Skip to main content

AI Feedback Loop

Qarion collects structured feedback on every AI-generated suggestion — descriptions, field classifications, quality check recommendations, and anomaly explanations. This feedback loop helps administrators measure AI accuracy and identify areas for improvement.

Providing Feedback

Inline Actions

When the AI generates a suggestion (e.g., a product description), you see inline feedback controls:

  • Thumbs Up — The suggestion is accurate as-is
  • Thumbs Down — The suggestion is not useful
  • Accept — Apply the suggestion directly
  • Edit — Modify the suggestion before applying (the original and edited values are both recorded)
  • Reject — Dismiss the suggestion entirely

Every action is logged along with the AI feature that produced it, the associated AI log entry, and an optional comment explaining why the suggestion was accepted or rejected.

Adding Comments

When giving a thumbs down, editing, or rejecting a suggestion, you can add a free-text comment. These comments are especially valuable for understanding patterns — they help identify whether suggestions are failing due to missing context, incorrect assumptions, or formatting issues.

Administrator Dashboard

Feedback Stats

Navigate to Administration → AI Feedback to view aggregated statistics across all AI features:

MetricDescription
Total CountTotal number of feedback entries for the period
Acceptance RatePercentage of positive outcomes (thumbs up + accepted + edited)
Thumbs Up / DownCount of each rating action
Accepted / Edited / RejectedCount of each outcome action

Stats can be filtered by feature and by time period (last 7, 14, or 30 days).

Exporting Data

Click Export to download all feedback entries as CSV or JSON. Exports include every field: feedback ID, feature, action, comment, AI log reference, user ID, original value, edited value, and timestamp. This data is useful for fine-tuning prompts, identifying systematic inaccuracies, and reporting on AI adoption metrics.

Supported Features

Feedback collection is integrated into the following AI features:

FeatureFeedback Trigger
Description GenerationAfter generating a product description
Field DescriptionAfter generating a column-level description
Tag SuggestionsAfter AI suggests classification tags
Quality Check SuggestionsAfter AI recommends quality rules
Anomaly ExplanationAfter generating a root-cause hypothesis
AI ClassificationAfter bulk field classification suggestions