How Call Scoring Works in Empower

Empower evaluates customer calls using a combination of AI-generated insights and statistical analysis. The goal is to provide an accurate, actionable score for every call to support coaching, quality assurance, and performance tracking.

 

 

🧩 Core Components

 

Each call is evaluated across six components, each scored on a 0 to 5 scale and assigned a weight (1 to 5):

 

 

 

1. Objection Handling & Product Knowledge (AI-based)

 

This component is derived by averaging two independent AI-generated scores:

• Objection Handling:
  
  • The LLM identifies client objections and the agent’s response/resolution steps.
  • Based on how well objections are handled, a score from 0 to 5 is generated.

 

• Product Knowledge:
  
  • The LLM checks if the agent’s statements and answers align with a predefined product knowledge base.
  • It scores accuracy, completeness, and clarity.

 

🧠 While users only see the final score, a detailed internal analysis is performed to increase reliability using a "reasoning before answering" methodology.

 

 

2. Call Mood (Statistical)

 

• Raw mood is a number between -1 (negative) and +1 (positive).
  
  

It is normalized to a 0 to 5 scale using:

Mood Score = ((Mood + 1) / 2) * 5

 

 

3. Topics / "Moments" (Statistical)

 

• Based on a list of "must-have" moments provided in the scorecard.

Score is calculated as:

• Moments Score = (Number of Detected Moments / Total Expected Moments) * 5

 

 

4. Script Adherence (AI-based)

 

The LLM:

1. Groups semantically similar speech turns.
2. Matches them to the steps in the predefined call script.
3. Scores based on how many steps were covered and how well.

 

✅ This AI-driven semantic mapping ensures flexibility beyond keyword matching.

 

 

5 & 6. Call Duration & Speech Rate (Statistical)

 

Both are scored based on whether they fall within admin-defined acceptable ranges.

• If in range → Score = 5

 

If outside → Penalized proportionally:

• Score = 5 - (deviation from range midpoint / total range)

 

Example:

• Expected speech rate: 100–200 WPM
• Actual speech rate: 50 → Score = 2.5 (halfway below)
• Actual speech rate: 250 → Score = 2.5 (halfway above)

 

 

🧮 Final Score Computation

 

Each component contributes to the final call score based on its assigned weight:

Final Score = 

• (Score₁ × Weight₁ + Score₂ × Weight₂ + ... + Score₆ × Weight₆) / 
• (Weight₁ + Weight₂ + ... + Weight₆)

 

📌 Each component has a max score of 5, but the influence on the final score depends on its weight (1–5).

 

 

✅ Summary

 

 

This scoring system ensures a fair and balanced assessment combining explainable data-driven insights with deeper AI analysis.