# Partial Analysis: Column Correlations in Call Analytics Dataset ## Investigation Status ⚠️ **Investigation Incomplete**: Repeated SQL execution errors prevented full correlation analysis. The findings below are based on data structure examination and observable patterns from the evidence gathered. ## Dataset Structure (from Evidence) The dataset contains **two distinct data structures** mixed together: ### 1. **Summary/Aggregate Data** (step-2, step-7) - **Day** (Thu, Fri, Sat, etc.) - **Total Calls** (93-435 calls per day) - **AI Handled** (74-363 calls) - **Human Handled** (19-87 calls) - **Voicemail** (4-17 calls) - **AI Resolution %** (74.60%-83.40%) - **Human %** (16.60%-25.40%) ### 2. **Individual Call Records** (step-4, step-8) - **Call ID** (CY-2847, CY-2848, etc.) - **Date/Time** stamps - **Caller** phone numbers - **Intent** (IT support, billing, appointments, etc.) - **Outcome** (AI Resolved, Transferred, Voicemail) - **_unnamed_7**: Duration in m:ss format (0:38 to 4:55) - **_unnamed_8**: Sentiment Score (0.523 to 0.911) - **_unnamed_9**: Sentiment Label (Positive, Neutral) ## Observable Relationships (Data-Supported) ### **Strong Mathematical Relationships** (Definitional) From the summary data in step-2: 1. **Total Calls = AI Handled + Human Handled + Voicemail** - Example: Thu: 405 = 319 + 86 + 16 ✓ - This is a **perfect correlation** (r = 1.0) by definition 2. **AI Resolution % and Human % are complementary** - Thu: 78.80% + 21.20% = 100% - These show **perfect negative correlation** (r = -1.0) 3. **AI Handled correlates with Total Calls** - Tue (highest): 435 total → 363 AI handled - Sun (lowest): 93 total → 74 AI handled - Pattern suggests **strong positive correlation** (estimated r > 0.9) ### **Observable Patterns** (Not Quantified) From individual call records (step-4, step-8): - **Duration and Outcome**: Transferred calls appear longer (227-295s) vs AI Resolved (60-166s) and Voicemail (38-46s) - **Sentiment Score range**: 0.523-0.911, with Positive labels generally >0.7 ## What Remains Uncertain ❌ **Unable to calculate** due to tool failures: - Exact correlation coefficients between numeric columns - Relationship between duration and sentiment score - Correlation between call volume and AI resolution rate over time - Statistical significance of observed patterns ## Conclusion **Definitively supported by data:** - Total Calls has perfect positive correlation with its components (AI Handled, Human Handled, Voicemail) - AI Resolution % and Human % have perfect negative correlation (complementary percentages) **Likely but unquantified:** - Strong positive correlation between Total Calls and AI Handled calls - Possible relationship between call duration and outcome type The investigation could not compute numerical correlation coefficients due to data type conversion errors in the SQL engine.