Agent Skills: Visualization Choice & Reporting

Visualization Choice & Reporting

Overview

Visualization choice & reporting matches visualization types to questions and data, then creates narrated dashboards that highlight signal and recommend actions.

Three core components:

1. Chart selection: Match chart type to question type and data structure (comparison → bar chart, trend → line chart, distribution → histogram, relationship → scatter, composition → treemap, geographic → map, hierarchy → tree diagram, flow → sankey)

2. Visualization best practices: Apply perceptual principles (position > length > angle > area > color for accuracy), reduce chart junk, use pre-attentive attributes (color, size, position) to highlight signal, respect accessibility (colorblind-safe palettes, alt text), choose appropriate scales (linear, log, normalized)

3. Narrative reporting: Lead with insight headline, annotate key patterns, provide context (vs benchmark, vs target, vs previous period), interpret what it means, recommend next actions

When to use: Data analysis, dashboards, reports, presentations, monitoring, exploration, stakeholder communication

Workflow

Copy this checklist and track your progress:

Visualization Choice & Reporting Progress:
- [ ] Step 1: Clarify question and profile data
- [ ] Step 2: Select visualization type
- [ ] Step 3: Design effective chart
- [ ] Step 4: Narrate insights and actions
- [ ] Step 5: Validate and deliver

Step 1: Clarify question and profile data

Define the question you're answering (What's the trend? How do X and Y compare? What's the distribution? What drives Z? What's the composition?). Profile your data: type (categorical, numerical, temporal, geospatial), granularity (daily, user-level, aggregated), size (10 rows, 10K, 10M), dimensions (1D, 2D, multivariate). See Question-Data Profiling.

Step 2: Select visualization type

Match question type to chart family using Chart Selection Guide. Consider data size (small → tables, medium → standard charts, large → heatmaps/binned), number of series (1-3 → standard, 4-10 → small multiples, 10+ → interactive/aggregated), and audience expertise (executives → simple with insights, analysts → detailed exploration).

Step 3: Design effective chart

For simple cases → Apply Design Checklist (clear title, labeled axes, legend if needed, annotations, accessible colors). For complex cases (multivariate, dashboards, interactive) → Study resources/methodology.md for advanced techniques (small multiples, layered charts, dashboard layout, interaction patterns).

Step 4: Narrate insights and actions

Lead with insight headline ("Revenue up 30% YoY driven by Enterprise segment"), annotate key patterns (arrows, labels, shading), provide context (vs benchmark, target, previous), interpret meaning ("Suggests product-market fit in Enterprise"), recommend actions ("Double down on Enterprise sales hiring"). See Narrative Framework.

Step 5: Validate and deliver

Self-assess using resources/evaluators/rubric_visualization_choice_reporting.json. Check: Does chart answer the question clearly? Are insights obvious at a glance? Are next actions clear? Create visualization-choice-reporting.md with question, data summary, visualization spec, narrative, and actions. See Delivery Format.

Question-Data Profiling

Question Types → Chart Families

| Question Type | Example | Primary Chart Families | |---------------|---------|------------------------| | Trend | How has X changed over time? | Line, area, sparkline, horizon | | Comparison | How do categories compare? | Bar (horizontal for names), column, dot plot, slope chart | | Distribution | What's the spread/frequency? | Histogram, box plot, violin, density plot | | Relationship | How do X and Y relate? | Scatter, bubble, connected scatter, hexbin | | Composition | What are the parts? | Treemap, pie/donut, stacked bar, waterfall, sankey | | Geographic | Where is it happening? | Choropleth, bubble map, flow map, dot map | | Hierarchical | What's the structure? | Tree, dendrogram, sunburst, circle packing | | Multivariate | How do many variables interact? | Small multiples, parallel coordinates, heatmap, SPLOM |

Data Type → Encoding Considerations

• Categorical (product, region, status): Use position, color hue, shape. Bar length better than pie angle for accuracy.
• Numerical (revenue, count, score): Use position, length, size. Prefer linear scales; use log only when spanning orders of magnitude.
• Temporal (date, timestamp): Always use consistent intervals. Annotate events. Show seasonality if relevant.
• Geospatial (lat/lon, region): Use maps for absolute location; use tables/charts if geography not central to insight.

