Overview
We designed and shipped a production retrieval-augmented AI assistant for a streaming provider — grounded in subscriber, billing, and content data — that resolves the majority of tier-1 support queries autonomously, with evaluation, guardrails, and cost controls built in.
The challenge
The OTT provider’s support costs were ballooning as the subscriber base scaled, and customers were frustrated by slow, repetitive tier-1 responses. Off-the-shelf chatbots hallucinated and could not access real account context.
Our approach
We built a RAG pipeline over the provider’s knowledge base and account systems, with a vector store for retrieval, strict grounding guardrails, and an evaluation harness to measure accuracy before every release. Human handoff was designed in for the cases the assistant should not handle.
Outcomes
- 62% of tier-1 tickets deflected
- 4.6/5 customer satisfaction on AI chats
- $4.2M estimated annual savings
- Faster resolution for subscribers