Home › Topics › Dashboards: Build, Buy, or Both › What Dashboards Are For

What a Dashboard Is For (and What It Isn't)


I have built the forty-panel dashboard. Twice. It starts as eight sensible tiles. Then someone asks for the CPU of one server, and you add it, because the tool makes adding a panel a nine-second job. Then the temperature of the comms closet. Then a pie chart, for reasons nobody wrote down. Three months later nobody can tell from the screen whether the network is healthy, and the person who built it has quietly stopped looking too.


The tool is not at fault, and neither are you. Nobody told you what a dashboard is for. Here is the definition this guide hangs on: a dashboard is a screen that answers one question, for one audience, at a glance, without anyone clicking. That is the entire job. Colors, layout, refresh rate, choice of panels: all of it follows from that one sentence.


By the end of this guide you will be able to name the three audiences a dashboard serves and the question each is asking. You will be able to explain how a dashboard differs from an alert and from a report. You will be able to apply design rules that work even if you have never designed anything. And you will be able to diagnose the saddest object in IT, the dashboard nobody looks at.


In this guide

  • A dashboard answers one question
  • The three audiences and their questions
  • Dashboards, alerts, and reports are different tools
  • Design principles for people who are not designers
  • The dashboard nobody looks at
  • Purpose, audience, refresh, and lifetime at a glance
  • A worked example: splitting the "everything" screen
  • Where to go next

A dashboard answers one question


A dashboard is a page made of panels. A panel is one visual element: a single number, a graph of a value over time, a colored up/down indicator, a table, a gauge, or a map. A monitoring tool feeds those panels from the data it collects — ping results, SNMP counters, disk space, service states. It redraws them on a schedule called the refresh interval. Mechanically, that is all. The interesting part is what you put on it.


The word came from the car, and the car got it right. A car dashboard shows speed, fuel, engine temperature, and a few warning lights. It does not show the service history, last Tuesday's tire pressure, or a graph of your average speed by month. It answers exactly one question — "can I keep driving right now?" — at a glance, because you are supposed to be looking at the road. A good monitoring dashboard has the same discipline: one question, answered so plainly that a stranger can read it in five seconds.


The test for any screen you build or inherit: stand three meters back. Can a stranger tell whether things are okay? If they have to lean in, hover, or ask what the green line means, it is not doing its job. A dashboard that tries to answer every question answers none, because the one answer that matters is buried under thirty-nine that do not.


The three audiences and their questions


"One question for one audience" sounds restrictive until you notice that most small networks have only three audiences, each asking a question that barely changes. One dashboard per audience is most of what you will ever need.


The wall screen: "Is anything broken right now?"

This is the dashboard on the television in the IT office, or in a NOC (network operations center). In a small organization, that is a fancy name for the corner where somebody keeps an eye on things. Nobody clicks it. It is read from across the room, in passing, by whoever happens to be there. Its question is binary: is anything on fire? So its panels are mostly state indicators — big colored tiles for the critical services and sites, green when fine, red when not. A few headline numbers are fine if readable at a distance. Graphs mostly are not, because a squiggle you cannot read from the doorway is decoration.


A good wall screen is boring nearly all of the time. That is the point. When a tile turns red, everyone in the room notices within seconds, without an email or a page. Its refresh interval is seconds to a minute, and its time range is "now."


The admin's morning view: "What changed, and what is trending wrong?"

This is the dashboard you look at up close, with coffee, while clicking. It covers the last day or week. And it is made of graphs rather than tiles, because its job is to show movement. That might be a disk filling up or error counters climbing on a port. It might be an interface that flapped at three in the morning and recovered before anyone woke up. Or it might be a WAN link whose latency is a little higher every day. Nothing on it is necessarily broken. It exists to catch what will be broken next week if nobody acts.


The morning view is interactive. You hover over a spike to see the exact time, zoom in on the odd hour, click through to the device. It refreshes every few minutes, which is plenty, since the things it shows move slowly.


The manager's view: "Are we meeting our commitments?"

Your manager, your director, or the business owner does not care that GigabitEthernet1/0/24 logged input errors. They care whether the internet was up during business hours and whether the phone system met its availability target. They care whether the file server has room for the rest of the year, and whether the money spent on IT is buying stability. The manager's view answers in the manager's language: availability of named services over the month and response time against an agreed target. It shows the count of incidents and capacity headroom. The classic form of this screen is the network health scorecard, which condenses the whole network into a short list of graded lines.


This view refreshes daily or weekly and covers a month or a quarter. It rarely needs a graph; a table of services with a status and a trend arrow often does better. It is also the dashboard most likely to be read by someone who cannot ask what a panel means. So every label has to stand on its own. (The reader will not ask. The reader will conclude.)


Dashboards, alerts, and reports are different tools


Monitoring tools produce three kinds of output, and a lot of dashboard misery comes from using one where another belongs. The short version: alerts interrupt, dashboards inform, reports summarize.


