Most dashboards are built outward from the data: here is everything we can measure, laid out in a grid. Most dashboards that work were built inward from a question: what does the person opening this need to know in the next five seconds?

Start with the question

Write the question as a sentence before you open a design tool. "Is the system healthy right now?" "Did we miss any calls today?" "What needs my approval?" A dashboard that answers one question well beats one that answers ten badly.

If you cannot write the question, you do not yet know what the first screen is for.

Choose a handful of headline numbers

A row of four to six numbers at the top does most of the work. Each should be:

  • A number a person would act on, not just one that is available.
  • Compared with something, such as the last period, a target or a limit. A bare number is trivia.
  • Labelled in the user's words, not the database's.

In the Callwise dashboard, for example, the first row answers the owner's question by showing calls answered, leads qualified, appointments booked and response time, each against the previous period. The Tulip Systems flight console opens on a similar row of readings for vehicle state, control source, confidence and margin, because an operator's first question is "does anything need me?"

Make exceptions louder than routine

A scannable list needs a visual difference between normal and notable. Use a small, consistent set of status colours, and keep them meaningful: if everything is amber, nothing is. Pair colour with a word or an icon so it works for people who cannot distinguish it.

Design the other states

A dashboard is rarely shown with perfect data. Design the empty state (what a new user sees on day one), the loading state, the error state and the state where a number is unexpectedly huge or has a very long label. These are the screens people actually meet, and they decide whether the product feels finished.

Filter, do not decorate

Tables with filters for the handful of ways people slice the data (status, date range, owner) are more useful than extra charts. Add a chart only if a trend is the answer to the question.

A short test

Show the first screen to someone who has not seen it for five seconds, then ask what they would do next. If the answer is "I'm not sure", the screen is answering a different question from the one you wrote down.