Data Analysis Workflow
Analysis usually fails because the question is unclear, definitions differ or correlation is mistaken for causation. This skill locks the decision question and verifies step by step.
Workflow
- 2
Align definitions before querying
Define numerator, denominator, deduplication, time boundaries and exclusions. After AI writes SQL, validate against a known total before segmenting.
- 3
Analyze and visualize
Check distributions and outliers before averages, and compare against at least two baselines. Keep only dimensions that support the conclusion instead of stacking 20 cuts on one page.
- 4
Write conclusion and next experiment
Write three layers: what was observed, possible causes and recommended action. Separate correlation from causation and include a validation plan and success criteria.
Recommended prompts
SQL Query Prompt
Turn a schema and a business question into readable, verifiable SQL.
Data Analysis Prompt
Start from a business question and get a framework, metrics and validation steps.
AI Evaluation Rubric Prompt
Design scorable evaluation dimensions, weights and decision rules for model output.
FAQ
Can I analyze data without SQL?
Natural-language query tools help, but you still need to understand definitions and deduplication. Otherwise the SQL looks right but the result is unusable.
How small is too small a sample?
It depends on effect size and variance. Look at confidence intervals rather than point estimates; flag insufficient samples instead of making strong claims.
Related skills
Dashboard Design
Design a dashboard that supports one decision and can be read at a glance without misleading.
Survey Research
Go from research question to survey design, sampling and insight without leading questions or invalid responses.
Long-Form Blog Production
A repeatable pipeline from topic selection and research to a 3,000-word draft and fact-check.