Dashboards examples worth studying before you start

These are twelve panel shapes, not twelve companies. Each one has a question it answers well, and a way it goes wrong when it is used for a question it does not fit. Choosing the shape is most of the design work, and it happens before any styling decision.

They are grouped by what the reader is trying to do, because that is how the choice is actually made.

What to take away

• Pick the shape from the reader's task, not from the data's type. The same column supports several of these.
• Half of these fail the same way: they are used to explain when they were built to notice, or the reverse.
• The failure mode listed with each shape is the thing to check before you publish it.

For noticing that something is off

The single number with a comparison. One figure, large, with its change beside it. Answers "is this where it should be." Fails when the comparison is left off, which turns it into a fact the reader cannot judge.

The run chart with an expected band. A time series with an agreed range drawn behind it. Answers "is this outside normal." Fails when the band is fitted to recent data instead of agreed in advance, at which point it moves to accommodate whatever happens.

The exception list. A table containing only the items that breach a rule, with the reason. Answers "what needs attention now." Fails when the rule is too loose and the list becomes long enough to ignore, which happens quickly.

The queue age histogram. Counts bucketed by how long items have been waiting. Answers "is anything stuck." Fails when it is replaced by an average wait, which stays flat while the tail grows.

For explaining what happened

Small multiples. The same chart repeated once per segment, at the same scale. Answers "is this everywhere or somewhere." Fails when the panels use independent scales, which makes a small segment look like a large one.

The composition chart over time. Stacked areas or bars showing shares. Answers "what is this total made of, and is the mix changing." Fails for reading any single component other than the bottom one, because the eye cannot follow a band whose baseline moves.

The waterfall. A bridge from one total to another through the contributions that changed it. Answers "why did this move." Fails when the components overlap or do not sum to the difference, which is common and rarely checked.

The funnel. Counts at successive stages. Answers "where do people stop." Fails whenever entry to a later stage does not require the earlier one, and fails again when the stages are measured over different windows.

For comparing things to each other

The ranked bar chart. Categories sorted by value. Answers "who is largest." Fails when the tail is long, at which point everything below the top few is a row of stubs conveying nothing.

The scatter with one point per entity. Two measures plotted against each other. Answers "which entities are unusual on both." Fails when the reader treats the visible relationship as causal, and fails harder when overlapping points hide density.

The leaderboard. A ranked table of people or teams. Answers "who is ahead" and is the most socially expensive shape on this list. Fails whenever the ranking is driven by territory, tenure, or account size rather than by anything the person controls.

For following a group over time

The cohort grid. Rows are start periods, columns are age, cells are the measure. It borrows its logic from a cohort study and answers "is this getting better for people who joined later." Fails when immature cohorts are compared to mature ones, so the most recent rows always look worst. Mark the incomplete cells.

Choosing between them

The reader's task	Reach for	Do not reach for
Spot a problem quickly	Exception list, run chart with band	Composition chart, scatter
Understand a change	Waterfall, small multiples	Single number, leaderboard
Compare entities	Ranked bars, scatter	Funnel, cohort grid
Judge whether things improve	Cohort grid	Single number with month over month
Find what is stuck	Queue age histogram	Any average

The one pattern worth internalizing: shapes built for noticing carry very little information on purpose, and shapes built for explaining carry a lot. Putting an explaining shape on a monitoring page is how a dashboard becomes something people stop opening.

Before you commit to a shape

Sketch it with fake extremes. What does this look like when one category is ninety percent of the total, when a segment has three members, when the newest period is half complete? Every failure mode above is visible in that sketch, and none of them is visible on the tidy sample data you are building with.

Related reading on this site

For deciding what kind of page these panels belong on, see dashboard design. For the layout and default errors that spoil good shapes, see nine dashboard mistakes. Several failure modes above are properties of the measure rather than the picture: see metric shapes and their traps. For panels that show a projected future rather than a recorded past, see forecasting, and for the fixed window versions of these, see reporting.

Common questions

Which shape should a page start with?

The single number with a comparison, in the top left, answering the question the reader came for. Everything else on the page exists to explain or qualify it.

Are pie charts missing on purpose?

Yes. Anything a pie chart shows is read more accurately from a ranked bar or a composition chart, and the one case it is genuinely good at, two or three parts of a whole, is served just as well by a sentence.

How many shapes should one page use?

Three or four. A page with eight different chart types asks the reader to learn eight reading conventions before getting any information.

What about a map?

Use one when geography is the question, not when geography is merely an attribute. Maps encode area rather than value, so a large sparse region dominates a small dense one that matters more.