Most dashboards get plenty of attention in their first week and very little after that. It’s rarely because the charts are ugly. It’s because the dashboard answers questions nobody is asking, or takes too long to read, or shows a number that disagrees with finance’s spreadsheet.
Good data visualisation is less about styling than about decisions: who is looking, what they need to decide, and the fastest honest way to show them.
Start with the decision, not the data
Before choosing a single chart, write down three things for each view:
- Who looks at it, and how often. An owner on Monday morning, a sales rep before calls, a finance lead at month end.
- The question they bring. “Which accounts need a call this week?”
- What they’ll do differently depending on the answer.
If you can’t fill in the third line, the chart is a curiosity rather than a tool. Leave it out, or move it to a secondary tab.
This also settles the most common design argument: one dashboard for everyone, or several. Different people make different decisions, so they need different views of the same data. Our demo writes one briefing for the owner and one for each sales rep, all built on the same data model, so their numbers always agree.
Pick the chart for the comparison
Every chart is a comparison: this month against last, this region against that one, actual against target. Choose the chart that makes that comparison easiest to see.
Research on graphical perception, starting with William Cleveland and Robert McGill’s work in the 1980s, found that people judge position along a common scale most accurately, then length, then angle and area, with colour intensity among the least accurate. In practice:
- Change over time: a line chart.
- Comparing categories: a bar chart, sorted by value, usually horizontal so the labels are readable.
- Parts of a whole: a stacked bar or a sorted bar of shares. Pie charts only work for two or three slices.
- Exact values people will look up: a table. A well-formatted table is a legitimate visualisation, not a failure to make a chart.
- A headline number: always with its comparison, against last period or against target. Never on its own.
Avoid gauges, 3D effects and dual-axis charts unless you have a strong reason. Each makes comparisons harder or easier to misread.
Context turns numbers into information
“£42,300 revenue this month” tells you nothing on its own. Is that good? Compared with what?
Give every headline number a reference point: the same period last year, the target, the average of peers. And mark the events that explain the shape of a line. In our demo, the North region’s wholesale revenue halves after its sales rep leaves. Plotted by month, the timing is plain to see; a note marking the departure makes it impossible to miss.
Titles can do more work too. “Revenue by region” describes the chart. “North wholesale revenue has halved since the rep left” tells the reader what to see. Use finding-style titles for reports about a specific story. For general views that refresh every day, a descriptive title with a line of context is safer, because a finding in the title goes stale the moment the data moves.
Remove whatever doesn’t help the comparison
Every element on the screen should earn its place.
- Make colour mean something. Grey for context, one accent colour for what matters. Keep red and green for bad and good, and never rely on them alone: around 1 in 12 men have some form of red-green colour blindness.
- Keep colours consistent. If wholesale is green on one chart, it’s green on every chart.
- Label directly. A label at the end of a line beats a legend the reader has to cross-reference.
- Start bar charts at zero. Line charts can start elsewhere if the axis is clearly labelled.
- Round sensibly. £42.3k, not £42,317.84.
Overview first, detail on demand
Ben Shneiderman’s rule from 1996 still holds: overview first, zoom and filter, then details on demand. Put the few numbers that answer “are we on track?” at the top, the charts that explain why underneath, and the detailed tables at the bottom or one click away.
Keep filters few and obvious. A dashboard with twelve filters has become a query tool, and the people who want a query tool should be given a proper one.
Make it fast. A dashboard that takes twenty seconds to load gets opened less, whatever is on it. Build it on modelled, pre-aggregated tables rather than raw data, and check it on a phone, because that’s where many owners look first.
Trust is a design feature
The quickest way to kill a dashboard is one number that disagrees with finance. After that, people doubt every number on it. Three things protect trust:
- One data layer underneath. The dashboard, the finance report and any AI tool use the same definitions, so they can’t disagree.
- Visible freshness. Show when the data was last refreshed and what period it covers.
- Definitions one click away. What’s in “revenue”, how an “active customer” is counted.
Where AI fits
AI doesn’t replace charts. It changes how people arrive at them.
Most people don’t want to browse a dashboard. They want to know what changed and what to do about it. In our demo, rule-based checks run against the data to find what deserves attention: key accounts that have gone quiet, accounts ordering less, leads nobody followed up. An AI model writes those findings up as a short briefing for each reader. Every number in the text is checked against the data before it’s published, and every point embeds the chart that proves it.
For questions the dashboard doesn’t cover, people can ask an AI assistant in plain English. It answers well when the data underneath is well modelled and described, which we cover in semantic data foundations for AI.
The chart becomes the evidence behind a sentence rather than something each reader has to interpret alone. That raises the bar for chart design: each one has to make its point at a glance.
Measure use, and prune
Most BI tools record who opens what. Look a month after launch. Views nobody opens should be fixed or removed. A small dashboard that gets used beats a comprehensive one that doesn’t.
Before you ship a dashboard
- Each view has a named audience, question and decision.
- Each chart type fits the comparison it makes.
- Every headline number has a reference point.
- Colour means something, and works for colour-blind readers.
- Freshness and definitions are visible.
- The numbers reconcile with the source of truth.
- It loads in a few seconds, on a phone too.
- There’s a date in the diary to check usage.
See it for yourself
Our live demo has a working dashboard and an AI-written owner briefing, both open with no sign-up. If your team has a dashboard nobody opens, tell us about it. It’s usually fixable.