Analytics Reporter

Transforms raw data (CSV, Excel, SQL results, JSON) into structured business intelligence: summary statistics, trend analysis, KPI dashboards, customer segmentation, and actionable recommendations with statistical confidence levels.

Quick Start

1. Validate the dataset for nulls, missing columns, duplicates, and sample size.
2. Select the analysis type from the decision table below.
3. Execute the corresponding implementation (see reference files).
4. Generate the structured report using the output template.

Workflow

Step 1: Data Validation

Before any analysis, validate the input dataset:

import pandas as pd

def validate_dataset(df):
    report = {
        'rows': len(df),
        'columns': list(df.columns),
        'null_counts': df.isnull().sum().to_dict(),
        'null_pct': (df.isnull().mean() * 100).round(2).to_dict(),
        'dtypes': df.dtypes.astype(str).to_dict(),
        'duplicates': df.duplicated().sum()
    }
    # Flag columns with >20% nulls as quality risks
    report['quality_warnings'] = [
        col for col, pct in report['null_pct'].items() if pct > 20
    ]
    return report

Validation gates — stop and notify the user if:

• Required columns are missing

• 20% nulls in key metric columns

• Sample size too small for statistical significance (n < 30 for hypothesis tests)
• Date ranges are inconsistent or contain future dates

Step 2: Select Analysis Type

Step 3: Execute Analysis

Load the relevant reference file for the selected analysis type. Each file contains fully executable, copy-paste-ready code with clear input/output contracts.

• analytics/kpi-dashboard.sql — Monthly revenue KPI metrics
• analytics/rfm-segmentation.py — RFM scoring and segment definitions
• analytics/marketing-attribution.sql — Multi-touch attribution by channel and campaign
• analytics/ab-test.py — Two-proportion z-test with ship/no-ship recommendation

Step 4: Generate Report

Structure every output using this template:

## [Analysis Title] — [Date]

### Executive Summary
- **Primary Finding**: [Insight + quantified impact, e.g. "Revenue grew 14% MoM (±2%, 95% CI)"]
- **Statistical Confidence**: [p-value / CI / sample size]
- **Top Action**: [Single most important next step]

### Key Metrics
| Metric | Value | vs Prior Period | Status |
|--------|-------|-----------------|--------|
| ...    | ...   | ...             | ...    |

### Detailed Findings
[Supporting analysis, segment breakdowns, trend charts]

### Recommendations
1. **Immediate (0–30 days)**: [Action + expected impact]
2. **Medium-term (30–90 days)**: [Initiative + measurement plan]
3. **Strategic (90+ days)**: [Change + evaluation criteria]

### Data Quality Notes
- Sources: [list]
- Rows analysed: [n]  |  Nulls excluded: [n]  |  Date range: [start – end]
- Warnings: [any quality flags from Step 1]

Critical Rules

1. Always run data validation (Step 1) before analysis. Surface quality warnings explicitly.
2. Report statistical confidence (p-value, CI, or sample size) for every quantitative claim.
3. Connect every finding to a business action — no insight without a recommendation.
4. Use NULLIF / null guards in SQL to avoid divide-by-zero silent errors.
5. State assumptions clearly — document any filters, exclusions, or transformations applied.