This guide explains every metric and score in simple terms. Use this to get the most value from your call analytics.
CallTranscribe is an AI-powered tool that listens to phone calls between your team (agents) and customers/patients, then provides detailed analysis to help you:
Converts speech to text so you can read what was said without listening to the entire call
Measures how well the conversation went using objective metrics
Provides specific feedback on what went well and what can be improved
Estimates likely outcomes like customer satisfaction and conversion probability
The AI uses highly deterministic settings. This means if you analyze the same call twice, you will get very similar (nearly identical) results. This ensures your evaluations are fair and consistent across all calls and agents.
Our AI uses strict evaluation criteria. We believe in high standards that drive real improvement. A score of 90+ is rare and exceptional - most good calls score in the 70-80 range. Don't expect inflated scores; expect honest, actionable feedback.
Throughout the analysis, you will see scores from 0 to 100. Here is what they mean:
Truly outstanding. Flawless execution, exceeded expectations, built excellent rapport, no missed opportunities. This is rare and worth celebrating. Use as a training example.
Strong performance with only minor issues. The agent did most things right. Reinforce this behavior while noting small areas for polish.
Solid performance with some clear improvement areas. This is a competent agent who can grow with targeted coaching.
Did the job but nothing special. Several areas need improvement. This agent needs coaching to move to the next level.
Significant issues that need training. Multiple areas require attention. Schedule coaching session soon.
Serious concerns. This call had major issues that need immediate review and correction. Manager escalation recommended.
Inflated scores feel good but don't drive improvement. Our strict scoring ensures:
If an agent scores 72 in Empathy, it means they showed good understanding of the customer's feelings, but there were specific moments where they could have been more supportive. This is a solid score - the agent is competent but has room to become excellent. Check the coaching feedback for specific suggestions on how to move from “good” to “very good.”
These are objective measurements about how the conversation flowed. They help identify patterns and issues.
What it shows: How much time each person spent talking during the call.
Why it matters: If the agent talks too much (60%+), they may not be listening to the customer. If they talk too little (30% or less), they may not be providing enough value.
Ideal range: Agent should speak 40-50% of the time.
Total Questions: How many questions the agent asked during the call.
Open Questions: Questions that require detailed answers. Example: “How have you been feeling lately?” or “What brings you in today?”
Closed Questions: Questions with yes/no answers. Example: “Is the pain constant?” or “Did you take the medication?”
Why it matters: Good agents ask more open questions to understand the customer deeply, then use closed questions to confirm specific details.
Ideal: More open questions than closed questions, especially early in the call.
What it shows: How many times someone started speaking while the other person was still talking.
Why it matters: Frequent agent interruptions suggest poor listening skills and can frustrate customers. Customer interruptions may indicate confusion or strong emotions.
Ideal: Less than 3 interruptions by the agent per call.
Average Response Time: How quickly the agent responds after the customer finishes speaking.
Why it matters: Too fast (under 1 second) suggests the agent isn't fully listening. Too slow (over 4 seconds) creates awkward silences.
Ideal: 1-3 seconds response time.
What it shows: Speaking speed of both the agent and customer.
Why it matters: Speaking too fast (180+ WPM) makes it hard for customers to follow. Speaking too slow (under 100 WPM) can seem disengaged.
Ideal: 120-150 WPM for professional conversations.
Every good call follows a natural structure. The analysis breaks down the call into these phases:
The first few seconds where the agent introduces themselves and sets the tone.
Good example: “Good morning! Thank you for calling ABC Clinic. My name is Priya. How may I help you today?”
What to look for: Warm, professional, includes name, asks how to help.
Understanding what the customer needs through questions and active listening.
Good example: Asking about symptoms, duration, previous treatments, lifestyle factors.
What to look for: Open questions, paraphrasing to confirm understanding, empathy.
Presenting options, explaining treatments, or providing information.
Good example: Clearly explaining what the treatment involves, expected outcomes, and costs.
What to look for: Clear language (no jargon), checking for understanding, addressing concerns.
Addressing concerns, questions, or hesitations the customer raises.
Good example: “I understand the cost is a concern. Let me explain our payment options...”
What to look for: Acknowledging the concern, providing reassurance, offering alternatives.
