Home › Topics › Dashboard Design: The NOC Wall vs. the Boss's Browser › Visual Hierarchy

Visual Hierarchy: Making the Important Thing the Big Thing


Every dashboard has a hierarchy whether you designed one or not. The eye lands somewhere first, moves somewhere second, and gives up somewhere third. If you did not decide where those places are, the layout tool decided for you. Usually, it did so by putting the first panel you built in the top-left corner. On most dashboards I have inherited, the top-left panel was the one the builder was most excited about on day one. On those dashboards, the panel that mattered was wherever there was room when it was added.


Visual hierarchy is the craft of deciding, on purpose, what the reader sees first. It uses three tools — position, size, and colour — and none of them requires design training or an opinion about fonts. It requires you to know which panel matters most and to be willing to make it the biggest thing on the screen.


By the end of this guide you will be able to read a screen the way a stranger's eye does. You will be able to run a five-second test with a colleague. You will be able to pick the right visual for each question and choose colours that work for a reader who cannot see them. You will be able to rearrange a cluttered dashboard so the important thing is the big thing. There is a decision table for choosing visuals, a worked example with six panels, and a design-rules checklist you can copy.


In this guide

  • What hierarchy means on a screen
  • Position, size, and colour as a language
  • The five-second test
  • Choosing the visual
  • Colour rules that survive a colour-blind reader
  • Scales, time ranges, whitespace, and labels
  • Worked example: rearranging six panels
  • The design-rules checklist
  • Where to go next

What hierarchy means on a screen


Visual hierarchy is the order in which a reader's eye visits the parts of a screen: first, second, third, and then everything else as a blur. It is not a metaphor. Put a page in front of someone and their eye will go to one spot, then another. That sequence depends on where things are, how big they are, and which of them are a different colour from the rest. Your job is to make that sequence match the order of importance.


The whole library rests on one definition — a dashboard answers one question, for one audience, at a glance. That definition is set out in the guide on what dashboards are for. Hierarchy is what "at a glance" means in practice. A glance is one eye movement, perhaps two. If the answer to the question is not where the first or second eye movement lands, the glance has failed, however good the panels are.


A useful way to think about it is a newspaper front page. The headline is the biggest text, at the top. The main story has the largest picture. The weather is in a small box in the corner, where you can find it if you want it. Nobody has to explain this layout to a reader. Monitoring dashboards are usually laid out like a newspaper where every story is the headline.


Position, size, and colour as a language


Three properties of a panel tell the eye how important it is, and they act like grammar: readers apply the rules without knowing they exist.


Position: top-left is first

Readers of left-to-right languages start at the top-left of a screen, move across the top, and then work downwards. Attention fades toward the bottom-right. The top-left panel is read by everyone; the bottom-right panel is read by people who were looking for it. Put the panel that answers the dashboard's question at top-left, or across the whole top row. Put things only useful to someone investigating at the bottom.


Size: bigger is read first

The eye goes to the largest thing on a screen before it goes anywhere else. It treats things of the same size as being of the same importance. If every panel is the same size, you have told the reader that nothing matters more than anything else. In that case, they will believe you. A grid of twenty identical panels is not a hierarchy; it is a spreadsheet with pictures.


A rule that works: one panel at four units, two panels at two units, and everything else at one. The four-unit panel is the answer to the question. The two-unit panels are the two things most likely to explain a bad answer. The one-unit panels are context. If you cannot decide which panel gets the four units, you have not decided what the dashboard is for. In that case, no amount of layout will hide that.


Colour: the exception, not the rule

Colour draws the eye to whatever is different from its surroundings, so it works only when most of the screen is not coloured. One red panel on a screen of muted greys is impossible to miss. One red panel on a screen with orange, blue, purple, and two shades of green is a red panel in a crowd. Reserve strong colour for state — something is wrong, something needs attention — and leave everything else quiet.


This is why the default palette of most graphing tools is unhelpful for monitoring. It gives every line a bright, distinct colour so you can tell them apart. That is fine for a chart in a report. It is poor for a screen whose purpose is that one line should stand out when it misbehaves. A dashboard where everything is bright is a dashboard where nothing is.


The five-second test


The five-second test is the simplest way to find out what your hierarchy actually says, as opposed to what you meant it to say. You show the screen to someone for five seconds, take it away, and ask what they saw. The gap between their answer and your intention is the work you have left to do.


Run it like this, with a colleague who has not seen the screen before:


  • Write down, privately, the one thing you want them to notice first and the one question the screen answers.
  • Show them the screen for five seconds. Count aloud if you have to; people cheat by looking longer.
  • Cover the screen and ask three questions: "What did you see first?" "Is anything wrong?" "What is this screen for?"
  • Write their answers down without correcting them.
  • Compare with what you wrote. Any mismatch is a hierarchy fault, not a colleague fault.
  • Fix the largest mismatch, then repeat with a different colleague.

