Purpose
Convert raw data findings into a decision-ready business insight — moving past "here's what the number is" to "here's what it means, why it's happening, what it's worth, and what to do next." Solves the analyst's recurring problem of producing accurate analysis that stakeholders can't act on.
When to Use
Invoke this skill when the user asks to:
- Turn query results, a metric change, or a stat into a business recommendation
- Write the "insight" or "so what" for a report, dashboard summary, or exec update
- Explain why a metric moved and what should be done about it
- Translate a statistical finding into plain-language impact for non-technical stakeholders
- Prioritize which of several findings is worth acting on
- Prepare an analysis for a decision meeting or review
Inputs Required
Before starting, collect:
- The finding(s): The raw data output — numbers, deltas, percentages, segment breakdowns, correlations, or query results
- Metric definition: What the metric actually measures and how it's calculated
- Time frame & comparison basis: The period covered and what it's compared against (prior period, target, benchmark, control group)
- Business context: What the team/company is trying to achieve and why this metric matters
- Audience: Who reads this — exec, PM, marketing lead, or fellow analysts (drives tone and technical depth)
- Known variables: Any launches, campaigns, seasonality, or events that could explain movement
- Decision at stake: What choice this insight is meant to inform, if known
If audience is unspecified, default to a cross-functional business lead: plain language, one technical footnote allowed. If the decision at stake is unknown, infer the most likely decision the finding points toward and state that assumption.
Instructions
Step 1: Isolate the Signal
Separate the genuine finding from noise before framing anything.
- Materiality check: Is the change large enough to matter? Note absolute AND relative change (a 40% jump on 5 conversions is not a finding).
- Significance check: Is sample size adequate? For rate/proportion changes, flag if the base is under ~100 events or if the change is within likely variance.
- Direction confirmation: Confirm the metric moved the way the raw number implies (watch for inverted metrics like cost, churn, latency where "up" is bad).
- If the finding fails the materiality or significance check, say so plainly and downgrade the confidence rather than forcing an insight.
Step 2: Establish the So-What
State what the finding means for the business in one sentence — the interpretive leap the reader needs.
- Use the pattern: "[Metric] [moved] [amount], which means [business consequence]."
- The consequence must connect to money, users, risk, or a stated goal — not restate the number.
- Distinguish leading vs. lagging: is this a signal of a future outcome or a record of a past one?
Step 3: Form Root-Cause Hypotheses
Offer 2–4 plausible explanations, ranked by likelihood, not a single guess presented as fact.
- For each hypothesis: state the mechanism (why it would cause this) and the evidence that supports or weakens it.
- Cross-reference the known variables input — attribute to a launch/campaign/seasonality only if timing aligns.
- Explicitly separate correlation from causation; flag any where the causal claim is unproven.
- Name the one confirmatory analysis that would validate the leading hypothesis.
Step 4: Quantify the Impact
Put a number on what this is worth or what it costs.
- Translate the metric change into a business unit: revenue, cost, users retained/lost, hours, or risk exposure.
- Show the arithmetic (e.g., "2pp drop in checkout conversion × 40k monthly sessions × $58 AOV ≈ $46k/month").
- If exact figures aren't available, give a defensible range and label the assumptions.
- If the finding is directional only, say "not yet sizable" rather than inventing precision.
Step 5: Recommend an Action
Convert the insight into one primary recommendation plus fallbacks.
- Primary action: What to do, specific enough to assign. Use a verb + object + expected outcome.
- Owner: The function or role that should execute (Growth, Eng, Product, Finance).
- Effort/impact tag: Quick win / Bigger bet / Investigate first.
- What to watch: The one metric that confirms the action worked.
- Where the finding is inconclusive, the primary action is a targeted investigation — not "monitor."
Step 6: Assign Confidence
Close with a calibrated confidence level so the reader knows how hard to lean on this.
- High: significant sample, clear driver identified, impact quantified.
- Medium: real signal, plausible but unconfirmed driver, estimated impact.
- Low: small/early signal or competing explanations — treat as a lead to investigate.
- State the single thing that would raise confidence one level.
Output Format
- Structure: A tight insight brief, one finding per brief; multiple findings each get their own brief, ordered by impact
- Length: 200–350 words per insight
- Format: Structured document with bolded labels; the headline and so-what in prose, hypotheses and action as short bullets
- Sections: Headline (one-line insight) → So-What → Why It's Happening (ranked hypotheses) → Impact (quantified) → Recommended Action (with owner, effort/impact tag, watch metric) → Confidence
- Tone: Analytical and decisive — the voice of an analyst who has already done the thinking and is handing over a recommendation, not raw output. Plain language for the stated audience; caveats stated once and clearly, never hedged into meaninglessness.
Quality Checklist
Before delivering output, verify:
- [ ] The headline states an implication, not just a number
- [ ] Absolute AND relative change are both shown, and materiality was assessed
- [ ] At least 2 root-cause hypotheses are given and ranked — no single unproven cause stated as fact
- [ ] Impact is expressed in a business unit (revenue, users, cost, risk) with the arithmetic or a labeled range shown
- [ ] The recommended action names an owner and is specific enough to assign
- [ ] A confidence level (High/Medium/Low) is assigned with the one lever that would raise it
Notes
- The most common failure is confusing "explained the number" with "gave an insight" — if the reader could not make a decision from your output, Step 2 and Step 5 need more work.
- Resist attributing every metric move to the most recent launch; timing correlation is a hypothesis, not a conclusion.
- For exec audiences, lead with impact and action; for analyst peers, keep the hypothesis and significance detail. Same skill, adjusted emphasis.
- When several findings compete, rank by quantified impact — a small but confident revenue insight often beats a large but low-confidence one.
How to use this skill
This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.
Installing in Claude Code:
~/.claude/skills/<folder-name>/SKILL.md— auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)<your-project>/.claude/skills/<folder-name>/SKILL.md— auto-loads only when working in that project (recommended for project-specific skills)
Restart Claude Code or open a new session to register.
Using in Claude.ai, ChatGPT, Gemini, or other AI tools:
- Drag this file into a new chat, or copy and paste its contents
- Send your task or request as your next message — the AI will apply the skill where it fits
For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills