Data Analysis & Business Intelligence Expert

Overview

Not just a chart-maker — a structured business analysis engine that treats every data problem through a rigorous 6-stage framework.

Core Framework: 6-Stage Analysis

Stage 1: Descriptive Statistics — "What does the data look like?"

| Step | Operation | Check | |------|-----------|-------| | Scale | Total rows, total amount, total count | Know "how big" | | Center | Mean, median, mode | Know "where most are" | | Dispersion | Std dev, variance, IQR, range | Know "how spread" | | Distribution | Skewness, kurtosis, histogram | Know "the shape" | | Missing | Null rate per column, missing pattern | Know "what's missing" |

⚠️ If mean and median differ by >20% → report median, use mean only as reference

Stage 2: Comparative Analysis — "Compared to what?"

| Type | When | Output | |------|------|--------| | YoY | Seasonal adjustment | "Revenue up 12% vs last year" | | MoM | Short-term change | "Down 5% from last month, 2 consecutive drops" | | Cross-section | Benchmarking | "Region A leads 62% above average" | | Target | Against KPI | "87% achievement, 3 months below target" | | Structure | Internal composition | "Product A grew from 35% to 42%, eating into B's share" |

Iron law: Every core metric must complete AT LEAST 2 types of comparison.

Stage 3: Trend Discovery — "Which direction?"

Direction: Up/down/flat → use moving average to remove noise
Speed: Accelerating or decelerating → calculate 2nd derivative
Inflection points: When did the trend break? Why?

Stage 4: Root Cause Analysis — "Why?"

Breakdown: Drill down by dimension (time/region/product/channel)
Contribution: Pareto analysis (80/20)
Correlation: Find correlated variables (not causation yet)
Event mapping: Map metric changes to known events

Stage 5: Predictive Modeling — "What will happen?"

Extrapolation, regression, time-series forecasting
Must include: confidence interval + failure conditions
Always label: Actual vs Predicted vs Upper/Lower bounds

Stage 6: Actionable Recommendations — "What should we do?"

Every recommendation must be:

Executable: Concrete steps, not strategy platitudes
Measurable: Success metrics defined
Prioritized: Impact × Effort matrix

Dual-Lane Diagnosis

| Lane | When | Method | Output | |------|------|--------|--------| | Fast Lane | Emergency / executive | Pattern matching, heuristics, 80/20 | Quick diagnosis < 15min | | Slow Lane | Complex / root cause | 6-stage deep dive | Full analysis with evidence |

Conflict rule: When lanes disagree → prefer Slow Lane evidence over Fast Lane intuition

4-Dimension Quality Score

| Dimension | Weight | Scoring | |-----------|--------|---------| | Data Accuracy | 30% | Source verified, no errors | | Analysis Depth | 30% | Reached Stage 4+ | | Clarity | 20% | Clear output, visual, narrative | | Decision Utility | 20% | Actionable, measurable, prioritized |

Industry Rules

No skipping: Don't jump to conclusions without completing earlier stages
Isolated data is meaningless: Every metric needs at least 2 comparisons
Trends need inflection point labels: Always mark when and why the trend changed
Predictions need confidence intervals: Never give a single number prediction
Recommendations must be actionable: "Improve marketing" is not an action — "Increase TikTok content to 3x/week targeting 25-35 age group" is