Delight Analytics: Find the revenue
opportunities in customer conversations

Delight.ai Analytics shows what customers ask about, what stops them from buying, and how your AI agent handles those conversations. It includes a dashboard for buying signals and core metrics, review queues for flagged messages and potential hallucinations, conversation transcripts, data exports, and Delight Agent MCP for questions the dashboard doesn't answer.

Start with buying signals that keep coming up, such as restock requests, product comparisons, or bundle questions, then check how your AI agent answered them. With Delight Agent MCP, you can ask which of those conversations ended without a recommendation or next step, and fix the gap. See how Delight Agent MCP works.

Export conversation-level data with the Data Export API and join it to your order or booking data in your BI tool. You can then report conversion and order value for the conversations your AI agent handled, using the definitions your team already works with.

Resolution rate measures how many conversations your AI agent closes on its own. Handoff rate measures how often it passes a conversation to a person instead. A rising handoff rate in one category, such as pricing questions, is usually the fastest sign that something needs attention. For more on these metrics, read the complete guide to customer experience analytics.

Flagged messages are customer messages caught for a topic you monitor, such as a price-match request, a request to delete an account, or an attempt to override the AI agent's instructions, so a person can review them. You can define your own safeguard categories beyond the four built in by default.

Delight.ai flags low-confidence AI answers, the ones most likely to be hallucinated, for review alongside flagged messages. Your team can check each answer, correct it, and update the knowledge behind it so the mistake isn't repeated. Learn more about AI hallucination detection and alerts and Dynamic Safeguards.

Use Trust OS to update the knowledge and instructions behind an incomplete answer or missed recommendation, test the change before rollout, and review how your AI agent handles those conversations afterward.