Build a dashboard.

Key terms

Dashboard

In Stephen Few’s definition, “a visual display of the most important information needed to achieve one or more objectives; consolidated and arranged on a single screen so the information can be monitored at a glance.”

KPI

Key performance indicator: one of the few metrics a team has agreed shows whether it’s meeting its goals.

Context

What a number is compared with, such as a target, last week, or last year. Without it, a number can’t tell you if things are good or bad.

Trend

How a number changed over time, usually shown as a line.

Data definition

A written note of what a metric means: its formula, period, data source, and owner.

Alert

A message sent automatically when a metric crosses a level you set, so someone looks right away.

The idea

Many dashboards fail quietly. Someone builds a screen with forty charts, it gets shown once in a meeting, and nobody opens it again. Not because the data was wrong. Because nobody could tell, at a glance, whether things were fine.

A good dashboard answers one question quickly: is this on track, and if not, where should I look? This lesson shows you how to build one, step by step, using Stephen Few’s well-known guidance on dashboard design and the metrics you chose in lesson 4. The example is the checkout and delivery team at the made-up furniture store.

The lesson, in six parts

1. 1Start with who it’s for, and what they’ll decide.

   Stephen Few, a data visualization expert, wrote a definition that’s still widely quoted: a dashboard is “a visual display of the most important information needed to achieve one or more objectives; consolidated and arranged on a single screen so the information can be monitored at a glance.” Notice where it starts. With objectives, not with data.

   Few also points out that a dashboard “quickly tells you what’s happening, but not why it’s happening.” Finding the why is the job you learned in lesson 5. So before you pick a single chart, answer three questions:

   • Who will look at it? A team checking daily, a leader checking monthly, or everyone watching a new launch?
   • What will they decide or do after looking? Fix something today, change a plan, or report up?
   • How often will they look, and how fresh must the data be?

   Who looks	How often	What it should show
   The checkout and delivery team	Every morning	Yesterday’s numbers against normal, errors and failures, anything that needs fixing today
   Leaders	Monthly	Progress toward the quarter’s goals, trends, the few things that changed
   A launch team	Daily, for a few weeks	The new feature’s adoption, its main metric, and its guardrails

   Common kinds of dashboard, described as general practice. The team names are from the made-up store.

   Write the answer at the top of your plan, in one line: “This dashboard helps [who] decide [what], every [how often].” If you can’t fill it in yet, that’s your next conversation: ask the people who’ll use it before you build.

2. 2Choose a few metrics, tied to goals.

   Now pick the numbers, and be strict. Few’s advice is to “include only the information that you absolutely need.” A useful pattern: one main metric (the primary metric from lesson 4) that shows whether the goal is being met, a few inputs that explain it, guardrails that catch harm, and a few health checks that tell you something is broken.

   Google’s HEART paper connects this to lesson 4. Its goals, signals, metrics process ends with building “specific metrics to track on a dashboard.” So if you’ve done the metric plan, most of your dashboard is already chosen.

   Role	What it answers	For the made-up checkout and delivery team
   Main metric	Are we meeting the goal?	First-try delivery rate: share of deliveries completed on the first attempt, per week
   Inputs	What’s moving it?	Share of orders with a confirmed address and phone number; share of orders with a delivery date chosen at checkout
   Guardrails	Is anything getting worse?	Returns because of damage or fit; delivery cost per order
   Health checks	Is something broken?	Payment error rate; page load time; support tickets about delivery
   Customer view	How do customers feel?	CSAT (customer satisfaction score) after delivery

   All metrics are examples for a made-up store. Use the formulas from lesson 4.

   A good test for each candidate: if this number changed tomorrow, would anyone do something different? If not, leave it off. It can live in a detailed report one click away.

3. 3Lay it out like a page.

   In languages read left to right, people tend to start a screen at the top left, the way they read a page. Few’s advice is direct: “The most important data ought to be prominent. Data that requires immediate attention ought to stand out. Data that should be compared ought to be arranged and visually designed to encourage comparisons.” He shows a dashboard that spends the top left, “the most prominent position”, on a logo and menus.

   He also lists “exceeding the boundaries of a single screen” as the first of thirteen common pitfalls. If people have to scroll or click through tabs to see the whole picture, it’s no longer at a glance. A layout that works for most product dashboards:

   • Top left: the main metric, large, with its target and last period beside it.
   • Top row: the inputs and guardrails, smaller.
   • Middle: the trend over time for the main metric, with the target drawn as a line.
   • Bottom: the breakdowns that help you see where a change happened, such as by platform or funnel step.
   • Put related numbers next to each other, so a number and its trend are read together.

