skills/mckinsey/vizro/dashboard-design

dashboard-design

SKILL.md

Building Vizro Dashboards

A structured workflow for creating effective dashboards with Vizro.

How to Use This Skill

CRITICAL: Use this skill BEFORE implementation. After completing Steps 1-3, proceed to the dashboard-build skill for implementation and testing.

IMPORTANT: Follow steps sequentially. Each step builds on the previous.

Copy this checklist and track your progress:

Dashboard Development Progress:
- [ ] Step 1: Understand Requirements (define end user, dashboard goals, document decisions)
- [ ] Step 2: Design Layout & Interactions (wireframes, filter placement)
- [ ] Step 3: Select Visualizations (chart types, colors, KPIs)
- [ ] Next: Use dashboard-build skill for implementation and testing

Interaction style: When gathering requirements or making design decisions, use the AskUserQuestion tool to present options as numbered choices. This enables interactive selection rather than walls of text. Break complex decisions into focused questions with 2-5 clear options each.

Do not skip steps. Handle partial context as follows:

  • User has data but no requirements → Start at Step 1
  • User has requirements but no data → Ask for data or suggest sample data
  • User has wireframes → Validate Step 1 decisions, then proceed from Step 2
  • User has visual designs/mockups → Validate Steps 1-2 decisions, then proceed from Step 3
  • User asks to "just build it" → Explain value of steps, offer to streamline but not skip, ask for data or suggest sample data

For simple dashboards (single page, less than 5 charts): Steps 1-3 can be abbreviated but not skipped entirely.


Spec Files: Documenting Decisions

IMPORTANT: Each step produces a spec file in the spec/ directory to document reasoning, enable collaboration, and allow resumption in future sessions. Create the spec/ directory at project start.


Step 1: Understand Requirements

Goal: Define WHAT information is presented and WHY it matters.

Key Questions to Discuss

  1. Users: Who are the end users of this dashboard? Per user type: What decisions do they need to make? What task/job do they need to accomplish?
  2. Goals: What is the current problem to solve? What is the goal of this dashboard?
  3. Data: What sources are available? What's the refresh frequency?
  4. Structure: How many pages or views? What's the logical grouping?

Design Principles

  • Limit KPIs: 5 primary metrics per page maximum
  • Clear hierarchy: Overview → Detail → Granular (max 3 levels)
  • Persona-based: Different users may need different views
  • Decision-focused: Every metric should inform a decision

REQUIRED OUTPUT: spec/1_information_architecture.yaml

Save this file BEFORE proceeding to Step 2:

# spec/1_information_architecture.yaml
dashboard:
  name: [Name]
  purpose: [One sentence goal]
pages:
  - name: [Page Name]
    purpose: [What question does this answer?]
    kpis: [List of 3-5 key metrics]
data_sources:
  - name: [Source Name]
    type: [csv/database/api]
decisions:
  - decision: [What was decided]
    reasoning: [Why this choice was made]

Validation Checklist

Before proceeding to Step 2:

  • Every page has a clear, distinct purpose
  • KPIs are measurable and actionable
  • Data sources are accessible
  • User has confirmed the structure

Detailed guidance: See information_architecture.md; Anti-patterns: See common_mistakes.md section "Step 1: Requirements Mistakes"


Step 2: Design Layout & Interactions

Goal: Define HOW users navigate and explore data.

Vizro Navigation Architecture

Tier 1: Global Navigation
├── Multi-page sidebar (automatic in Vizro)
└── Page selection

Tier 2: Page-level Controls
└── Filters/Parameters in left collapsible sidebar

Tier 3: Component-level
├── Container-specific filters/parameters
├── Cross-filter, cross-highlight interactions
└── Export actions

Layout Strategy

Optimal Grid Configuration:

  • Always use row_min_height="140px" (at page or container level)
  • 12 columns recommended (not enforced) - flexible due to many divisors (1, 2, 3, 4, 6, 12)
  • Control height by giving components more rows

Component Sizing (based on 12-column grid, height = rows × 140px):

Component Columns Rows Height
KPI Card 3 1 140px
Small Chart 4 3 420px
Large Chart 6 4-5 560-700px
Table 12 (full) 4-6 560-840px

Exceptions - size based on content to render:

  • Text-heavy Card → treat like a chart (3+ rows)
  • Small Table (less than columns) → doesn't need full width
  • Button → 1 row is enough

Layout Rules:

  • Place 2-3 charts maximum per row (side-by-side)
  • Full-width ONLY for time-series line charts
  • Give charts minimum 3 rows (use *[[...]] * 3 pattern)
  • Use -1 for intentional empty cells

Filter Placement & Selectors

Filter needed across multiple visualizations?
├─ YES → Page-level (left sidebar)
└─ NO → Container-level (top of the container)

Choose appropriate selectors - don't default to Dropdown:

Data Type Selector Example
2-4 options RadioItems Region (N/S/E/W)
5+ options Dropdown Category (many)
Numeric range RangeSlider Price ($0-$1000)
Single number Slider Year (2020-2025)
Date DatePicker Order date
Multi-select Checklist Status (Active, Pending)

