Data Analysis

• You need to debug a KPI, investigate a metric spike or drop, or explain why a number changed.
• You're running an A/B test, cohort analysis, or funnel review and need statistically sound interpretation — not just a raw number.
• You're preparing an executive report or stakeholder brief and need findings framed as a decision, not a data dump.
• You're working with SQL, Python notebooks, spreadsheets, or BI tool exports and the hard part is analytical judgment: what to measure, how to compare it, and what it actually means.

• You only need mechanical spreadsheet formatting or raw SQL syntax help with no analytical layer — the companion sql or csv skills may be sufficient.
• You need to build or configure a live dashboard or reporting pipeline; consider the dashboard or business-intelligence skills instead.
• The data quality is too poor or the sample size too small to support any reliable inference — this skill will flag that and block unsupported conclusions rather than produce false confidence.

Before touching any data, the skill anchors every analysis to a concrete decision: who owns it, what would change if the result is X vs Y, and what the relevant timeframe is. Analysis without a clear decision is treated as incomplete.

Every calculation is locked to an explicit metric contract: entity, grain, numerator, denominator, time window, timezone, filters, exclusions, and source of truth. Ambiguities are surfaced before results are presented, preventing silent definition drift.

The skill validates sample size sufficiency, fair comparison groups, multiple-comparison risk, practical vs. statistical significance, and uncertainty quantification. Results are presented as ranges (e.g. "12–18% lift") rather than false point estimates.

The skill maps analytical questions to the right method: hypothesis tests for comparisons, regression for prediction, cohort analysis for retention, segmentation for group differences, and anomaly detection for unusual patterns — each with the correct key outputs.

Every result is structured as: answer, evidence, confidence level, caveats, and recommended next action. Stakeholder-facing outputs translate technical findings into business implications rather than leading with methodology.

The skill actively flags analysis pitfalls such as reused KPI names after definition changes, mixing aggregation grains in one chart, showing percentages without underlying counts, and post-hoc narrative hunting — before they corrupt a decision.

After running a product experiment, use this skill to validate sample size, check for novelty effects, compute effect size with confidence intervals, and produce a stakeholder brief that states whether the result is decision-ready or requires further testing.

When a key metric moves unexpectedly, the skill walks through metric contract verification, segment decomposition, time-grain consistency, and confounder checks to identify whether the movement is real signal or an artifact of definition or data quality issues.

Convert raw query results or notebook outputs into structured decision briefs — leading with the insight, quantifying uncertainty, stating what the data cannot tell you, and recommending the next action — suitable for leadership or cross-functional stakeholders.

Analyze how different user cohorts behave over time, produce retention curves broken down by acquisition period or segment, and interpret whether observed differences are statistically meaningful or within noise.