   The drawing at the top of this page shows that layout for the made-up store. Notice what isn’t there: no logo, no decoration, no number without a comparison.

4. 4Give every number context.

   A number alone can’t tell you if things are good or bad. Few’s example is a sales total shown on its own: “Compared to what? Is this good or bad? How good or bad? Are we on track? Is this better than before?” His fix is to add one or more comparisons, “such as a target or some history”, plus a quick visual sign of whether the value is good, satisfactory, or bad.

   Two more of Few’s thirteen pitfalls belong here. Don’t show more precision than people need: $3.8M reads faster than $3,848,305.93. And show the difference directly when the difference is the point. If people need to know how far revenue is from budget, show “10% under” instead of two numbers they have to subtract in their heads.

5. 5Pick the chart by the question it answers.

   In a 2004 article on choosing graphs, Few groups almost every message a chart can carry into seven types, including change over time (time-series), ranking, parts of a whole, and how two things relate (correlation), and gives the best way to show each. This lesson uses the four that come up most on product dashboards:

   Question	Use	Few’s guidance
   How did it change over time?	A line chart	Lines to show the overall pattern. “Always place time on the horizontal axis.”
   Which is biggest or smallest?	Bars, sorted	Bars only. Sort descending to highlight the highest values, ascending to highlight the lowest.
   What share is each part?	Bars, not a pie	Bars only. In “Common Pitfalls”, he shows that a pie chart often makes parts hard to compare without reading the labels.
   Do two things move together?	A scatter plot	Points with a trend line.

   From Stephen Few, “Eenie, Meenie, Minie, Moe: Selecting the Right Graph for Your Message” (2004), and his “Common Pitfalls in Dashboard Design” (2006).

   Avoid the decoration that dashboard tools make easy: 3-D effects, dials that look like car gauges, a different color for every bar. Few’s rule for color is to keep it “subdued and neutral, except when you are using color to highlight something.” And start bar charts at zero. Few shows a bar chart whose scale doesn’t start at zero, where revenue looks about four times costs, when it was actually less than double.

6. 6Define every number, set alerts, and keep it alive.

   A dashboard is only trusted if everyone reads it the same way. Next to it, or one click away, keep a short data definition for every metric:

   Field	Example for one metric (made-up store)
   Name	First-try delivery rate
   Definition	Share of deliveries completed on the first attempt
   Formula	Deliveries completed on the first attempt ÷ deliveries with a first attempt in the period, × 100
   Period and refresh	Weekly, Monday to Sunday; updated every morning
   Source	The delivery partner’s status data
   Owner	The delivery product manager
   Target	Set by the team for this quarter

   Then decide which numbers deserve an alert. Set the trigger outside the normal ups and downs, so it doesn’t fire every day. To find them, look back over the last few weeks, note the highest and lowest values on ordinary days, and set the trigger a little past them. Send the alert to someone who can act, and say in the message what to check first. An alert nobody acts on trains people to ignore alerts.

   Finally, treat the dashboard like a product. Show it to the people it’s for, and watch how they use it. Every few months, ask three questions, and remove what nobody looks at:

   • Is it opened regularly, by the people it was built for?
   • Has it led to a decision or a fix in the last month?
   • Is there any number no one can explain? Define it or remove it.

With AI: AI is changing how people use dashboards. Some analytics tools now let you ask a question in plain words and get a chart, or summarize what changed this week. That can make a fixed dashboard less central for exploring, but no less useful for watching the few numbers that matter. And an AI summary is only as good as the definitions underneath it: if “active user” means two things, the summary can mix them up without warning you.

PURPOSE: This dashboard helps ________ decide ________, every ________.

MAIN METRIC (top left)
  Name and formula:
  Target:
  Compared with (last period, last year):

INPUTS (what moves it):
  -
  -

GUARDRAILS (must not get worse):
  -

HEALTH CHECKS (something is broken):
  -

LAYOUT
  Top left:
  Top row:
  Middle (trend over time):
  Bottom (breakdowns: platform, region, funnel step):

CHARTS: line for time, sorted bars for comparisons and shares (skip pies), no 3-D, bars start at zero.

DATA DEFINITIONS: name, definition, formula, period, source, owner, for every number.

ALERTS
  Metric and trigger:
  Sent to:
  What to check first:

REVIEW DATE (remove what nobody uses):

Ask AI about this lesson

All lessons