An alert is a message that finds you — an email, a text, a page — because something needs a human now. It is allowed to wake you up. What separates a useful alert from noise is covered in the anatomy of a good alert. A dashboard is the opposite: you go to it. It never wakes anyone. It sits there being true, read when someone chooses to read it. A report is a document — a PDF, a scheduled email, a printed page — covering a period that has already ended. It is read once, later, usually by someone making a decision.


The confusion comes in three shapes. The first is the dashboard used as an alert: someone is assigned to watch a screen and react when a tile turns red. That is a person doing an alert rule's job, badly, because people blink, take lunch, and go home. If you catch yourself watching a dashboard to notice failures, you have found an alert you have not written yet. The second is the report used as a dashboard: an outage discovered by reading last month's availability PDF, by which time it is history. The third is the alert used as a dashboard: fifty emails a day nobody acts on. But those emails give the team "a feel" for the network. That feel costs you the one email that mattered.


At Tollgate Foods the night supervisor was asked to "keep an eye on the screen" instead of anyone writing alert rules. The WAN tile went red at one in the morning and stayed red for nearly three hours. Meanwhile the supervisor was on the loading dock, where the work was. The outage was logged the next morning from a driver's voicemail. The post-incident note observed that the dashboard had "functioned correctly", which it had.


OutputWho initiatesTime horizonQuestion it answersWhat goes wrong when misused
AlertThe system, to youThis momentDoes a human need to act now?Used as a feed: noise trains people to ignore it
DashboardYou, to the systemNow to the last weekWhat is the state, what is moving?Used as an alert: someone has to stare at it
ReportA scheduleA finished periodWhat happened, and did we meet the target?Used as a dashboard: problems found weeks late

Tip: before building any dashboard, write its question on a sticky note and stick it on the monitor. If a panel does not help answer the note, it goes on a different dashboard or nowhere. This sounds childish and works better than any design guideline you will ever read.


Design principles for people who are not designers


You do not need a design background. You need five habits, most of them about restraint.


One question per screen

The moment a screen serves two questions, it serves two audiences, and every design decision becomes a compromise. The wall-screen reader wants big tiles, the morning-view reader wants dense graphs, and both end up squinting. When a second question creeps in, make a second dashboard. Dashboards are cheap. Confusion is not.


Top-left is prime real estate

In languages read left to right, eyes land on the top-left of a screen first and drift right and down. Whatever sits in that corner gets read; whatever sits bottom-right gets read last or never. So put the single most important answer top-left: the internet link, the core switch, the availability number the manager cares about. Not the logo, not the clock. Most tools fill the grid in the order you added panels. That is the order you thought of them, which is almost never the order of importance. Rearrange on purpose.


Color means state, and only state

On a monitoring dashboard, color is a language with four words. Green means fine, amber means degraded, red means broken, and grey means unknown or no data. Never use those colors for anything else. If the second line on a traffic graph is red because that was the default palette, a glance at the wall screen says "something is broken" when nothing is. Use neutral colors for graph series and reserve the state colors for state. There are two smaller points. Not everyone sees red and green the same way, so pair color with a label or shape. And make grey mean something. A grey tile is not "fine," it is "we cannot see this," which is its own kind of problem.


Fewer panels than you want

Every panel you add makes every other panel smaller and less noticed. There is no free panel. A wall screen works best with a single-digit count of large tiles. A morning view can carry a dozen graphs if they share a theme. Beyond that the screen becomes a catalog. The test for each panel: what decision does it support? If you cannot name one, move it to a secondary page for the day you need it.


Consistent time ranges and labeled units

All the graphs on one screen should share the same time range and refresh interval. Two graphs side by side, one showing six hours and the other twenty-four, look like they disagree even when showing the same event. Set the range once at the dashboard level and let every panel inherit it. And label everything: "Latency 40" means nothing until the reader knows it is milliseconds and whether 40 is good. Show units on every axis. Draw the warning threshold as a line so a reader can see at once whether the current value is near it.


The dashboard nobody looks at


Every organization has one: a dashboard built with care, looked at daily for two weeks, then never opened again. I found one of mine by searching for the name of a switch we had replaced.


  • Step 1: ask who it is for. Name a person or a role. If the honest answer is "everyone," it is for no one, and nobody feels responsible for reading it.
  • Step 2: ask what question it answers. If the answer takes more than one sentence, the dashboard is several dashboards stuck together, and each of the audiences finds it mostly irrelevant.
  • Step 3: check the view count. Most tools record when a dashboard was last opened and by whom. A screen with no views in a month is not a dashboard; it is a screensaver.
  • Step 4: check whether it is still true. Panels whose data source was renamed or whose device was replaced sit there looking calm because they have nothing to show. Quietly empty panels earn distrust fast, and distrust is why people stop looking. This rot gets its own guide in the maintenance burden nobody budgets for.
  • Step 5: check whether the answer arrives some other way. If the question is "is anything down?" and the answer already reaches the team as an alert, the dashboard is redundant.
  • Step 6: ask the intended audience whether they would notice if it vanished. If not, retire it, with the etiquette described in the guide to dashboard sprawl.