Chart Selection Guide

| Question Type | Chart Types | When to Use | |---------------|-------------|-------------| | Comparison | Bar (horizontal), Column, Grouped bar, Dot plot, Slope chart | Categorical → Numerical. Horizontal bar for long names/ranking. Grouped for 2-3 metrics. Slope for before/after. | | Trend | Line, Area, Sparkline, Step, Candlestick | Time → Numerical. Line for continuous trends. Area for cumulative/part-to-whole. Sparkline for inline. Step for discrete changes. | | Distribution | Histogram, Box plot, Violin, Density plot | Numerical → Frequency. Histogram for shape/outliers. Box for quartiles across groups. Violin for full density. | | Relationship | Scatter, Bubble, Hexbin, Connected scatter | Numerical X → Numerical Y. Scatter for correlation. Bubble for 3rd/4th variable (size/color). Hexbin for dense data. | | Composition | Treemap, Pie/Donut, Stacked bar (100%), Waterfall, Sankey | Parts of whole. Treemap for hierarchy. Pie for 2-5 categories (part-to-whole key). Waterfall for cumulative. Sankey for flow. | | Geographic | Choropleth, Bubble map, Flow map | Spatial patterns. Choropleth for regions. Bubble for precise locations. Flow for origin-destination. | | Multivariate | Small multiples, Heatmap, Parallel coordinates | Many variables. Small multiples for consistent comparison. Heatmap for matrix (time×day). Parallel for dimensions. |

Design Checklist

Essential Elements

• [ ] Insight headline title: Not "Revenue by Month" but "Revenue Up 30% YoY, Driven by Enterprise"
• [ ] Clear axis labels with units: "Revenue ($M)", "Month (2024)", not just "Revenue", "Date"
• [ ] Legend if multiple series: Position near chart, use direct labels on lines when possible
• [ ] Annotations for key points: Arrows, labels, shading for important events/patterns
• [ ] Source and timestamp: "Source: Analytics DB, as of 2024-11-14" builds trust

Perceptual Best Practices

• [ ] Start Y-axis at zero for bar/column charts (to avoid exaggerating differences)
• [ ] Use position over angle/area (bar > pie for accuracy, scatter > bubble when size isn't critical)
• [ ] Colorblind-safe palette: Avoid red-green only; use blue-orange or add patterns
• [ ] Limit colors to 5-7 distinct hues (more requires legend lookup, slows comprehension)
• [ ] Use pre-attentive attributes (color, size, position) to highlight signal, not decoration

Declutter

• [ ] Remove chart junk: No 3D, no gradients, no heavy gridlines, no background images
• [ ] Mute non-data ink: Light gray gridlines, thin axes, subtle colors for reference lines
• [ ] Use white space: Don't cram; let data breathe

Accessibility

• [ ] Alt text describing insight: "Line chart showing revenue grew from $2M to $2.6M (30% increase) from Q1 to Q4 2024, with Enterprise segment contributing 80% of growth."
• [ ] Sufficient contrast: Text readable, lines distinguishable
• [ ] Patterns in addition to color for critical distinctions (dashed/solid lines, hatched fills)

Narrative Framework

Structure: Headline → Pattern → Context → Meaning → Action

1. Headline (one sentence, insight-first):

• Not: "This chart shows monthly revenue."
• But: "Revenue grew 30% YoY, driven by Enterprise segment."

2. Pattern (what do you see?):

• "Q1-Q2 flat at $2M/month, then steady climb to $2.6M in Q4."
• "Enterprise segment grew 120% while SMB declined 10%."

3. Context (compared to what?):

• "vs target: 15% above plan"
• "vs last year: Q4 2023 was $2.0M, now $2.6M"
• "vs industry: Our 30% growth vs 10% industry average"

4. Meaning (why does it matter?):

• "Suggests product-market fit in Enterprise; SMB churn indicates pricing mismatch."
• "If sustained, Q1 2025 could hit $3M/month."

5. Action (what should we do?):

• "Prioritize: Hire 2 Enterprise AEs, launch SMB annual plans to reduce churn."
• "Monitor: Enterprise win rate, SMB churn by plan type."