REQUIRED OUTPUT: spec/2_interaction_ux.yaml

Save this file BEFORE proceeding to Step 3:

# spec/2_interaction_ux.yaml
pages:
  - name: [Must match Step 1]
    layout_type: grid  # or flex
    grid_columns: 12
    grid_pattern: [[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3]] # Component placement

    containers:
      - name: [Container Name]
        has_own_filters: true/false
    filter_placement:
      page_level: [columns with selector types]
      container_level: [columns with selector types]
wireframe: |
  [ASCII wireframe for ALL pages and tab views]
decisions:
  - decision: [What was decided]
    reasoning: [Why this choice was made]

Validation Checklist

Before proceeding to Step 3:

  • Layout follows Vizro constraints
  • Filter placement is intentional and documented
  • User has been presented ASCII wireframes for every page and approved them

Wireframes & examples: See layout_patterns.md; Anti-patterns: See common_mistakes.md section "Step 2: Layout Mistakes"


Step 3: Select Visualizations

Goal: Choose appropriate chart types and establish visual consistency.

Chart Type Quick Reference

Data Question Recommended Chart
Compare categories Bar chart (horizontal preferred)
Show trend over time Line chart (12+ points)
Part-to-whole (simple) Pie/donut (2-5 slices ONLY)
Part-to-whole (complex) Stacked bar chart
Distribution Histogram or box plot
Correlation Scatter plot

Chart Anti-Patterns (Never Use)

  • 3D charts, Pie charts with 6+ slices, Dual Y-axis, Bar charts not starting at zero

Color Strategy

Primary Rule: Let Vizro handle colors automatically for standard charts.

When to specify colors:

  • Semantic meaning (green=good, red=bad)
  • Consistent entity coloring across charts
  • Brand requirements

Vizro Semantic Colors — two palettes available, pick one and use consistently:

# Option A: Teal/Green palette (softer, recommended for chart-heavy dashboards)
positive_color = "#00B5A9"  # Darkgreen-500
negative_color = "#EA5748"  # Red
warning_color = "#FFC107"  # Yellow
sum_color = "#3E495B"  # Grey

# Option B: Blue palette (bolder, recommended when positive = primary brand blue)
positive_color = "#097DFE"  # Blue-500
negative_color = "#EA5748"  # Red
warning_color = "#FFC107"  # Yellow
sum_color = "#3E495B"  # Grey

Semantic colors can be used in charts where the meaning is inherent to the visualization (e.g., waterfall charts for increase/decrease, bar charts showing profit vs loss). Use them for KPI status indicators, notifications, and any chart where positive/negative semantics are core to the message.

KPI Card Pattern

Use kpi_card() for simple metrics, kpi_card_reference() for comparisons. Use reverse_color=True when lower is better (costs, errors). NEVER put kpi_card or kpi_card_reference as a custom chart or re-build KPI cards as custom charts, use the built-in kpi_card and kpi_card_reference in Figure model instead. Only accept exceptions for when the KPI card is strictly not possible, for example when dynamically showing text as a KPI card.

Chart Title Pattern

IMPORTANT: Titles go in vm.Graph(title=...), NOT in plotly code.

REQUIRED OUTPUT: spec/3_visual_design.yaml

Save this file BEFORE proceeding to implementation (dashboard-build skill):

# spec/3_visual_design.yaml
visualizations:
  - name: [Chart Name]
    type: [bar/line/scatter/etc]
    needs_custom_implementation: true/false
    reason: [if custom: has_reference_line/needs_data_processing/etc]

color_decisions:
  - component: [Name]
    reason: [Why non-default color]
    colors: [List of hex codes]

kpi_cards:
  - name: [KPI Name]
    value_column: [column]
    format: [e.g., '${value:,.0f}']
    has_reference: true/false

decisions:
  - decision: [What was decided]
    reasoning: [Why this choice was made]

Validation Checklist

Before proceeding to implementation (dashboard-build skill):

  • Chart types match data types (no pie charts for time series)
  • No anti-patterns used
  • Custom chart needs are identified
  • Color usage is consistent and intentional

Chart decision trees: See chart_selection.md; Anti-patterns: See common_mistakes.md section "Step 3: Visualization Mistakes"

Reference Files

File When to Read
information_architecture.md Step 1: Deep dive on requirements
layout_patterns.md Step 2: Wireframes, component sizing
chart_selection.md Step 3: Chart decision trees
common_mistakes.md All steps: Anti-patterns to avoid

Quick Reference: Vizro Components

Components: Dashboard, Page, Container, Tabs, Graph, Figure, AgGrid, Card, Filter, Parameter, Selector, Button

Key Imports: import vizro.models as vm, from vizro import Vizro, import vizro.plotly.express as px, from vizro.tables import dash_ag_grid, from vizro.figures import kpi_card, kpi_card_reference, from vizro.models.types import capture

Weekly Installs
36
Repository
mckinsey/vizro
GitHub Stars
3.6K
First Seen
Jan 24, 2026
Installed on
opencode33
gemini-cli33
cursor33
github-copilot31
codex31
amp28