US B2B SaaS Cut L1 Support Volume 64% with RAG-Grounded Chatbot

The Challenge

Support ticket volume was growing 18% QoQ but the support team could not hire fast enough. 60%+ of L1 tickets were repetitive questions answered in the knowledge base, but customers preferred chatting to searching.

Our Solution

Iedeo built an in-product AI chatbot grounded in the customer's help docs, API reference, changelog and resolved-ticket history. Conversations escalate to live agents with full context when confidence drops or the customer asks. Deployed via embedded widget + Slack integration.

Architecture & Stack

• Embedded React widget with branded theme
• Anthropic Claude 3.5 Sonnet via AWS Bedrock (US-east-1)
• Hybrid retrieval: pgvector + BM25, reranked with Cohere rerank-3
• Knowledge sources: Notion docs + Intercom resolved tickets + GitHub changelog + API reference
• Live agent handoff via Intercom with full conversation transcript
• Eval suite of 247 scenarios run on every prompt change
• Observability: LangFuse for trace inspection

Timeline

2 weeks discovery → 5 weeks build → 2 weeks parallel-run pilot → cutover

“We had quotes from US firms at $180K. Iedeo shipped in 7 weeks for $42K, with the same engineering rigour and better observability tooling than what the bigger vendors offered.”