Example Full Narrative:

Headline: Enterprise revenue up 120% YoY while SMB declined 10%, resulting in overall 30% growth.

Pattern: Revenue grew from $2M/month (Q1) to $2.6M (Q4). Enterprise segment contributed $1.5M in Q4 (up from $680K in Q1), while SMB dropped from $1.3M to $1.1M.

Context: Total revenue 15% above plan. Enterprise growth (120%) far exceeds industry average (25%). SMB churn rate doubled from 5% to 10% in Q3-Q4.

Meaning: Strong product-market fit in Enterprise; SMB pricing or feature set may be misaligned. Enterprise is now 58% of revenue vs 34% in Q1, reducing diversification.

Actions:

1. Prioritize: Hire 2 Enterprise AEs for Q1, double down on Enterprise playbook
2. Fix: Launch SMB annual plans (Q1) to reduce churn; interview churned SMB customers to identify gaps
3. Monitor: Enterprise win rate, SMB churn by plan type, revenue concentration risk

Delivery Format

Create visualization-choice-reporting.md with these sections:

1. Question: The question you're answering with data (e.g., "How has revenue trended over the past year?")

2. Data Summary: Source, time period, granularity, dimensions, size (e.g., "Analytics DB, Jan-Dec 2024, monthly, revenue by segment, 24 rows")

3. Visualization:

• Chart type selected (e.g., "Multi-line chart with annotations")
• Rationale (why this chart? question type, data structure, chart advantages)
• Design decisions (Y-axis scale, labels, annotations, colors)
• Chart specification (embed image, code, or detailed spec with axes, series, annotations)

4. Narrative: (Headline → Pattern → Context → Meaning → Action structure from above)

• Headline: Insight-first one-liner
• Pattern: What you see
• Context: vs benchmark/target/history
• Meaning: Why it matters
• Actions: What to do next

5. Validation: Self-check with rubric (Clarity ✓, Accuracy ✓, Insight ✓, Actionability ✓, Accessibility ✓)

6. Appendix (optional): Raw data, alternatives considered, statistical tests, assumptions

See resources/template.md for full template with examples.

Common Mistakes

Chart Selection Errors

❌ Pie chart for >5 categories: Hard to compare angles accurately ✓ Use horizontal bar chart: Position on common scale is more accurate

❌ Line chart for categorical data: Implies continuity that doesn't exist (e.g., revenue by product) ✓ Use bar chart: Discrete categories

❌ 3D charts: Perspective distorts values, adds no information ✓ Use 2D with color/size: Clearer, more accurate

Design Mistakes

❌ Y-axis doesn't start at zero (bar chart): Exaggerates differences ✓ Start at zero for bar/column: Accurate visual proportion

❌ Dual Y-axes with different scales: Misleading correlations ✓ Use small multiples or index to 100: Compare shapes, not scales

❌ Rainbow color scheme: Not colorblind-safe, no perceptual ordering ✓ Sequential (light→dark) or diverging (blue→white→red) palette

Narrative Failures

❌ Title: "Revenue by Month": Descriptive, not insightful ✓ "Revenue up 30% YoY, driven by Enterprise": Insight-first

❌ No context: "Revenue is $2.6M" (vs what?) ✓ Add benchmark: "Revenue $2.6M, 15% above $2.25M target"

❌ Pattern without meaning: "Revenue increased" (so what?) ✓ Interpret: "Revenue up 30%, suggests Enterprise product-market fit, informs 2025 hiring plan"

❌ No actions: Ends with "interesting pattern" ✓ Recommend: "Hire 2 Enterprise AEs, investigate SMB churn"

Resources

• Simple cases: Use resources/template.md for question profiling → chart selection → narrative
• Complex cases: Study resources/methodology.md for dashboards, small multiples, interactive visualizations, advanced chart types
• Self-assessment: resources/evaluators/rubric_visualization_choice_reporting.json

Further reading:

• "Storytelling with Data" by Cole Nussbaumer Knaflic (chart choice, decluttering, narrative)
• "The Visual Display of Quantitative Information" by Edward Tufte (principles, chart junk, data-ink ratio)
• "Show Me the Numbers" by Stephen Few (dashboard design, perceptual principles)