Skip to main content
Build a high-quality benchmark dataset by reviewing AI-scored calls and flagging discrepancies. This feedback refines Reddy’s QA logic and improves scoring accuracy over time.
Review and mark complete 20 calls to establish a solid benchmark dataset. Focus on calls representing typical scenarios across different agents and call types.

​
Understanding QA Score Icons

Before evaluating calls, understand the scoring status indicators:
  • R icon: Automatically scored by Reddy AI
  • No icon: Not yet scored, configured for manual scoring
  • Verification icon (checkmark): Score verified by someone at your organization

​
Review Process

1

Find the call

From the Dashboard, find the Training Session or Live Call you want to review and click the timestamp.
Dashboard showing training sessions and live calls list
2

Review QA items

On the QA Scorecard tab, review each QA item. When you find a score that should be different, click it.QA scorecard showing items to review
Only QA items with the R icon (AI-scored) can be corrected. Items with no icon require manual scoring.
3

Provide feedback

Select the correct score and explain why this grade is more accurate.Modal for providing QA correction feedback
The more detail you provide, the better Reddy can learn your organization’s scoring standards. Include specific transcript excerpts or timestamps when possible.
4

Mark Complete (Optional)

After reviewing all QA items, click Mark Complete to confirm the scorecard review.
Mark Complete button on QA scorecard to confirm review is finished
Marking complete tells Reddy that unchanged scores were accurate, which trains the AI to improve scoring precision for your organization over time.
Correcting individual QA items (Steps 2-3) is valuable feedback even if you can’t review the entire scorecard. Marking complete is a bonus step that provides the strongest training data when you have time to verify all items.

​
Tips for Effective QA Reviews

Mark complete full scorecards when possible for the strongest training data. The combination of corrected items plus confirmation that other items were accurate provides the most valuable feedback for AI improvement.
Don’t hesitate to correct individual items even if you can’t review the entire scorecard. Every correction helps improve scoring accuracy.
Provide clear reasons for corrections. Include transcript excerpts, timestamps, or note edge cases to help Reddy understand your organization’s specific quality standards.
QA items configured for manual scoring (no icon) won’t have AI grades. Score these as you review to complete the full scorecard picture.
Apply the same scoring criteria across all reviews to ensure Reddy learns a consistent quality benchmark.

​
Prerequisites

Before evaluating QA scorecard accuracy, upload real customer calls to test your scorecard against actual interactions. Use these calls to verify that Reddy’s AI grades match your quality standards and to identify any scoring logic that needs refinement. Upload Call Recordings