← Guides

Building

How to build dashboards

This is what I do all day: systems and dashboards that give an organisation one version of the truth. Most dashboards fail for the same ten reasons.

  1. Start with the decision, not the data.

    If nobody changes what they do because of a number, that number does not belong on the dashboard. Ask “what will you do differently on Monday?” before you write a line of SQL.

  2. Interview the people who’ll open it.

    Not the person who commissioned it. The three people who’ll have it up on a second monitor at 8am have a completely different set of questions, and theirs are the ones that matter.

  3. One source of truth, first.

    Two dashboards that disagree destroy trust in both, permanently. Fix the pipeline and the definitions before you make anything pretty -- you cannot design your way out of contradictory numbers.

  4. Five numbers above the fold.

    A screen with thirty tiles is a screen nobody reads. Pick the five that answer “is this healthy?” and push everything else one click down.

  5. Every number needs a comparison.

    £40,000 is trivia. £40,000, up 12% on last month and 4% under target, is information. A bare figure makes the reader do work they will not do.

  6. Name the metric what the business calls it.

    If the floor says “time to hire”, the tile says time to hire, not “req_cycle_duration_avg”. Your schema is not the user interface.

  7. Show how fresh it is.

    A visible “updated 6:04am” prevents the single most damaging question a dashboard can face, which is “is this current?” asked in a meeting with no answer.

  8. Design for the phone.

    Executives look at dashboards in the back of taxis. If the top three numbers don’t survive a 375px screen, they don’t survive at all.

  9. Kill the charts nobody opens.

    Log the usage and be ruthless about it. Every unused chart is maintenance cost and visual noise charged against the charts that do get used.

  10. Write the definitions down.

    A one-page data dictionary next to the dashboard -- what each metric means, what it excludes, who owns it. It ends more arguments than any visualisation you will ever build.