About this skill

Decision-Oriented Data Analysis

When SQL, spreadsheets, notebooks, dashboards, or exported tables contain many numbers, but it is hard to judge which conclusions are trustworthy and what action should change, this skill provides an analysis framework. It shifts focus from “producing results” to “supporting decisions,” and is useful for KPI debugging, A/B experiment interpretation, funnel and cohort analysis, anomaly reviews, metric quality checks, and executive reporting.

How It Works

It starts by clarifying the decision question and metric contract: who owns the decision, what result would change the decision, entity grain, numerator and denominator, time window, timezone, filters, exclusions, and source of truth must be explicit. It then separates extraction, transformation, and interpretation, so business assumptions are not hidden inside SQL, formulas, or notebook code. The analysis emphasizes statistical rigor: sample size, fair comparison groups, multiple comparisons, effect size, confidence intervals, and sensitivity checks all matter. Outputs should lead with the conclusion, evidence, confidence, limitations, and next steps; when samples are weak, definitions drift, or confounders are unresolved, conclusions should be downgraded rather than overstated.

Boundaries

It does not replace data cleansing, permission governance, BI engineering, or dashboard implementation. It helps form auditable analytical conclusions from existing data. If the source is unreliable, the sample is too small, or the goal is to prove a predetermined conclusion, the skill should surface risk and preserve uncertainty.

Screenshots

Use Cases

Debug KPI anomalies by checking numerator, denominator, time window, and filters before judging business changes.
Interpret A/B tests by checking sample size, multiple comparisons, effect size, and confidence intervals.
Review cohort retention by fixing cohort definitions and time grain before drawing conclusions.
Prepare executive briefs with metric definitions, evidence, confidence, and limitations.

Best For

Data analysts owning KPI dashboards who need to explain metric definitions and anomaly drivers.
Product managers running A/B tests who need to judge whether results are strong enough to change launch decisions.
Operations leads writing monthly reviews who need cohort, funnel, and anomaly findings turned into recommendations.
BI engineers reviewing query logic who need extraction, cleanup assumptions, and conclusions separated.