Dashboard Layout Principles
Designing for the room the dashboard will be read in — reading order, colour discipline, and what to leave out.
A dashboard's job is to let someone decide something faster than they could otherwise. Most dashboards fail at that not because the charts are wrong but because they were designed without deciding who was reading them, from how far away, and what they were deciding.
Start with the decision#
Before placing a single panel, answer three questions:
- Who reads this? An operator, a supervisor, a plant manager, an executive.
- From how far away? Arm's length, across a desk, across a control room.
- What decision does it support? "Should I intervene now?" is a different dashboard from "Was last month acceptable?"
Everything else follows. A dashboard that cannot answer these is usually a collection of charts someone had rather than a designed thing.
| Audience | Interval | Elements | Distance |
|---|---|---|---|
| Control room | Hourly, current shift | 3–5, large | Across the room |
| Supervisor | Shiftly or daily | 6–10, tables useful | A desk |
| Plant management | Daily, month to date | 6–8, trends and comparisons | A desk |
| Executive | Monthly, year to date | 3–5, highly aggregated | A meeting or a printout |
Summary to detail#
The strongest organising principle. Most important information top-left, supporting detail below and to the right.
Readers of left-to-right languages scan a screen in a rough Z: top-left first, across, then down. Put the number that matters most where the eye lands first, and do not make someone hunt for the headline.
Within a dashboard:
+---------------------------------------------+
| Heading — what this is, what period |
+----------+----------+----------+------------+
| Widget | Widget | Widget | Widget | < the state of things now
+----------+----------+----------+------------+
| Main trend chart | < how it got there
+---------------------+-----------------------+
| Pareto / breakdown | Detail table | < why
+---------------------+-----------------------+
Status, then trend, then explanation. It matches the order questions are actually asked in.
Colour is information, not decoration#
The most important discipline on a dashboard.
Reserve colour for meaning. If red means "out of specification", then nothing else on the dashboard may be red. A red chart border, a red heading, or a red series in a palette destroys the signal — the reader can no longer tell an alarm from a design choice.
Drive colour from limits and enumerations, never by hand. A hard-coded colour is a statement that will eventually be false: it stays green while the value goes out of specification. A limit-driven colour stays true with no maintenance. See Measure Limits and Thresholds.
Use one convention across every dashboard. Red out of specification, amber marginal, green good, grey no data. A reader should not have to relearn the scheme per page.
Grey is a colour with a job. "No data" needs to be visually distinct from "zero" and from "good". A missing value shown as green is actively misleading.
Do not rely on colour alone. Some readers cannot distinguish red from green. Pair colour with a value, a symbol, or position so the information survives without it. A widget showing a red Fail works; one showing a red blank square does not.
What to leave out#
The hardest part of dashboard design, and the highest-value.
Anything nobody acts on. If no decision changes based on a panel, it is consuming attention that the panels that matter need. Delete it.
Duplicate framings. A widget and a table cell and a chart legend all showing today's production is three uses of space for one fact.
Precision nobody uses. Six decimal places on a temperature, or a total to the nearest unit when decisions are made to the nearest hundred.
Gauges beyond the second one. A screen of twelve gauges looks impressive and reads poorly. Two or three for genuine emphasis; widgets or bullets for the rest. See Widgets and Gauges.
Charts answering a question nobody asked. Every panel should correspond to something a real person wants to know.
The removal test. Take a working dashboard and ask, of each panel, "if this were gone, what decision would get worse?" Anything with no answer should go. Dashboards accumulate panels and almost never lose them, so this needs doing deliberately.
Label the period#
A shockingly common failure: a dashboard whose panels cover different periods without saying so.
A widget showing today so far, beside a chart showing the last 30 days, beside a table showing yesterday, is three periods on one screen. Each is fine; together, unlabelled, they mislead.
State the period — in a Text panel heading, in the chart titles, or both. And where a value is a partial period, say so: Production (today so far) prevents the comparison against yesterday's complete figure that otherwise happens every morning.
Designing for a wall#
Signage is a different medium and needs a different design, not a scaled-up desktop one.
- Three or four elements, not twelve.
- Widgets and large trends. Tables are unreadable at distance.
- Colour carrying the message, because at four metres colour is read before numbers.
- No interaction. Nobody hovers for a tooltip or clicks a link. Everything must be visible.
- Eight to fifteen seconds dwell time in a loop.
And make sure the limits are tight enough that the display actually changes state. A wall screen that has been green for a month has trained everyone to ignore it, and it will be ignored on the day it turns red.
Designing for mobile#
Infohub dashboards are read on phones, often by people who are not at the plant.
Assume one column. A 12-wide panel and a 3-wide panel both become full width, so a layout that depends on side-by-side comparison falls apart. Where mobile access matters, build a dedicated dashboard: the three or four numbers someone wants when they are asked "how did we do", stacked vertically, and nothing else.
Consistency across the set#
Once there are twenty dashboards, consistency matters more than the cleverness of any one of them.
Same layout convention, same colour meanings, same naming, same period conventions. A reader who learns one area dashboard should be able to read every other one without thinking. Where the shape repeats per area, chart templates keep them aligned as they change.
A test worth running#
Show a finished dashboard to someone who knows the plant but has not seen it, and ask:
- What is this dashboard for?
- Is anything wrong right now?
- What period does it cover?
- What would you do about the worst thing on it?
If they cannot answer all four in under a minute, the dashboard needs work — and the specific question they stumble on tells you which part.