How to design a dashboard people can read at a glance
A practical guide to dashboards people can read in seconds: what to show, how to add context to numbers, which charts to use and how to test it.

Most dashboards fail in the first five seconds. Someone opens the page, sees fourteen cards, six charts and a table, and closes it again without learning anything. The data was all there. The design just never told them where to look or what it meant.
A good dashboard answers one question fast: "Is everything okay, and if not, where?" In this guide we walk through how to get there, from deciding what belongs on the screen to choosing charts, numbers and colours that people can read without squinting. It works whether you are designing an admin panel for a SaaS product, a client report, or an internal tool for a small team.
Start with the questions, not the data
The fastest way to build a bad dashboard is to open the database and ask "what can we show?" The answer is always "everything", and everything is unreadable. Start with the people instead.
Sit down with two or three real users and ask what they check first thing in the morning, what decision they make after looking, and what would make them worried. Write the answers as plain questions:
- Are we on track for this month's revenue target?
- Did anything break overnight?
- Which clients need attention today?
Each question becomes one area of the dashboard. If a metric does not answer any of the questions, it goes to a detail page or a report, not the main screen. This single step usually cuts the content in half.
Build a clear hierarchy in three layers
People scan dashboards top to bottom and left to right, so put the most important answer where the eye lands first. A layout that works for most products has three layers:
- Headline numbers. Three to five key figures across the top, each with a label, the current value and a comparison.
- Trends. One or two charts that explain why the headline numbers look the way they do.
- Details. A table or list of items that need action, such as overdue invoices or failed jobs.
Keep the top layer small. When you have eight "key" numbers, none of them are key. If stakeholders push back, offer a second tab or a filtered view rather than squeezing more cards into the first row.
Give every number context
A number on its own means nothing. "1,240 sign-ups" could be great or terrible. Every headline figure needs at least one point of comparison so the reader can judge it in a glance:
- Change against the previous period, shown as a percentage and an arrow.
- Progress towards a target, shown as a short bar or "82% of goal".
- A small sparkline showing the last few weeks.
Be careful with colour on these comparisons. Green for up and red for down only works when up is good. For churn, refunds or error rates, a rise is bad news, so the colour should follow meaning, not direction. Also round sensibly: "$48.2k" reads faster than "$48,213.57", and the extra precision rarely changes a decision.
Choose the simplest chart that answers the question
Most dashboard questions can be answered with four chart types. Pick by the question you are answering, not by what looks impressive:
- Line chart for change over time.
- Bar chart for comparing categories, sorted from largest to smallest.
- Stacked bar for parts of a whole, when there are only a few parts.
- Single number with a sparkline when the trend matters less than the current value.
Avoid pie charts with more than three slices, 3D effects and gauges. People are poor at comparing angles and areas, and gauges take a lot of space to show one number. Label lines directly at their end instead of using a separate legend, so readers do not have to match colours back and forth.
Use colour as a signal, not decoration
On a calm dashboard, colour should mean something. A practical approach is to keep most of the interface neutral (greys, one brand colour for the primary data series) and save strong colours for states that need attention.
If everything is highlighted, nothing is. Reserve red and amber for problems the reader should act on today.
Check that your status colours work for people with colour vision deficiencies. Pair colour with a second cue, such as an icon, a label like "Overdue" or a different line style. Test the contrast of chart labels and small text against WCAG AA, because dashboards are often read on laptop screens in bright offices.
Design the boring states too
Dashboards spend a lot of time in states designers forget to draw. Before handoff, make sure each card and chart has a design for:
- Loading: skeleton shapes that match the final layout, so the page does not jump.
- Empty: a short explanation of why there is no data yet and what to do about it.
- Error: which widget failed, when it last updated, and a way to retry.
- Stale data: a visible "Last updated" time, especially for numbers that sync once a day.
The last one matters more than it looks. People make decisions from dashboards, and a confident number that is two days old is worse than no number at all.
Make filters and time ranges obvious
Half of the "the dashboard is wrong" bug reports we have seen came down to a hidden filter. Show the active date range and filters in plain words near the top of the page, for example "Last 30 days · All regions". Use sensible defaults that match the most common question, and keep the same time range across all widgets unless a card clearly says otherwise.
On smaller screens, do not shrink everything. Stack the headline numbers first, let charts take the full width, and move detailed tables behind a "View all" link. A phone view that shows the three most important numbers well is more useful than a tiny copy of the desktop layout.
Test it with the five-second check
Before you call the design done, run a quick test. Show the dashboard to someone from the target team for five seconds, hide it, and ask: "How are things going, and what would you look at next?" If they cannot answer, the hierarchy or the comparisons need work.
Repeat the test with real data, not placeholder numbers. Real data has long names, zero values and outliers that break neat mockups, and it is better to find those in Figma than in production.
Conclusion
A readable dashboard is mostly about saying no. Start from the questions people actually ask, put a few headline numbers with context at the top, use simple charts and meaningful colour, and design the loading, empty and stale states as carefully as the happy path. Then check it with a real person for five seconds. If they can tell you what is going on, your dashboard is doing its job.


