Chart Selection Decision Guide

Pick the right chart type based on what you're trying to communicate. This guide covers the 90% case — when in doubt, use a bar chart.

The Decision Framework

Ask yourself: "What relationship am I showing?"

Decision Tree

What are you showing?
│
├─ A single important number?
│  └─ KPI CARD with trend indicator
│
├─ How something changes over time?
│  ├─ Continuous (daily/hourly)? → LINE CHART
│  ├─ Discrete periods (monthly)? → COLUMN CHART
│  └─ Cumulative? → AREA CHART
│
├─ Comparison between categories?
│  ├─ How many categories?
│  │  ├─ 2-5 → GROUPED BAR CHART
│  │  ├─ 6-15 → HORIZONTAL BAR (sorted)
│  │  └─ 15+ → TABLE or TOP-N BAR
│  └─ Part of a whole?
│     ├─ 2-4 parts → DONUT CHART (only exception to no-pie rule)
│     └─ 5+ parts → STACKED BAR (100%)
│
├─ Distribution of values?
│  ├─ One variable? → HISTOGRAM
│  ├─ Compare distributions? → BOX PLOT
│  └─ Density estimation? → VIOLIN PLOT
│
├─ Relationship between variables?
│  ├─ Two continuous? → SCATTER PLOT
│  ├─ Add a third via size? → BUBBLE CHART
│  └─ Many variables? → HEATMAP MATRIX
│
├─ Geographic data?
│  ├─ Regional aggregates? → CHOROPLETH MAP
│  ├─ Point locations? → SYMBOL MAP
│  └─ Routes/flows? → FLOW MAP
│
└─ Progress toward goal?
   ├─ Single metric vs target? → BULLET CHART
   └─ Multiple metrics vs targets? → TABLE with conditional formatting

Chart-Specific Guidelines

Line Charts

• Maximum 5-7 lines before it becomes unreadable
• Always include zero baseline for rate/percentage metrics
• Use solid lines for primary, dashed for comparison/forecast
• Mark significant events with vertical annotation lines
• Show confidence intervals as shaded bands

Bar Charts

• Sort by value (not alphabetical) unless there's a natural order
• Horizontal bars for labels longer than 10 characters
• Space between bars = 40-50% of bar width
• Use a single color unless encoding an additional dimension
• Always start axis at zero (unlike line charts, truncating bars is deceptive)

KPI Cards

Anatomy of a good KPI card:

┌─────────────────────────────┐
│  Revenue (MTD)              │  ← Label + time period
│  $2.4M                      │  ← Big number
│  ▲ 12% vs last month        │  ← Trend with comparison
│  Target: $2.8M              │  ← Context
└─────────────────────────────┘

Tables

• Use tables for detail/lookup, never as the primary visualization
• Right-align numbers, left-align text
• Add conditional formatting (data bars, color scales) for quick scanning
• Limit to 7-10 columns visible at once
• Include sorting and searching

Scatter Plots

• Always label axes clearly (not just field names)
• Add quadrant lines when showing 2x2 framework (high/low segmentation)
• Use size encoding sparingly (humans are bad at comparing areas)
• Color-encode a categorical dimension (max 5-7 categories)
• Add a trend line if showing correlation

Heatmaps

• Use sequential color scale for one-direction data (light → dark)
• Use diverging color scale for data with a meaningful center (red → white → green)
• Sort both axes meaningfully (not alphabetical)
• Include the actual values as text labels if cells are large enough

Color Usage by Chart Type

Common Mistakes by Chart Type

Pie Charts (Almost Always Wrong)

• Problem: Humans cannot accurately compare arc angles or areas
• Exception: Showing 2-3 parts where one dominates (75%/25% split)
• Alternative: Horizontal bar chart, always

Dual-Axis Charts

• Problem: Implies correlation between unrelated metrics; scale manipulation
• When acceptable: Same unit, related metrics (revenue + cost → profit context)
• Alternative: Two charts stacked vertically, aligned on x-axis

Stacked Area Charts

• Problem: Inner bands are hard to read because they don't share a flat baseline
• When acceptable: Showing total composition over time where total matters
• Alternative: Small multiples (one line chart per category)

Treemaps

• Problem: Hard to compare rectangles of similar size; no natural reading order
• When acceptable: Showing hierarchical part-of-whole with 2+ levels
• Alternative: Sunburst diagram or nested horizontal bars

