Принципы дизайна дашбордов
Карьерник — квиз-тренажёр в Telegram с 1500+ вопросами для собесов аналитика. SQL, Python, A/B, метрики. Бесплатно.
Зачем это знать
Дашборд в Tableau / Metabase / Looker — front-facing product аналитика. Плохой дашборд = игнорируется командой. Хороший — цитируется в решениях.
Design — not after-thought. Это 50% value analytics работы.
Принципы
1. Audience first
Кто смотреть? Что нужно understand?
- Exec: 3-5 hero numbers + trend
- PM: feature metrics, experiment results
- Ops: real-time, detailed
- Analyst: explorable, drilldown
Different audiences → different dashboards.
2. Hierarchy visual
Top — most important. Bottom — detail.
- Top-left: primary KPI (eye goes here first)
- Sections: separated clearly
- Indent: detail within section
3. Consistency
- Colors consistent (red = bad, green = good)
- Same time periods everywhere
- Same decimal places
- Same alignment
Inconsistency confuses.
4. Clear labels
«Revenue (in thousands, YoY)» > «rev_k»
Axis labels, chart titles, units — always.
5. Context numbers
«CR 5%» — too little. «CR 5% (target 6%, prev month 4.8%)» — actionable.
6. Less is more
Remove:
- Gridlines (if not helpful)
- Excessive legend
- Redundant labels
- 3D effects (never useful)
Chartjunk → confuses.
Chart choices
Trend → Line
Best для time series.
Compare categories → Bar
Horizontal bars for long labels. Vertical для many categories.
Composition → Stacked bar или Donut
Pie charts limited (< 5 parts).
Correlation → Scatter
Two numeric vars.
Distribution → Histogram / Box plot
Flows → Sankey
Funnels, migrations.
Map → Choropleth / Bubble
Geo data.
Table
When user нужен precise values.
Не use fancy chart where simple works.
Color usage
Meaningful
- Red/Green: bad/good
- Blue: neutral
- Categorical palette: для different items
Accessibility
Colorblind-safe palettes. ColorBrewer good reference.
Minimal
2-3 colors typically. More — distracting.
No rainbow
Heat maps — single gradient, not rainbow.
Interactivity
Filters
Date, segment, country — affect все charts.
Drill-down
Click on segment → detail view.
Tooltips
Hover → extra info.
Don't overuse
Too interactive → users get lost.
Layout
Grid-based
Align edges, consistent widths.
Logical flow
Left-top → right-bottom reading pattern.
Group related
Similar metrics together.
White space
Don't cram. Breathing room.
Performance
Load time
< 5 seconds ideal. Users abandon.
Optimize queries
- Aggregations
- Indexes
- Materialized views
- Caching
Progressive loading
Hero numbers first, detailed charts after.
Мобильная адаптивность
Execs check phone. Dashboard должен работать.
- Larger fonts
- Single column
- Skip charts не critical
Data freshness
Show «last updated» time.
- Real-time: streaming
- Daily refresh: standard
- Weekly: some metrics
Mismatch users expectation → distrust.
Anti-patterns
Too many metrics
30 metrics на one page → paralysis.
Max 10 metrics. Rest — drill-down.
Truncated axes
Y-axis starting от 80% чтобы 85% выглядит как huge. Manipulation.
Start от 0 (мostly) unless specific reason.
Pie charts с 10+ slices
Unreadable. Use bar.
Default colors
Tableau default > random. But still customize.
Data-ink ratio низкий
Minimize non-data ink. Cut chrome.
Validation
Before ship
Review с intended users. «What's the story here?» Если не get — iterate.
Watch usage
Analytics на dashboard (meta). Low clicks → redesign.
Update
Monthly review — still useful? Metrics same? Audience same?
Пример: sales dashboard
Top section (hero)
- Today's revenue (+MoM%)
- Orders count
- AOV
- Conversion rate
Trend section
- Daily revenue last 30 days (line)
- Comparison vs last month / year
Breakdown
- By product category (bar)
- By region (map)
- By traffic source (bar)
Detail
Sortable table top 20 products.
На собесе
«Как design dashboard?»
Explain:
- Understand audience
- Prioritize metrics
- Choose charts per data type
- Layout с visual hierarchy
- Test и iterate
Show design thinking, не только technical.
Связанные темы
- Executive dashboard
- Как документировать dashboard
- Dashboard простыми словами
- Как сделать dashboard в DataLens
- Визуализация данных на собесе
FAQ
One dashboard на все?
No. Different audiences → different.
Interactive обязательно?
Depends. Exec — static often ok. Analyst — interactive.
Real-time обязательно?
Rarely. Daily refresh usually enough.
Тренируйте визуализацию — откройте тренажёр с 1500+ вопросами для собесов.