AI support assistant cut first-response time by 78%.

The challenge

The client's six-person support team was drowning. Ticket volume had doubled in a year, first responses were taking around 14 hours, and CSAT was sliding. Agents spent most of their day answering the same forty questions about orders, returns, and product details — while genuinely complex cases sat in the queue.

Hiring more agents would have raised costs without fixing the underlying problem: almost two-thirds of tickets never needed a human at all.

The solution

We built a retrieval-augmented (RAG) AI assistant trained on the client's product catalog, policy documents, and order data — so every answer was grounded in their actual information, not generic model knowledge.

The assistant was embedded in the website and WhatsApp, connected to the order system for live status lookups, and wired into the helpdesk through n8n so unresolved conversations escalated to a human with full context. Guardrails, evaluation suites, and a human-review loop kept quality measurable from day one.

• RAG assistant grounded in catalog, policies, and live order data
• Web chat + WhatsApp channels with seamless human handoff
• n8n integration into the existing helpdesk and CRM
• Evaluation suite and guardrails monitored weekly