The test finds three faults reliably. The first is that they saw the biggest thing first and the biggest thing was not the important thing. The second is that they could not tell whether anything was wrong, which means state is not being carried by colour or words. The third is that they could not say what the screen was for, which usually means it is for several things. (It is always for several things. That is how it reached twenty panels.)


Choosing the visual


Each kind of panel answers one kind of question well and the others badly. So choose the panel from the question, not from what looks impressive. Here are the six you will use.


A single number is one value in large type: "Open incidents: 2". A gauge is a dial or bar showing one value against a fixed maximum. A sparkline is a tiny line graph with no axes, showing shape only. A line graph is a full graph of a value over time with axes and units. A bar chart compares several things at one moment. A table is rows and columns of values or words.


VisualQuestion it answersUse it whenAvoid it when
Single numberHow many, or how much, right now?The value means something on its own (a count, a percentage of a known total).The reader needs a threshold to judge it; then show a status word instead.
GaugeHow close to the limit?There is a real, fixed maximum the reader understands (disk capacity, a link's speed).The maximum is arbitrary or unknown; a gauge with a made-up scale is a decoration.
SparklineIs it rising, falling, or steady?Direction matters and the exact value does not; good beside a number.Anyone needs to read a value off it. It has no axis on purpose.
Line graphWhat happened, and when?The reader is investigating and will look closely; desk dashboards and the second page.The screen is read from a distance or in a glance.
Bar chartWhich of these is biggest?Comparing a handful of things at one moment (top talkers, busiest ports).Showing change over time; that is a line graph's job.
TableWhat is the status of each of these?A list of named things each with a word or two; the manager's page.More than about twelve rows, or any reading from a distance.

Two of those deserve a warning. A gauge implies that the far end of the dial is a real limit, and readers believe it. So a gauge of CPU going up to a hundred is honest. A gauge of latency going up to a number you picked because it looked round is not. And the pie chart is missing from the table because it answers "what share of a whole", which monitoring rarely asks. A pie chart of errors by port has never once helped anyone fix a port.


Colour rules that survive a colour-blind reader


Some of your readers cannot distinguish red from green, and you do not know which ones. That single fact settles most colour arguments, because it means colour can never be the only way a panel says something. A red tile must also say "DOWN". A green line must also be labelled. An amber cell must also carry the word "Watch". The colour is a highlight for the readers who can see it, and the word is the message for everyone.


The test is easy: look at the screen in greyscale. Most screenshot tools, image viewers, and print-preview dialogs can show a black-and-white version. If you can still tell which panels are in trouble, the screen passes. If the states have all become the same grey, they were being carried by hue alone. And hue is exactly what a colour-blind reader does not get.


Contrast matters as much as hue, and it has a formula. The contrast ratio between text and its background is calculated from their relative luminance. That is roughly how bright each colour is to a human eye. The web accessibility guidelines most designers work to ask for at least 4.5 to 1 for normal text. They ask for at least 3 to 1 for large text:


contrast ratio = (L1 + 0.05) / (L2 + 0.05)

L1 = relative luminance of the lighter colour (0 = black, 1 = white)
L2 = relative luminance of the darker colour
Pass: 4.5 or more for normal text, 3.0 or more for large text

You do not need to compute luminance by hand; any contrast checker will do it from two colour codes. What matters is that you check. The pale grey text that looks elegant on your monitor at your desk is invisible on a television across a bright room. It was never readable to anyone with weaker eyesight than yours.


Finally, keep the colour vocabulary tiny and consistent across every panel and every screen. Green, amber, red, grey, and one neutral for everything else. If red means "down" on one panel and "high traffic" on another, the reader must learn each panel separately, and they will not.


Scales, time ranges, whitespace, and labels


Four smaller rules do more than their size suggests, and each is broken on most dashboards you will inherit.


Consistent scales and time ranges

Two graphs side by side will be compared, so give them the same time range and, where the units match, the same vertical scale. A graph of one link over the last hour beside a graph of another over the last day invites a comparison that means nothing. A reader who spots that the "spike" on one is a flat line on the other has learned to distrust both. Fix the time range for the whole screen and print it once.


Whitespace is a tool

Empty space around a panel is what makes it visible. Panels crammed edge to edge merge into one texture, and the eye slides over texture. Leave a clear margin around the four-unit panel in particular; the space is part of the hierarchy. Resist the instinct that an empty corner is wasted. The empty corner is why the rest of the screen can be read.


Labels and units

Every panel has a title that says what it shows in words a reader outside the team would understand. Every axis and every number has a unit. "Throughput" is a title; "Internet link, head office, inbound, Mbit/s" is a label. A graph whose vertical axis has no unit is a graph the reader will guess at. There is a great deal of difference between kilobits and megabits.


The date of the data

Put the date and time the data was collected on every graph. Yes, every graph. Yes, the one on the screen that refreshes automatically, because one day it will stop refreshing and look exactly the same. The NOC wall guide covers the freshness marker for a whole screen. This rule is the per-panel version. It is the one that saves you when a screenshot of the panel turns up in an email three weeks later.