Wrapping up with clear next steps and a professional goodbye.
Good example: “So you're confirmed for Thursday at 3 PM. We'll send you a reminder. Is there anything else I can help with?”
What to look for: Summarizing agreed actions, confirming details, offering additional help.
Each segment is rated as Excellent, Good, Average, or Poor based on how well it was executed.
The AI evaluates the agent across multiple skills. Here's what each category measures:
How well did the agent start the call? Did they introduce themselves, sound welcoming, and set a positive tone?
How effectively did the agent understand the customer's needs? Did they ask good questions and listen actively?
How clearly did the agent explain options or recommendations? Was the information relevant and easy to understand?
When the customer raised concerns, how well did the agent address them? Did they acknowledge and resolve the issues?
How effectively did the agent wrap up? Were next steps clear? Did they ask if anything else was needed?
Did the agent show understanding of the customer's feelings and situation? Did they acknowledge emotions appropriately?
Was the agent easy to understand? Did they avoid jargon and explain things simply?
Did the agent follow required protocols? Did they make proper disclosures and avoid misinformation?
Things the agent did well. Use these as examples of good behavior to reinforce in training.
Specific things that could be done better. Focus training on these areas.
Chances to help the customer better or close a sale that were not taken. Learn from these for future calls.
Suggested phrases the agent could use in similar situations. These can be copied and practiced.
Serious issues that need immediate attention - like rude behavior, misinformation, or compliance violations.
The AI identifies important moments during the call that deserve attention. Here are the types of moments detected:
Customer expressed frustration, dissatisfaction, or a problem. These need careful handling.
Customer said something positive about the service, agent, or company. Great for morale!
Customer raised a concern or hesitation. How these are handled often determines the outcome.
Customer mentioned another company or service. Useful for competitive intelligence.
Money, costs, or pricing came up. Important for sales and objection handling analysis.
Customer agreed to something - an appointment, treatment, purchase, etc. Positive signal!
Customer had an “aha moment” where they understood something important.
Moment where the situation could escalate into a formal complaint if not handled well.
Each moment is tagged with importance:
Based on the conversation, the AI predicts likely outcomes. These are educated guesses, not guarantees.
What it means: How likely is the customer to take the desired action (book appointment, sign up, purchase)?
High (70-100%): Customer showed strong interest, agreed to next steps, seemed satisfied.
Medium (40-69%): Some interest but also hesitation. May need follow-up.
Low (0-39%): Significant barriers exist. Needs more work to convert.
What it means: How likely is the customer to leave or not return?
High Risk: Customer expressed strong dissatisfaction, unresolved issues, or intent to leave. Act immediately.
Medium Risk: Some concerns but not critical. Follow up to ensure satisfaction.
Low Risk: Customer seems happy and likely to return.
What it means: How likely is this to become a formal complaint?
High Risk: Customer was very upset, threatened to complain, or issue was not resolved. Manager should review.
Medium/Low Risk: Normal interaction with no significant escalation potential.
What it means: How satisfied is the customer likely to be based on this interaction?
This combines sentiment, issue resolution, agent behavior, and conversation flow to estimate overall satisfaction.
Predictions are based on patterns in the conversation. They are not 100% accurate. Use them as guidance for prioritizing follow-ups and identifying at-risk customers, but always apply human judgment.
Yes! The AI uses very consistent settings (low temperature), so analyzing the same call multiple times will produce nearly identical results. This ensures fair and consistent evaluations.
Yes. The AI understands English, Hindi, and mixed Hindi-English (Hinglish) conversations, which is common in Indian healthcare and business contexts.
Predictions are based on patterns in conversation and are generally reliable, but they are estimates, not guarantees. Use them to prioritize actions, not as absolute truths.
AI transcription is highly accurate but not perfect, especially with poor audio quality, heavy accents, or background noise. Always use the audio player to verify critical parts.
Typically 30-60 seconds per call, depending on the length. Longer calls take more time.
Yes! You can export the full analysis as JSON, the transcript as text, or the coaching report separately. Use the export buttons on the analysis page.
Most common audio formats: MP3, WAV, M4A, AAC, OGG, FLAC, MPEG, and WebM. Files should be under 20MB.