Dashboards people can read at a glance

Most dashboards answer the question "what data do we have?". Almost nobody opens one to ask that. They open it to ask "is anything wrong, and do I need to do something today?" — and then close it again in under a minute.

The gap between those two questions explains most of what goes wrong with dashboard design. A screen built from the data available will be complete, accurate, and useless at a glance. A screen built from the question will leave things out, which is the part that takes nerve.

What a glance can actually hold

A glance is about three seconds and four or five pieces of information. That is the budget. Everything beyond it is a second visit, and a second visit only happens if the first one earned it.

So the top of a dashboard is not a summary of everything. It is the shortest honest answer to "is this normal?" — and the rest of the page exists to explain that answer when it is no. Inverting that, which is the default, produces a wall of equally weighted tiles where nothing is findable because nothing is subordinate.

If every number on the screen is important, the screen has not been designed yet.

A number alone is not information

"1,284 orders" means nothing. It becomes meaningful only against something: last week, the same week last year, a target, the rest of the fleet, what you would expect given the season.

This is the single highest-leverage decision in a dashboard, and it is usually left to the viewer. The viewer then does the comparison badly, from memory, or not at all. A figure shown with its own baseline — a delta, a sparkline, a band of normal — turns a lookup into a judgement, which is the thing the person came for.

The corollary is that the default time range is a design decision, not a default. Whatever it is set to is what most people will read, forever, because almost nobody changes it.

The states that decide whether it is trusted

A dashboard is read quickly and acted on, so the ways it can be subtly wrong matter more than in most interfaces:

• Stale. The numbers were right four hours ago. If the screen does not say when it was last updated, every figure on it is an assertion the viewer cannot check.
• Partial. One source failed and the rest loaded. A total that silently excludes a region is worse than no total, because it looks complete.
• Empty. A new account, or a filter with no matches. These look identical to zero, and zero is a finding — "no data" and "a value of nothing" must not render the same way.
• Too small to mean anything. A conversion rate computed from nine visits will swing wildly and invite a decision it cannot support. Showing the denominator is often enough to stop that.

Colour is a sentence, not decoration

In a dashboard, colour is one of the few things read pre-attentively — before the viewer has consciously started reading. That makes it expensive to spend on decoration.

The useful rule is that colour should encode one thing consistently across the whole product, and the obvious candidate is attention: this needs you, this does not. Once red also means "revenue" in one chart and "error" in another, it stops being readable at a glance and becomes something the viewer has to decode per widget.

The same applies to the chart-type choice. A line for change over time, a bar for comparison between things, a number for a single fact. A donut is almost never the right answer, and the reason is mechanical rather than aesthetic: angles are hard to compare and the labels end up outside the shape anyway.

Density is not the enemy

There is a reflex to make dashboards airy, and for a marketing page that is right. For a screen somebody uses every day it often is not: whitespace that pushes the second row below the fold costs a scroll on every visit, forever.

Professional tools are usually dense and that is not a failure of taste — a trader, a dispatcher or a dealer managing stock wants more on screen, not less. The thing that makes density readable is not space, it is hierarchy: consistent alignment, a small number of type sizes, and ruthless subordination of the things that are not the answer.

Lowkar, a car marketplace built with React and Next.js, has surfaces of this kind: dealer listing flows where somebody is managing real stock rather than reading a summary once a week, and where the useful screen is a working one rather than a spacious one.

Every number should lead somewhere

The glance tells you something is wrong. The next thing a person does is ask why, and a dashboard that cannot answer that sends them to export a spreadsheet — at which point the dashboard has become a notification and the real work happens elsewhere.

So the rule is that every headline figure should be clickable into the rows behind it, with the filter already applied. Not a separate reports section with its own filters to reconstruct: the same question, one level deeper, carried over.

What to check on your own

1. 01Open it and time yourself. If you cannot answer "is anything wrong?" in three seconds, the top of the page is doing the wrong job.
2. 02Cover the top row. If nothing is lost, the top row is decoration.
3. 03Find a number with no comparison attached and add one, or remove it.
4. 04Check what it shows for a brand-new account with no data at all.
5. 05Unplug one data source and see whether the page admits it or quietly reports a smaller total.
6. 06Click a headline figure. If nothing happens, the next step is a spreadsheet.

If you have a dashboard people glance at and then go elsewhere to do the work, tell us what you are building.