At Harbourview Clinic the on-duty admin glanced at the operations screen during a Tuesday outage and saw nothing wrong. The four-unit panel at top-left was a bandwidth graph of the guest Wi-Fi. It was behaving beautifully. The clinical records link was the one thing the building could not run without. It was a one-unit tile at bottom-right, next to the printer counts. It had been red for twenty minutes. The review found that the Wi-Fi graph had been put top-left because it was the first panel built. That was on the day the dashboard was new and the guest Wi-Fi was the exciting project. The screen was rearranged in an afternoon. The Wi-Fi graph is now at the bottom, still behaving beautifully, where nobody minds.


Worked example: rearranging six panels


Here is a small desk dashboard for a single site, before and after, drawn as a grid of unit cells. The question it answers is "is the site's connection to head office healthy, and if not, why?" Six panels, no additions, no deletions; only position, size, and colour change.


BEFORE (all panels equal, order of construction)

+----------------+----------------+----------------+
| Guest Wi-Fi    | Printer        | Switch CPU     |
| clients (line) | queue (number) | (gauge)        |
+----------------+----------------+----------------+
| WAN link       | WAN latency    | Site status    |
| Mbit/s (line)  | ms (line)      | (tile)         |
+----------------+----------------+----------------+

AFTER (one at four units, two at two, three at one)

+---------------------------------+----------------+
| SITE STATUS                     | WAN latency    |
| (tile, word + colour)           | (sparkline)    |
|                                 +----------------+
|                                 | WAN Mbit/s     |
|                                 | (sparkline)    |
+----------+----------+-----------+----------------+
| Switch   | Guest    | Printer   |  (empty on     |
| CPU      | Wi-Fi    | queue     |   purpose)     |
| (number) | (number) | (number)  |                |
+----------+----------+-----------+----------------+

What changed and why. The site status tile answers the question. So it became the four-unit panel at top-left. It has a status word inside it so colour is not doing the work alone. The two WAN measures are what most often explain a bad status. So they became two-unit sparklines beside it. Shape is enough at a glance, and a click opens the full graph. Switch CPU, guest Wi-Fi, and the printer queue are context, so they shrank to single numbers along the bottom. The gauge went, because a number does the same job in less space. And the bottom-right cell was left empty, because it is the least-read spot on the screen and something would have been put there.


Run the five-second test on the "after" layout and the colleague says "site status, it is fine, this is about the branch connection." Run it on the "before" and they say "some graphs." That is the whole difference, and it needed no new data.


Tip: when you rearrange a screen people already use, tell them before you do it and leave the old version reachable for a fortnight. Even a better layout is a different layout, and the person who found the printer queue by muscle memory will report it as missing. The dashboard maintenance burden covers why every change to a shared screen costs more than the change itself.


The design-rules checklist


This is the list I run against any screen before it goes to an audience. It is about layout only; what to put on a wall or in front of a manager has its own guide in this pillar.


  • The question the screen answers is written down, and one panel answers it.
  • That panel is at top-left or across the top, and it is the largest panel on the screen.
  • Panel sizes follow one at four units, two at two units, the rest at one.
  • Each panel type was chosen from the question it answers, using the decision table.
  • No gauge has a maximum the reader would not recognise as a real limit.
  • Strong colour is used only for state; everything else is muted.
  • Every state shown by colour is also shown by a word or a shape.
  • The screen still reads correctly in greyscale.
  • Text and background contrast meets 4.5 to 1 for normal text and 3 to 1 for large text.
  • The colour vocabulary is the same on every panel and every screen.
  • All graphs on the screen share one time range, printed once at the top.
  • Graphs with matching units share a vertical scale.
  • Every panel has a plain-language title, and every axis and number has a unit.
  • Every graph shows the date and time of its data.
  • There is a clear margin around the largest panel, and at least one cell is deliberately empty.
  • A colleague has passed the five-second test on the final layout.

Where to go next


Hierarchy is the part of dashboard design that costs nothing and is skipped most often. That is because the layout tool will happily let you build a screen without one. Decide what matters, make it the big thing, put it top-left, and keep the colour for exceptions. Then show it to someone for five seconds and believe what they tell you.


The wall and the manager's page apply these rules at their two extremes. At one extreme, the NOC wall is hierarchy at three metres. And the boss's browser is hierarchy for a reader who will not ask. Dashboard wallpaper shows what happens to a good hierarchy when panels keep arriving. The design review checklist turns this list into a thirty-minute habit. They all belong to Dashboard Design: The NOC Wall vs. the Boss's Browser.


Put this into practice with SysPeep Monitor

Good hierarchy needs panels you can size, place, and colour by state rather than by default. SysPeep Monitor builds live dashboards and network maps from the devices it discovers and polls over SNMP, WMI, and SSH. Thresholds and dependency rules decide what counts as a problem before it reaches the screen. The free trial installs on a Windows machine if you want to try the six-panel rearrangement above on your own data.

Download Free TrialLearn About SysPeep