Data Visualisation Best Practices for Dashboards

By Grocito16 October 2026

Good data visualisation starts with the decision the dashboard supports, not the chart gallery. Choose the simplest chart that answers each question, keep axes honest, use a small and accessible colour palette, put the key numbers first and add context such as targets and comparisons. Then test with the people who will use it and fix what confuses them.

Key takeaways

Decide who will use the dashboard and what they will decide or do differently because of it.
Match the chart to the question: comparison, trend, composition, distribution or relationship.
Be honest with scales: bars start at zero, comparable charts share the same scale.
Use few colours, make sure they are readable by colour-blind users and keep their meaning consistent.
Show key numbers first, group related items, use whitespace and give every number context.
Agree metric definitions in one place, or two teams will show two different answers.

What should you decide before designing any chart?

Purpose comes before pixels. Ask:

Who is this for? An executive checking health weekly needs something different from an analyst investigating a drop.
What decision does it support? If nobody can name an action that follows from a chart, it is decoration.
How often is it used? Daily operational dashboards need fast scanning, while monthly review packs can carry more detail.
What does good look like? A number with no target or history cannot tell you anything.

Our guide to choosing the right metrics for business intelligence dashboards helps with the first step of picking what to show.

Which chart type fits which question?

Think in questions and then pick the form.

Comparison between categories: a bar chart, horizontal if the labels are long. Sort the bars so the pattern is obvious.
Change over time: a line chart for continuous trends. Columns can work for a small number of periods.
Composition, meaning parts of a whole: a stacked bar for a few parts, or a simple labelled figure. Be careful as soon as there are many parts.
Distribution, meaning how values are spread: a histogram or a box-style summary, which reveal outliers and skew that an average hides.
Relationship between two measures: a scatter plot, optionally with a trend line.
Single key figure: a large number with context, rather than a chart.
Detailed lookup: a table, because sometimes people need exact values.

When should you avoid pie charts, 3D effects and dual axes?

Pie charts are hard to read once there are more than a few slices, and people compare angles poorly. A sorted bar chart is usually clearer. A pie can be acceptable for two or three parts of a clear whole.
3D effects distort size and add nothing. Avoid them.
Dual-axis charts can imply a relationship that is only an artefact of the two scales. If you must use one, label both axes clearly. Often two small charts stacked together are more honest.
Decorative gauges and radar charts take space and make comparison difficult.

How do you keep axes and scales honest?

Bar charts should start at zero because the length of the bar represents the value. Cutting the axis exaggerates differences.
Line charts can use a non-zero start when the changes matter, but state it clearly.
When charts are meant to be compared, use the same scale on each.
Avoid changing the time window between charts without saying so.
Mark missing data and partial periods rather than letting a line fall off a cliff.

Trust is the dashboard's most valuable feature. Once people spot one misleading chart, they doubt all of them.

How should you use colour?

Use a limited palette. Grey for most things and one strong colour for what matters is a good starting point.
Keep meaning consistent: if blue is revenue on one chart, it should not mean cost on another. Reserve red and green for good and bad only if that is truly the meaning, and never rely on them alone.
Check contrast so text and marks are legible on the background.
Design for colour blindness: add labels, shapes or patterns so that colour is not the only signal.
Do not use a rainbow of colours for categories unless they genuinely need to be told apart.

How do you structure layout and hierarchy?

People read dashboards in a scanning pattern, usually starting top left.

Put the key numbers at the top.
Place trends and breakdowns below, grouped by topic.
Keep detail and tables at the bottom or on a drill-down page.
Use whitespace to separate groups rather than heavy borders.
Align items on a grid and keep sizes consistent.

If everything is the same size and colour, nothing stands out.

What about labels, annotations, sorting and small multiples?

Label directly on the chart where you can, instead of making people match a legend.
Write chart titles as statements or clear questions: "Orders by region this month" is better than "Chart 3".
Annotate unusual events, such as a campaign launch or a system outage, so people do not guess.
Sort bars by value unless there is a natural order, such as months or age bands.
Use small multiples, meaning a row of similar small charts with the same scale, to compare many categories without clutter. It beats one tangled chart with ten lines.

How do filters, drill-down and KPI cards work together?

Filters should answer common questions without overwhelming people. Offer a few well-chosen ones such as date range, region and product line, with clear defaults and a visible indicator of what is applied. Allow drill-down from summary to detail so people can follow up a surprise without leaving the page.

A KPI card should show the current value, the comparison (previous period, target or benchmark you have defined yourself) and the direction of change. A bare number of "4,200" means nothing without that context.

What performance and mobile points matter?

Tell users how fresh the data is and when it last refreshed. Real-time is expensive, and many decisions do not need it.
Keep load times short by limiting the number of heavy visuals per page and aggregating data in advance where possible.
On mobile, show fewer items, stack them vertically, use large tap targets and make the key numbers readable without zooming.
Do not shrink a desktop dashboard onto a phone and call it done.

How do you test a dashboard and govern definitions?

Sit with two or three real users and ask them to answer real questions using the dashboard. Note where they hesitate, misread or ask what a term means. Then fix those spots.

Governance matters as much as design. Keep a short glossary: what counts as a customer, an active user, a sale or a refund, and who owns each definition. If you are still working in spreadsheets, our guide to spreadsheets versus BI tools explains when it is time to move on, and our analytics and BI dashboard is built around the same principles.

What are the most common mistakes?

Too many charts on one page.
Vanity metrics that look good but drive no action.
No context: no target, comparison or time frame.
Inconsistent definitions across pages.
Heavy decoration and clashing colours.
Charts chosen because they look impressive.
Dashboards nobody owns, which slowly go out of date.

A worked example

Imagine an online retailer whose weekly dashboard has twenty-two charts, including a 3D pie chart of orders by city. The team rebuilds it around three questions: are we on target, what changed and where should we act? The top row shows sales, orders and returns against last week and the target. Below that sit one trend line and one sorted bar chart of category performance, with an annotated note for the festival sale. Detail moves to drill-down pages. Meetings get shorter because everyone reads the same page the same way.

FAQ

What is the best chart for showing change over time?

A line chart is usually the clearest choice for continuous trends because it shows direction and shape at a glance. For a small number of periods, columns also work. Add context such as a target or a previous period.

Why are pie charts discouraged in dashboards?

People find it hard to compare angles and areas, especially with many slices. A sorted bar chart makes the same comparison easier to read. A pie can still be acceptable for two or three parts of a clear whole.

How many charts should a dashboard have?

There is no fixed number, but fewer is generally better. Include only what supports the decisions the dashboard exists for, and move detail to drill-down pages.

How do I make a dashboard accessible?

Use sufficient contrast, do not rely on colour alone, add labels and text alternatives where possible and make text readable at normal sizes. Test with a colour-blind-friendly view and with keyboard navigation if the tool supports it.

Next steps

Pick one dashboard people already complain about. Write down who uses it and which decision it supports, remove everything that does not serve that decision and test the result with two real users. If you would like help designing a dashboard or the data pipeline behind it, you can contact us.