Lexey automatically assesses the quality of every assistant response in customer conversations.

How it works

Each assistant message is evaluated in real time by an AI assessor that scores it 1–5 and flags issues:

• Poor response — The response was unhelpful or incorrect.
• Missing knowledge — The agent lacked information to answer properly.
• Tone mismatch — The response didn't match the configured tone.
• Missed escalation — The agent should have escalated but didn't.
• Hallucination — The agent stated something not supported by its knowledge base.

Conversation-level QA scores are derived as an average of per-message scores.

Using QA data

• The Conversations tab shows QA score dots on each conversation and per-message scores in the detail view.
• The Manage tab shows an "Improve response quality" prompt when there are actionable quality issues.
• The config agent's knowledge gap detection cross-references QA flags with your existing knowledge articles and suggests specific improvements.

You don't need to interpret QA data yourself. Just tell the management chat what's going wrong — for example, "customers are getting wrong answers about refunds" — and the agent will review the relevant conversations, identify the issue, and suggest improvements to your knowledge base or configuration.