE
-80% · CS response time
(01) Challenge · The problem
“Starting point”
Over 200 customer support tickets came in daily, but 60% were repeat questions (delivery delays, refund policy, product specs). Three agents spent 80% of their day on this repetitive work, while complex issues piled up in queue.
(02) Solution · How we solved it
“Approach”
We built a Claude API agent orchestrated with LangGraph: inquiry → classification → RAG answer → confidence check → customer reply or human escalation. FAQs, policy docs, and past tickets are indexed in PGVector, so answers stay grounded in real sources. Agents only see what gets escalated.
(04) Outcomes · Results
Proven by numbers.
-80%
CS response time
Repeat inquiries auto-resolved; agents handle only complex cases
24/7
Unattended operation
Always on, including weekends and holidays
94%
Auto-resolution rate
AI closes the ticket in one turn without escalation
<30s
Average response time
Includes RAG retrieval and answer generation
(05) Stack · Tools used
Trusted tools only.
01 · AI
- ▸Claude 3.5 Sonnet
- ▸LangGraph
- ▸PGVector
- ▸Embedding (OpenAI)
02 · Backend
- ▸Python
- ▸FastAPI
- ▸Supabase
- ▸PostgreSQL
03 · Integrations
- ▸ChannelTalk
- ▸Slack (escalation)
- ▸Notion (FAQ source)
04 · Monitoring
- ▸Langfuse
- ▸Sentry
- ▸Custom dashboard
▸ Related · AI AGENTALL PROJECTS →
START / FREE CONSULT · NDA OK