Sales Transaction Data Analysis

Your EDA report is ready: /workspace/webserver/lib/smartvox/skill_hub/eda_report_sales_data_20251203_175852.html

Executive Summary

Dataset: 1,000 sales transactions × 4 features
Data Quality: Excellent - no missing values, no duplicates
Total Revenue: $6,473,793.27

Key Findings

1. Revenue Patterns

• Distribution: Heavily right-skewed (skewness: 3.63) with extreme outliers
• Average Transaction: $6,474 | Median: $4,512 (significant gap indicates outliers)
• Range: $546 to $82,722 (one extreme outlier at $82k - 12× the average)
• Top 10% Transactions: ≥$13,602 (driving significant revenue)
• Outliers: 72 high-value transactions (7.2%) - likely B2B or bulk orders

2. Units Sold

• Total Units: 28,592 across 1,000 transactions
• Average: 28.6 units | Median: 24 units
• Distribution: Moderately right-skewed (skewness: 1.38)
• Range: 0 to 129 units (2 transactions with 0 units - returns or errors?)
• Outliers: 46 high-volume transactions (4.6%)

3. Profit Margins

• Average Margin: 27.9% | Median: 28.6% (healthy, consistent)
• Range: -2.2% to 48.6%
• Distribution: Slightly left-skewed (skewness: -0.37) - few low-margin outliers
• Low Margin (<15%): 98 transactions (9.8%) - review for profitability
• Negative Margin: 1 transaction (-2.2%) - potential error or promotional loss

4. Discount Strategy

• Average Discount: 15.4% | Median: 3.3% (median much lower - most get small discounts)
• Distribution: Right-skewed (skewness: 1.12) - few get large discounts
• No Discount: Only 2 transactions (0.2%)
• High Discount (>50%): 108 transactions (10.8%) - aggressive promotions?

5. Correlation Analysis

CRITICAL FINDING: No significant correlations detected between any variables (all |r| < 0.3)

• Revenue ↔ Units Sold: -0.009 (essentially zero - counter-intuitive!)
• Revenue ↔ Discount Rate: -0.040 (weak negative - expected but minimal)
• Profit Margin ↔ Discount Rate: -0.013 (no relationship - surprising)

Interpretation: Revenue is driven by price per unit, not volume or discounts. Higher discounts don’t meaningfully increase units sold or reduce margins.

Data Quality Issues

1. Two Zero-Unit Transactions: Rows with units_sold = 0 but positive revenue (rows 658, 715) - likely returns, cancellations, or data errors
2. One Negative Margin Transaction: Row 606 with -2.2% margin - investigate if intentional (loss leader) or error
3. Extreme Revenue Outlier: $82,722 (row 739) - validate this transaction
4. High Kurtosis in Revenue: 26.93 (heavy tails) - extreme outliers dominate distribution

Recommendations

1. Data Cleaning

• Investigate and potentially remove/correct the 2 zero-unit transactions
• Validate the $82k revenue outlier (row 739)
• Review the negative margin transaction (row 606)

2. Discount Strategy Reassessment

• Finding: Discounts show NO correlation with units sold (-0.016) or revenue (-0.040)
• Action: Evaluate if 10.8% of transactions receiving >50% discounts are cost-effective
• Recommendation: A/B test reducing discount frequency - may maintain revenue with better margins

3. Focus on High-Value Transactions

• Top 10% of transactions (≥$13,602) drive significant revenue
• Identify characteristics of these 100 transactions to replicate success
• Separate B2B/bulk orders from retail for targeted analysis

4. Revenue Driver Analysis

• Critical Gap: Revenue doesn’t correlate with units sold
• Implication: Price per unit varies dramatically across transactions
• Next Step: Calculate revenue_per_unit = revenue / units_sold to identify premium vs. discount products

5. Modeling Considerations

• Log-transform revenue before regression modeling (extreme skewness: 3.63)
• Segment analysis by transaction size (retail vs. bulk) will be more meaningful than aggregate models
• Outlier treatment: Consider capping revenue at 99th percentile or using robust regression methods

6. Profitability Investigation

• 98 transactions (9.8%) have margins <15% - calculate total profit contribution
• If these low-margin transactions are high-volume, they may still be valuable
• If low-margin AND low-volume, consider raising prices or eliminating these products

Technical Notes

• File Location: /workspace/webserver/lib/smartvox/skill_hub/eda_report_sales_data_20251203_175852.html
• Analysis Method: Full statistical EDA with visualizations embedded
• Outlier Detection: IQR method (1.5× IQR from Q1/Q3)
• Report Format: Self-contained HTML with base64-encoded images (works offline)