What Makes a Good Dashboard?
A dashboard is not a data dump — it is a decision-support tool. The best dashboards answer a specific set of questions for a specific audience in as few visual elements as possible. Poor dashboards overload viewers with every available metric, use chart types that obscure rather than reveal patterns, and require the reader to do mental arithmetic to understand what they are looking at.
Good dashboard design is as much about what you leave out as what you include. Every element on a dashboard should earn its place by directly supporting a decision or action. If a chart cannot change what the viewer will do, it should be removed.
The Five Questions Before You Build
Before opening any BI tool, answer these questions. They determine every design decision that follows.
| Question | Why It Matters | Example Answer |
|---|---|---|
| Who is the audience? | Executives need summaries; analysts need drill-down; operations teams need alerts | VP of Sales — weekly review meeting |
| What decision does this support? | Without a clear decision, there is no way to judge what belongs on the dashboard | Where should we focus sales effort next quarter? |
| What is the primary metric? | One North Star prevents attention fragmentation | Revenue vs. target by region |
| What is the refresh cadence? | Real-time data costs more to maintain and is only useful if decisions are made in real time | Weekly refresh is sufficient for this use case |
| What action does a bad number trigger? | If no one can act on a red metric, that metric should not be on the dashboard | Flag region for pipeline review call |
Choosing the Right Chart Type
| Task | Best Chart Type | Avoid |
|---|---|---|
| Show change over time | Line chart | Bar chart for dense time series; pie chart |
| Compare values across categories | Horizontal bar chart (sorted) | 3D bar, pie/donut with many slices |
| Show part-to-whole proportions (≤5 categories) | Stacked bar or donut chart | Pie chart with many thin slices |
| Show distribution of a continuous variable | Histogram or box plot | Line chart; pie chart |
| Show relationship between two numeric variables | Scatter plot | Line chart (implies order) |
| Show a single KPI vs target | Big number + sparkline + variance label | Gauge chart (low data-ink ratio) |
| Show geographic distribution | Choropleth map (colour = intensity) | Bubble map when proportions matter more than location |
| Show correlation across many variables | Heat map | Matrix of scatter plots (too busy at scale) |
Layout and Visual Hierarchy
Viewers read dashboards in an F-pattern or Z-pattern — top-left first, then right, then down. The most important information should occupy the top-left position. Supporting context flows beneath or to the right.
A practical layout for an executive dashboard follows three tiers. The first tier (top row) contains three to five KPI scorecards: big numbers with trend indicators and variance vs. target. The second tier (middle) contains the primary trend chart — usually a time series of the main metric broken down by the most important dimension. The third tier (bottom) contains supporting detail charts that explain the drivers behind the top-level numbers. Filters and date pickers sit at the top right or in a left sidebar, never scattered across the canvas.
White space is not wasted space — it reduces cognitive load and directs attention. Resist the urge to fill every pixel. Padding between charts should be consistent (16px or 24px grids work well), and chart titles should describe the insight, not just the metric name. "Revenue is up 12% YoY in EMEA" tells the viewer what to think; "Revenue by Region" makes them figure it out themselves.
Colour Best Practices
Colour is the most misused element in dashboard design. Follow these rules to use it effectively rather than decoratively.
| Rule | Rationale |
|---|---|
| Use one primary colour with one accent | Multiple bright colours compete for attention; one accent draws the eye to what matters |
| Reserve red and green for performance signals only | Red = bad, green = good is universal; using these colours decoratively creates false alarms |
| Use sequential palettes for continuous data (light → dark) | Viewers intuitively read darker as more; avoid rainbow palettes that imply false ordering |
| Use diverging palettes when data crosses zero | Two-colour diverging scale (red → white → blue) clearly shows positive and negative values |
| Ensure 4.5:1 contrast ratio for text | Accessibility requirement; grey text on white background fails this test |
| Test for colour blindness (8% of men) | Red-green combinations are invisible to deuteranopes; use shape or pattern as a backup encoding |
Defining Metrics Consistently in SQL
A dashboard is only as trustworthy as the SQL behind it. Inconsistent metric definitions are the most common cause of dashboard credibility failures — when two charts on the same page show different revenue figures, viewers lose trust in all the numbers.
-- Canonical revenue metric: always use the same logic
-- Exclude refunded orders, internal test accounts, and VAT
WITH clean_orders AS (
SELECT
o.order_id,
o.user_id,
DATE_TRUNC('week', o.created_at) AS week,
o.region,
o.gross_revenue
- COALESCE(r.refund_amount, 0) AS net_revenue
FROM orders o
LEFT JOIN refunds r USING (order_id)
WHERE o.is_test_order = FALSE
AND o.status NOT IN ('cancelled', 'fraud')
)
SELECT
week,
region,
SUM(net_revenue) AS revenue,
COUNT(DISTINCT order_id) AS orders,
COUNT(DISTINCT user_id) AS customers
FROM clean_orders
GROUP BY 1, 2
ORDER BY 1, 2;
Centralise these definitions in a semantic layer (dbt metrics, Looker LookML, or a shared SQL view) so that every chart on every dashboard uses exactly the same logic.
Common Dashboard Mistakes
| Mistake | Impact | Fix |
|---|---|---|
| Truncated y-axis starting above zero | Exaggerates small differences; misleads viewers about magnitude | Always start bar chart y-axis at zero; use line charts for trend emphasis |
| Too many metrics | Viewer cannot identify what to focus on | Limit to 5–7 KPIs; move the rest to a detail page |
| Missing context (no target, no prior period) | A number without a reference point is meaningless | Always show vs. target or vs. same period last year |
| Unlabelled axes | Viewer cannot interpret the chart without asking questions | Always label axes with units; include a data-as-of date |
| Pie charts with many slices | Humans cannot accurately compare angles beyond ~5 segments | Use sorted horizontal bar chart instead |
| Interactive filters with no default | Dashboard loads blank; users do not know what to select | Always set sensible defaults (last 30 days, all regions) |
Summary
Great dashboards are built backwards from decisions, not forwards from available data. Define the audience and the question first, choose the minimal set of metrics that answer it, arrange them in a clear visual hierarchy, use colour purposefully rather than decoratively, and anchor every number in a comparison that gives it meaning. A dashboard that a busy executive can read in 30 seconds and act on immediately is worth infinitely more than a sprawling canvas of every metric the data team can compute.
Create a free reader account to keep reading.