The challenge

A 150-seat SaaS company with 40,000 users was receiving 1,200 support tickets per week. Over 60% were answerable from documentation. A three-person support team was overwhelmed, average first response time was 4 hours, and CSAT had fallen to 3.6 stars. They couldn't hire fast enough to match user growth.

Our approach

We built a RAG assistant grounded in their product documentation, help articles, and historical ticket resolutions. The knowledge base was indexed nightly, with semantic search to find relevant answers. The system was integrated into their existing Intercom workflow — handling tier-1 queries automatically and escalating with context to human agents for tier-2.

The solution

The assistant resolves questions about product features, account management, and common error states using cited documentation sources. Unresolved queries escalate to the human queue with a pre-written summary of what was tried. A weekly report shows deflection rate, CSAT by resolution path, and knowledge gaps driving escalations.

Results

First-week deflection rate of 38%, reaching 45% by week four as the knowledge base was refined. Average first response time dropped from 4 hours to under 2 minutes for auto-resolved tickets. CSAT rose from 3.6 to 4.7 stars. The support team now handles complex, high-value issues instead of FAQs.

"The bot handles FAQs better than half my team used to. My support engineers are finally doing support engineering, not answering the same questions."

James Okafor

Head of Customer Success, SaaS Co.