Hybrid AI Contact Center Analytics Engine

Contact centers generate two kinds of data that don't naturally connect. There's the quantitative side — queue length, call times, escalation counts, buried in Contact Trace Records — and the qualitative side — what customers and agents actually said, locked inside transcripts and Contact Lens summaries. Most operational questions require both, and no single tool spans them.

A question-answering system that takes a natural-language question — "Are there specific agents who handle escalations better than others based on customer satisfaction?" or "Which queues are consistently taking the longest to resolve, and why?" — figures out which parts require quantitative data, which require qualitative, pulls from the right source for each, and assembles a unified answer.

1. Classify — An LLM parses the question and determines which parts are quantitative, qualitative, or neither.
2. Query the numbers — For quantitative parts, it generates Athena queries against Connect CTR data.
3. Retrieve the context — For qualitative parts, it runs RAG over call transcripts and Contact Lens summaries using Titan embeddings.
4. Assemble — Nova Pro combines the structured metrics and retrieved call context into a single natural-language answer.

Wrap it in an API, add a QnA mode that helps callers self-serve while waiting in queue, and move it from a local prototype to a deployed serverless footprint.