E-commerce · Public demo
24/7 email triage agent for e-commerce
Automated classification and first response on 2,000 daily customer care emails. Human team only intervenes on the most complex 25%.
Transparency note: The cases published here are technical implementation scenarios: internal PoCs, pilots with synthetic or anonymized data, and reference architectures. As real customer projects reach production, we replace them with named cases with permission.
The challenge
An e-commerce with 30,000 SKUs handled 2,000 daily customer care emails (order tracking, returns, size changes, complaints). Average response time was 8 hours, the team was burning out, and customers were frustrated.
Our solution
Multi-step agent built with LangGraph: reads every incoming email, classifies it by intent (tracking, return, complaint, general), retrieves order data from Shopify, composes a contextual response. Simple requests are answered directly. Complex ones get a draft assigned to the best-suited human operator.
Tech stack
Measured results
- 75% of emails resolved without human intervention
- First response time: from 8 hours to ~8 seconds
- Team time reduced 60%, focus on complex cases
- Customer Satisfaction stable above 4.2/5
- Cost per interaction reduced ~4x vs benchmark
- Demonstrated scalability to 10,000 emails/day in load tests
Lessons learned
- •Design the human routing first: agents learn from how humans handle complex cases
- •Persistent agent memory is decisive on multi-turn conversations
- •The team only trusts the agent when they know they can always override it