Dashboard Layout Patterns

Pattern 1: KPI + Trend + Detail (Most Common)

┌──────────┬──────────┬──────────┬──────────┐
│  KPI 1   │  KPI 2   │  KPI 3   │  KPI 4   │  ← Status row
├──────────┴──────────┴──────────┴──────────┤
│                                            │
│           Main Trend Chart                 │  ← Analysis
│                                            │
├─────────────────────┬──────────────────────┤
│  Breakdown Chart 1  │  Breakdown Chart 2   │  ← Detail
├─────────────────────┴──────────────────────┤
│              Detail Table                   │  ← Export
└────────────────────────────────────────────┘

Pattern 2: Comparison Layout

┌──────────────────────┬─────────────────────┐
│   Current Period     │   Previous Period    │  ← Side by side
├──────────────────────┴─────────────────────┤
│              Variance Chart                  │  ← What changed
├────────────────────────────────────────────┤
│              Top Movers Table               │  ← Biggest changes
└────────────────────────────────────────────┘

Pattern 3: Funnel / Journey Layout

┌────────────────────────────────────────────┐
│              Funnel Visualization           │  ← The flow
├────────────┬───────────────┬───────────────┤
│  Stage 1   │   Stage 2     │   Stage 3     │  ← Stage details
│  Metrics   │   Metrics     │   Metrics     │
├────────────┴───────────────┴───────────────┤
│           Drop-Off Analysis                 │  ← Where we lose them
└────────────────────────────────────────────┘

Pattern 4: Geographic Layout

┌─────────────────────────┬──────────────────┐
│                         │  Region KPIs     │
│       Map               │  ┌────────────┐  │
│                         │  │ North: $2M  │  │
│                         │  │ South: $1M  │  │
│                         │  └────────────┘  │
├─────────────────────────┴──────────────────┤
│           Regional Trend Chart              │
└────────────────────────────────────────────┘

Tool-Specific Recommendations

Tableau

• Use container-based layouts for responsive behavior
• Leverage parameter actions for interactivity without separate filter controls
• Small multiples via row/column shelves, not duplicated worksheets
• Use Level of Detail (LOD) expressions instead of table calculations when possible

Looker

• Single-value tiles for KPIs with value_format for readability
• Use listen filters to link tiles, not cross-filtering
• Merged results for combining explores in one tile
• Dashboard-level filters with required_fields for performance

Power BI

• Use bookmarks for guided navigation between views
• Sync slicers across pages for consistent filtering experience
• Custom tooltips for progressive disclosure on hover
• Field parameters for user-controlled measure switching

Dashboard Design Principles

A field-tested framework for building dashboards that people actually use. These principles apply regardless of tool (Tableau, Looker, Power BI) or domain.

Principle 1: One Dashboard, One Question

Every dashboard must have a single primary question it answers. If you can't state that question in one sentence, your dashboard is trying to do too much.

Good primary questions:

• "Are we on track to hit this quarter's revenue target?"
• "Which marketing channels are delivering qualified leads efficiently?"
• "Where are users dropping off in the onboarding flow?"

Bad primary questions:

• "How is the business doing?" (too broad)
• "What happened?" (too vague)

Implementation

Place the answer to the primary question in the top-left quadrant of the dashboard. This is where the eye lands first (F-pattern reading). Use a large KPI card or a single trend line with a clear up/down indicator.

Principle 2: The 5-Second Rule

A viewer should understand the dashboard's main message within 5 seconds of looking at it. This means:

The Hierarchy of Visual Weight

1. BIG NUMBERS (KPI cards)      → Instant comprehension
2. Trend lines                   → Direction and velocity
3. Bar charts                    → Comparison
4. Tables                        → Detail (put below the fold)

Place elements in this order from top to bottom. Heavy-reading elements (tables, detailed breakdowns) go below the fold.

Principle 3: Progressive Disclosure

Structure information in three layers:

Layer 1: Summary (Above the Fold)

• KPI cards with period-over-period comparison
• One or two key trend charts
• Status indicators (green/yellow/red)
• No interaction required to understand the situation

Layer 2: Analysis (Below the Fold or One Click Away)

• Breakdown charts (by segment, region, product)
• Comparison views (budget vs actual, this year vs last)
• Filter interactions to slice the data
• Requires some interaction to explore

Layer 3: Detail (Drill-Through or Linked Dashboard)

• Full data tables with export capability
• Individual record-level views
• Exception lists and anomaly details
• Reached through explicit drill-down action

Principle 4: Consistent Visual Encoding

Once you assign a meaning to a visual property, never break that contract.

