Responses & Insights

Once your deployment is active and participants complete interviews, Third Ear automatically collects, structures, and processes response data.

 

1. What Is a Response?

A Response is a single completed participation event within a deployment.

Each response contains:

• Structured question–answer pairs
• Timestamps
• Transcript (voice mode)
• Optional audio recording
• Processing status

Responses are tied to a specific deployment but based on the interview structure.

 

 

2. Response Lifecycle

Every response progresses through system states:

• IN_PROGRESS – Participant has started but not finished
• COMPLETED – Interview finished successfully
• PROCESSING – AI pipeline is analyzing transcripts
• FAILED – Processing error (rare, retry supported)

Processing begins automatically after completion. You can monitor status in real time.

 

 

3. Reviewing Individual Responses

To inspect a response:

1. Open the deployment.
2. Click on a response.
3. Review transcript, answers, and metadata.

For voice interviews, you can:

• View the full diarized transcript
• Play the audio recording (if enabled)
• Search within transcript
• Copy text

For text interviews, responses appear in structured Q&A format.

 

 

4. What Is Raw Data?

Raw data includes:

• Verbatim transcripts
• Structured question–answer JSON
• Participant metadata
• Audio file reference (if applicable)

Raw data represents exactly what participants said — without interpretation. Third Ear ensures traceability from insights back to verbatim quotes.

 

 

5. What Is an Insight?

An Insight is not a single response.

An insight is an AI-generated, cross-response finding derived from pattern detection across multiple transcripts. Insights include:

• Aggregated summary
• Supporting quotes
• Participant references
• Confidence score

Insights are always grounded in underlying transcripts.

 

 

6. Insight Generation Pipeline

After responses are completed:

1. Transcripts are normalized.
2. Speaker turns are identified (diarization).
3. Responses are grouped by question.
4. AI extracts patterns and themes.
5. Confidence scores are calculated.

This processing happens automatically in the background. Large datasets may take several minutes depending on volume.

 

 

7. Understanding Confidence Scores

Each insight includes a confidence score between 0.00 and 1.00.

Confidence reflects:

• Number of confirming responses
• Consistency across participants
• Quality of supporting quotes
• Model certainty

Confidence ranges:

• 0.80–1.00 → High confidence
• 0.50–0.79 → Medium confidence
• 0.25–0.49 → Low confidence
• 0.00–0.24 → Very low confidence

Low confidence does not mean incorrect — it may indicate insufficient data or divergent opinions.

 

 

8. Categorical Analysis & Distributions

For structured questions (e.g., multiple choice, rating scales), Third Ear automatically provides:

• Frequency distributions
• Percentages
• Participant breakdowns
• Sorted answer ranking

This allows quantitative patterns to appear alongside qualitative insights.

 

 

9. Exploring Transcripts

You can browse transcripts individually or across the dataset.

Features include:

• Speaker-labeled turns
• Search functionality
• Copy/export capability
• Audio sync (if available)

Transcripts remain accessible even after insights are generated. This ensures transparency and human oversight.

 

 

10. Asking Additional Questions (Interactive Insights)

You may generate additional insights by asking new analytical questions.

For example:

• “What are the most common frustrations mentioned?”
• “How do experienced users differ from new users?”

The system processes your query across transcripts and produces structured findings.

 

 

11. Exporting Data

You can export:

• Raw responses (JSON / CSV if enabled)
• Transcripts
• Insight summaries
• Categorical breakdowns

Exports allow integration with external tools or reporting workflows.

 

 

Best Practices for Insights

• Wait for multiple completed responses before interpreting patterns.
• Always review supporting quotes.
• Treat low-confidence insights as research prompts, not conclusions.
• Combine categorical trends with qualitative depth.
• Use exports to document findings in stakeholder reports.

 

 

If you get stuck, open Support or reach out to your team workspace admin.