root-cause-analysis
Use when investigating a KPI movement, spike, drop, or concentration in a dataset and the AI should structure evidence, candidate causes, confidence, and next checks in a disciplined way.
Root Cause Analysis
Purpose
Diagnose a KPI movement in a way that is analytically disciplined and decision-ready.
When to use
Use this skill when:
- a metric moved unexpectedly
- the user asks "why did this change," "what drove this drop," or "what explains these numbers"
- a dataset, export, or KPI table shows a clear spike, drop, or concentration that needs explanation
- someone wants an explanation, not just a trend report
- there are multiple candidate drivers to evaluate
When not to use
Do not use this skill when:
- you only need to define the metric
- there is not enough evidence to go beyond framing and hypothesis generation
Required thinking discipline
- Keep observation, inference, hypothesis, and recommendation separate.
- Include measurement risk as a possible explanation.
- Avoid causal certainty unless evidence supports it.
- Evidence constraint: Every conclusion must cite specific data — a number, a rate, a segment, or a timeframe. Do not speculate without evidential basis. If data is insufficient, state what is missing rather than asserting an unsupported inference.
Workflow
- State the metric, timeframe, and baseline.
- Summarize what changed in direct observational terms.
- Break down the change by key segments or driver components.
- Evaluate candidate causes against the available evidence.
- State confidence level for each likely driver.
- End with recommended next checks or actions.
Output format
- Metric change summary
- Key observations
- Likely drivers
- Alternative explanations
- Confidence and uncertainty
- Recommended next steps
Good example
Observation: refund-adjusted revenue fell 9% week over week, concentrated in two EU markets.
Inference: the decline is unlikely to be broad-based demand weakness.
Hypothesis: a payment routing issue is the leading candidate because authorization rate fell at the same time.
Recommendation: confirm processor logs and compare fallback success by country.
Bad example
Revenue dropped because the pricing update upset customers.
Why this is bad:
- it jumps from change to cause
- it does not show evidence
- it ignores competing explanations
Practical notes
- If evidence is weak, say "leading hypothesis" instead of "root cause."
- If multiple mechanisms exist, rank them instead of forcing one explanation.
- For generic asks, this is the default diagnostic mode once the metric, timeframe, and baseline are clear.
Optional variants
- Product incidents: emphasize instrumentation, rollout timing, and affected cohorts.
- Operations issues: emphasize bottleneck stage and exception categories.