The cause is almost always one of four. It was built for a project that ended. It was built to impress someone during a demo. It was built because the empty grid was inviting. Or it was built before anyone had a question for it, on the theory that one would show up later. None of these are crimes. All of them are fixed by going back to the sticky note.


Purpose, audience, refresh, and lifetime at a glance


The three core audiences cover most of what a small network needs, but you will meet two other kinds of dashboard. A troubleshooting view is built in a hurry during an incident to watch one device or service closely. A project view tracks a migration or rollout while it happens. Both are legitimate. Both are temporary, and the trouble starts when they are treated as permanent. The incident ends; the troubleshooting view does not.


DashboardAudienceQuestionRefreshTime rangeTypical lifetime
Wall screenWhoever is in the room; on-callIs anything broken right now?Seconds to a minuteNowYears; changes rarely
Morning viewAdminsWhat changed? What is trending wrong?A few minutesLast day to last weekMonths; evolves with the network
Manager's viewManagers, stakeholdersAre we meeting our commitments?DailyMonth or quarterAs long as the commitment exists
Troubleshooting viewThe admin working an incidentWhat is this device or service doing right now?SecondsLast hourHours; disposable
Project viewThe project teamIs the rollout on track?VariesThe project so farDelete at project end

Refresh, time range, and lifetime all fall out of the audience and the question. If a screen's stated audience is "managers" and it refreshes every ten seconds, something is mislabeled.


A worked example: splitting the "everything" screen


Suppose you inherit the dashboard from the start of this guide. It is called "Network Overview," it has thirty-eight panels, it is on the office TV, and nobody can say what it is for.


First, list the panels and tag each with the question it answers. That produces eleven up/down tiles for servers, switches, and the firewall (is anything broken?). It produces nine traffic graphs for WAN and uplink ports (what is moving?). It produces six disk-usage gauges (what is trending wrong?). It produces four monthly availability numbers for phones, email, the internet link, and the file server (are we meeting commitments?). There is a topology map, a clock, and a weather widget someone added as a joke. And there are five panels that show "No data" and have for as long as anyone remembers.


Second, group by question. Three questions surfaced, plus decoration and debris. That is three dashboards. The clock, the weather, and the dead panels go nowhere.


Third, build the wall screen. The eleven state tiles become "Wall." The internet link and core switch go top-left, because if either is red nothing else matters. Tiles are enlarged until readable from the door. The topology map stays, since a map with red nodes reads at a distance. But it moves to the right where it supports the tiles rather than competing. Refresh: thirty seconds. Time range: now. Eleven tiles is a lot, so the three least critical merge into one "branch offices" tile that turns red if any branch does.


Fourth, build the morning view. The traffic graphs and disk gauges become "Morning," with the range set to seven days at the dashboard level so every graph agrees. The gauges become graphs, because a gauge shows where a disk is and a graph shows where it is going, and "going" is the question. Warning thresholds are drawn as lines, and the series lose the default palette that used red for the second line.


Fifth, build the manager's view. The four availability numbers become "Service Commitments," a single table with one row per service. Each row shows the target, the achieved value for the month so far, and a trend arrow. The table refreshes daily, carries a plain-language label on every row, and has a one-line note under it. The note says how availability is measured, because the reader cannot ask.


Sixth, retire the original. The old screen is exported to a file, and kept with the monitoring configuration. It is deleted after the team has lived with the new ones for a few weeks. Anyone who bookmarked it gets a note pointing to the replacements. The office TV now shows eight big tiles and a map, and a visitor can finally tell whether the network is fine.


The whole exercise takes an afternoon. No new data was collected, no new checks were written, nothing was bought. The same panels were sorted by the question they answer, and three dashboards people look at replaced one that nobody did.


Where to go next


A dashboard answers one question for one audience at a glance. Everything else follows. The room, the admin, and the manager each get a screen. Refresh, time range, and lifetime fall out of the question. Alerts interrupt, dashboards inform, reports summarize. The pie chart, wherever it ended up, is still not answering anything. The next question is where to build: in the dashboards your monitoring tool ships with, or in a separate flexible tool. That tradeoff is the subject of built-in dashboards versus flexible dashboard tools. The rest of Dashboards: Build, Buy, or Both covers keeping the screens true once they exist.


Put this into practice with SysPeep Monitor

If you would like a starting point that already answers the basic questions, SysPeep Monitor ships with live dashboards, network maps, and reports in a single web console. These are fed by the devices it discovers on your network. The free trial installs on a Windows machine in a few minutes. That is a low-effort way to see what a default overview screen looks like before deciding what to build yourself.

Download Free TrialLearn About SysPeep