EcomBot Docs

Quality Audit & Self-Improvement

Audit your chatbot's performance: automatically detect sales frictions, hallucinations, and continuously improve its responses.

EcomBot features an automated background analysis system that evaluates the quality of the conversations handled by your chatbot on a daily basis. This intelligent audit detects sales obstacles, missed opportunities, visitor sentiment, and potential inaccuracies to propose concrete optimizations.


Why Audit Conversation Quality?

To maximize your conversions, your virtual assistant must be flawless. Instead of manually reviewing hundreds of transcripts, EcomBot performs a daily semantic analysis to isolate priority improvement points based on several key criteria:

  1. Sales Obstacles & CRO: Identifies moments where the AI could have steered the customer toward a purchase or suggested a cross-sell more effectively.
  2. Sentiment & Satisfaction: Analyzes visitor frustration or enthusiasm to highlight conversations that require special attention.
  3. Accuracy (Hallucinations): Ensures that the chatbot remained strictly faithful to your catalog data and knowledge base rules.
  4. Relevance (Knowledge Gaps): Detects unanswered questions or cases where the customer did not get the information they were looking for.
  5. Proactivity: Evaluates whether the AI took the initiative to assist the customer or rescue a failed search query.

How the Automatic Analysis Works

Every night, our secure cloud infrastructure processes the previous day's chat sessions:

  • Global Scoring: A performance score is assigned to each project.
  • Extraction of Fixes: The AI identifies missing information and formulates improvement suggestions.
  • Self-Correction: EcomBot generates corrective guidelines to be integrated into your chatbot's global instructions to refine its future behavior.

How to Apply Suggested Improvements

To review audit results and optimize your agent:

  1. Go to your EcomBot Dashboard, under Insights (Améliorations).
  2. Browse the list of self-improvement suggestions.
  3. For each proposed correction, review the detailed analysis (the detected weakness and the suggested prompt adjustment).
  4. Click Apply Recommendation to instantly inject the fix into your agent's instructions.

The corrections are applied starting from the very next conversation, in a completely transparent manner.