Color Assignments

Green  = Good / On Track / Growth
Red    = Bad / Behind / Decline
Blue   = Primary / Current Period
Gray   = Secondary / Comparison Period
Orange = Warning / Needs Attention

Typography

KPI Value:      28-36pt, Bold, Dark
KPI Label:      12-14pt, Regular, Medium Gray
Chart Title:    14-16pt, Semi-Bold, Dark
Axis Labels:    10-12pt, Regular, Medium Gray
Annotations:    11pt, Italic, Dark Gray

Spacing

Section padding:     24px
Card internal pad:   16px
Between cards:       12px
Chart margins:       16px top, 12px sides

Principle 5: Mobile-Aware Design

Design for the narrowest screen first, then expand.

Responsive Breakpoints

Mobile Simplification Rules

1. KPI cards: Show value + trend arrow, hide sparklines

2. Charts: Show one at a time with swipe navigation

3. Filters: Collapse into a filter drawer, show active count

4. Tables: Convert to card layout (one row = one card)

Principle 6: Filter Design

Filters are the primary interaction model. Design them carefully.

Filter Placement

• Global filters (date range, business unit): Top of page, always visible
• Section filters (segment, category): Top of the section they control
• Chart-specific filters (outlier toggle): Inside the chart container

Filter Defaults

Every filter must have a sensible default:

• Date: Current month or last 30 days (not "all time" — that kills performance)
• Segments: All selected (inclusive default)
• Comparisons: Same period last year

Cross-Filter Behavior

Document how filters interact:

• Does selecting a bar in Chart A filter Chart B? (Use this sparingly)
• Do date range changes affect all charts? (Usually yes)
• Are there filter dependencies? (Region → City cascading)

Principle 7: Annotation and Context

Numbers without context are meaningless. Always provide:

Comparison Points

• Period-over-period (vs last month, vs last year)
• Target/budget/forecast comparison
• Benchmark (industry average, top quartile)

Annotations

• Mark significant events on time series (product launch, outage, campaign start)
• Add brief text explanations for anomalies
• Include "last updated" timestamps

Thresholds

Define and display thresholds for KPIs:

Revenue Growth:    Green >10%, Yellow 0-10%, Red <0%
Customer Churn:    Green <2%, Yellow 2-5%, Red >5%
NPS Score:         Green >50, Yellow 30-50, Red <30

Principle 8: Performance and Load Time

A dashboard that takes 10 seconds to load won't be used daily.

Performance Rules

1. Limit to 10 queries per dashboard (fewer is better)

2. Pre-aggregate where possible (use materialized views / extracts)

3. Default date range should query <1M rows

4. Use incremental refresh for real-time-ish data

5. Lazy-load below-the-fold content

Anti-Patterns That Kill Performance

• Calculated fields on unbounded date ranges
• Cross-database joins at query time
• Blending more than 3 data sources
• LOD calculations (Tableau) on high-cardinality dimensions
• Unindexed filters on large tables

Principle 9: Naming Conventions

Be boringly consistent with names.

Dashboards

{Domain} - {Focus} - {Audience}

Example: Marketing - Campaign Performance - Team Lead

KPI Cards

{Metric Name} ({Time Period})

Example: Revenue (MTD), New Users (Last 7 Days)

Charts

{Metric} by {Dimension} — {Time Period}

Example: Conversion Rate by Channel — Monthly

Filters

Select {Dimension}

Example: Select Region, Select Date Range

Principle 10: Documentation and Handoff

Every dashboard needs a companion spec. Minimum documentation:

## Dashboard: [Name]
**Primary Question:** [What this dashboard answers]
**Audience:** [Who looks at this and how often]
**Data Sources:** [Tables/views used]
**Refresh Cadence:** [How often data updates]
**Owner:** [Who maintains this]
**KPI Definitions:**
  - [Metric 1]: [exact calculation logic]
  - [Metric 2]: [exact calculation logic]
**Known Limitations:**
  - [What this dashboard does NOT show]
  - [Data delays or gaps]

Anti-Pattern Gallery