Data Analysis Specialist

Purpose

This skill equips the AI system with a structured, rigorous methodology for performing end-to-end data analysis. It enables the agent to ingest raw datasets, clean and preprocess data, execute exploratory data analysis (EDA), perform statistical evaluation, and generate clear, actionable insights with accompanying data visualizations.

When to use

Use this skill when:

• A user provides a raw dataset (CSV, JSON, Excel, etc.) and requests insights, patterns, or trends.
• There is a need to validate hypotheses, identify anomalies, or evaluate metrics over time.
• The task requires translating complex statistical findings into plain-English business recommendations and high-quality charts.

Instructions

1. Workflow

Phase 1: Data Assessment & Integrity Check

• Load and Inspect: Read the dataset and check its shape (rows, columns), data types, and basic descriptive statistics.
• Identify Anomalies: Explicitly check for missing values, duplicate rows, and extreme outliers.
• Data Rectification: Handle missing values logically (e.g., imputation via median/mean for continuous variables vs. mode for categorical variables, or dropping rows if appropriate). Document these cleaning choices.

Phase 2: Exploratory Data Analysis (EDA)

• Univariate Analysis: Examine the distribution of key variables using histograms or box plots.
• Multivariate Analysis: Evaluate correlations between key metrics using correlation matrices or scatter plots to find hidden dependencies.
• Feature Aggregation: Group data by relevant categorical segments (e.g., time periods, regions, user types) to discover localized trends.

Phase 3: Statistical Formulation & Insights

• Hypothesis Testing (If Applicable): Use appropriate statistical models (e.g., A/B testing evaluation, t-tests, or linear regression) to confirm if trends are statistically significant.
• Insight Synthesis: Move from numbers to narrative. Highlight the "What" (the trend) and the "Why" (the driving force).

Phase 4: Reporting & Visualization

• Chart Selection: Choose the correct chart for the job (e.g., Line charts for temporal trends, Bar charts for comparisons, Scatter plots for relationships).
• Final Delivery: Structure the output with a Key Takeaways Executive Summary, followed by Detailed Breakdown, Visualizations, and Next Steps/Recommendations.

2. Constraints & Guardrails

• No Data Hallucination: Rely only on the provided dataset. Never invent data points, metrics, or trends that cannot be statistically proven from the source file.
• Acknowledge Bias/Limitations: If the sample size is too small or missing data is too pervasive to draw a reliable conclusion, explicitly state this limitation.
• Chart Best Practices: Always include clear labels for axes, titles, and legends. Avoid overly cluttered charts.

3. Concrete Example

User Prompt:

"Analyze this dataset of customer churn (churn.csv) and tell me why we are losing users."

Agent Execution Flow:

1. Assessment: "Dataset loaded with 10,000 rows. Identified 2% missing values in TotalCharges; imputed using the column median. 15 duplicate rows removed."
2. EDA: Aggregated churn rate by ContractType. Noticed Month-to-Month contracts have a 42% churn rate compared to Two-Year contracts at 5%.
3. Statistical Insight: "Ran a chi-square test confirming that contract type is a statistically significant predictor of churn ($p < 0.05$)."
4. Final Output Format:
   • Executive Summary: Month-to-month subscribers are the primary driver of customer attrition.
   • Supporting Visual: [Bar chart showing Churn Rate % by Contract Type]
   • Recommendation: Introduce a promotional discount incentive to transition Month-to-Month users into 6-month or 1-year commitments.