Contents

Data Engineering › Serving & Analytics

Dashboard Design

Building dashboards people actually use.

Also known as: designing dashboards, dashboard layout, dashboard best practice

A dashboard is a single screen that answers a recurring question. The design work is deciding which question that is, and then removing everything that does not serve it.

Why it matters

Most dashboards accumulate. Someone asks for one more chart, then another, and a year later the page has forty tiles, no obvious reading order, and a load time that makes people wait for the number they wanted. The classic failure is not a bad chart; it is a dashboard with no question at the top.

Rules that carry most of the value

  • One question per dashboard, stated in the title. Not “Sales” — “are we on track this quarter, by region”. A dashboard’s title is its specification, and if you cannot write the question, the dashboard has no purpose yet.
  • Answer at the top, detail below. The reader gets the answer in the first second and only scrolls if they want to know more.
  • A small number of KPIs, chosen with kpi design and defined once with metric definitions. Conflicting numbers on one page are worse than no numbers, because they cost you the reader’s trust in the others.
  • One chart per panel, each doing a job from chart selection. A panel with four charts in it is four half-attention charts.
  • Show the comparison the reader needs: against target, against the same period before (period over period), against the other regions. A number with no reference point is not an answer.
  • Say when the data was last updated and where it came from. A dashboard nobody trusts is worse than an empty page — see data freshness.
  • Make it fast. A dashboard that takes a minute to load gets opened once a week instead of daily, which quietly changes the decisions it feeds. That is query cost and aggregate tables as a design constraint, not just an engineering one.

The trade-off

Scope against usefulness. A dashboard for one question serves a handful of people and gets used. A dashboard for everyone gets opened once and then remembered as “the dashboard”. Keep the number of dashboards smaller than the number of decisions, and retire what nobody opens.

Drill-down from a tile to its detail is covered in drill-down — that is how a dashboard stops being a status page and starts being a place to work. Where the goal is letting people answer questions without you, that is self-service analytics, and the recurring-report form is a bi dashboard.