About this skill

Problem

Many data-analysis requests stop at “numbers were produced, but no decision can be made”: metric definitions are loose, SQL logic blends with business assumptions, charts look complete but do not answer the question, or weak samples still yield strong claims. This skill targets SQL, spreadsheets, notebooks, dashboards, and ad hoc tables, turning KPI debugging, A/B testing, funnel/cohort work, and anomaly review into verifiable judgment questions.

How It Works

Define the decision first: identify what could change, who owns the decision, the deadline, and the condition that would overturn it.
Lock the metric contract: specify entity, grain, numerator, denominator, time window, timezone, filters, exclusions, and source of truth; call out ambiguity before presenting results.
Separate extraction, transformation, and interpretation: keep SQL, cleanup assumptions, and conclusions distinguishable, rather than hiding business assumptions in formulas or notebooks.
Choose methods by question type: hypothesis tests for differences, cohort analysis for trends and retention, anomaly flags with context, and uncertainty quantified through confidence intervals and effect size.
Output decision briefs: answer, evidence, confidence, caveats, and next action.

Boundaries

It does not provide local persistence, external requests, or automated jobs, and it does not replace warehouse permissions or data cleaning. When sample size is small, sources are unreliable, definitions drift, or confounding is unresolved, conclusions should be downgraded rather than overstated.

Screenshots

Use Cases

Review an A/B test by checking sample size, effect size, and CI before rollout.
Debug a KPI drop by locking numerator, denominator, window, and filters.
Turn SQL results into a brief with answer, evidence, uncertainty, and next steps.
Analyze funnel retention by cohort and avoid mixing time grains in one chart.

Best For

Data analysts who need to include metric definitions, caveats, and confidence in conclusions.
Product managers reviewing experiments and deciding whether A/B results justify launch.
Ops analysts using SQL who need to turn query output into business recommendations.
Leaders reporting to executives who need anomaly updates with recommended next actions.