analytics-insights

name	analytics-insights
description	Deep dive into product analytics — investigate a question, surface insights, build a data narrative. Use when you need to go beyond dashboards to understand what's happening.

Analytics & Insights

Go beyond dashboards to answer a specific product question in 1-2 hours. Claude helps you structure the investigation, identify the right cuts of data, spot patterns, and build a narrative that drives action. You provide the product context that turns numbers into meaning.

When to Use

• You have a specific question that your dashboards don't answer
• A metric moved and you need to understand why
• You're building a data-backed case for a product decision
• Leadership asked "what are the numbers telling us?" and you need depth, not a screenshot
• You need to segment data to understand different user behaviors
• Pre-launch baseline analysis or post-launch impact measurement

When NOT to Use

• You're doing a routine weekly check — use Metrics Review
• You're diagnosing a specific churn or retention problem — use Growth & Retention Diagnostics
• You need to design an experiment — use Experiment Design

The AI-Native Approach

Step	Time	Claude Does	You Do
Frame the question	10 min	Sharpen vague questions into specific, answerable investigations	Provide the business context and what decisions hinge on this
Design the analysis	15 min	Identify metrics, segments, time ranges, and comparison groups	Confirm data availability and source reliability
Analyze the data	30 min	Structure the analysis, spot patterns, flag anomalies	Provide the actual data, add qualitative context
Build the narrative	20 min	Draft audience-specific narrative with insights and recommendations	Validate interpretation, add "so what" and "now what"

Process

Step 1: Frame the Question (10 minutes)

Vague questions get vague answers. Start by making the question specific enough to answer with data.

I need to investigate: [vague question — e.g., "How are users engaging with the new feature?"]

Help me sharpen this into specific, answerable questions:

For each question:
- What metric answers this?
- What segments matter? (user type, cohort, geography, plan tier, etc.)
- What time range is relevant?
- What comparison would be meaningful? (before/after, segment A vs. B, us vs. benchmark)
- What would a "good" answer look like? What would a "bad" answer look like?

What decision will this analysis inform?
What would we do differently based on what we find?

Step 2: Design the Analysis (15 minutes)

For each question, design the analysis:

Data needed:
- Metric: [what we're measuring]
- Source: [where the data lives — analytics tool, database, export]
- Segments: [how to cut the data]
- Time range: [start, end, comparison period]
- Granularity: [daily, weekly, monthly]

Analysis approach:
- Trend analysis: how has this metric changed over time?
- Segment comparison: how do different user groups behave?
- Cohort analysis: how do users who started at different times compare?
- Funnel analysis: where are users dropping off?
- Correlation: what other metrics move with this one?

Known limitations:
- Data quality issues to watch for
- Segments that are too small for meaningful comparison
- External factors that could confound the analysis

Step 3: Analyze the Data (30 minutes)

Here's the data: [paste data, describe charts, or share exports]

Analyze:
- What's the headline? (one sentence summary of the finding)
- What patterns are visible? (trends, seasonality, step changes)
- What anomalies stand out? (unexpected spikes, drops, or divergences)
- How do segments differ? (which groups behave differently, and how)
- What's statistically meaningful vs. just noise?

Dig deeper:
- Can we identify a root cause or contributing factor?
- Is this a one-time event or a structural change?
- What would we need to investigate further to be confident?

Flag:
- Where the data is strong (large sample, clean measurement)
- Where the data is weak (small sample, proxy metric, known instrumentation issues)
- What we still don't know

Step 4: Build the Narrative (20 minutes)

Numbers don't drive action. Stories backed by numbers do.

Build an insights narrative for [audience — team, leadership, board]:

Structure:
1. The question we investigated and why it matters
2. Headline finding (one sentence — what did we learn?)
3. Supporting evidence (2-3 key data points with context)
4. What this means (interpretation — connect data to product/business impact)
5. What we should do (specific recommendation with expected impact)
6. What we don't know yet (gaps, caveats, next investigations)

Tone:
- For the team: detailed, include methodology, show your work
- For leadership: headline first, evidence second, ask/recommendation third
- For board: strategic frame, biggest number, trend direction, confidence level

Include:
- Data visualizations recommendations (what chart type best tells this story)
- Comparison context (benchmarks, historical, goals)
- Confidence level (how sure are we about this finding?)

Output

• Sharpened investigation questions tied to decisions
• Analysis design with metrics, segments, and methodology
• Pattern analysis with anomaly detection and root cause hypotheses
• Audience-specific narrative with insights and recommendations

Common Pitfalls

1. Answering the wrong question. "How many users clicked the button?" is easy to answer but rarely useful. "Did the button change user behavior in a way that moves our success metric?" is the real question. Start with the decision, work backward to the data.
2. Cherry-picking data. It's tempting to highlight the metric that supports your thesis. Show the full picture — including the metrics that didn't move or moved the wrong way. Credibility comes from honesty.
3. Confusing correlation with causation. Two metrics moving together doesn't mean one caused the other. Be explicit about whether you're showing correlation, suggesting causation, or proving it with an experiment.
4. Drowning in data. More data doesn't mean better insights. Pick the 3-5 metrics that answer the question. Everything else is appendix material.
5. Insights without action. "Activation is down 5%" is an observation. "Activation is down 5% because new users aren't completing onboarding step 3 — we should investigate why and test a simplified flow" is an insight. Always end with "so what" and "now what."

Related Skills

• Metrics Review — for routine weekly product health checks
• Growth & Retention Diagnostics — for specific churn/retention investigations
• Experiment Design — when the analysis surfaces a hypothesis worth testing
• Stakeholder Alignment — when the insights need to drive a cross-functional decision