Early in my career I treated dashboards as the finish line. If the chart was accurate and the axis labelled, my job was done. It took a while to notice that nobody was actually using them.
The shift came when I started asking a smaller question before building anything: what decision is this supposed to change? A metric that doesn’t move a decision is decoration, no matter how rigorous it is.
Three habits have stuck with me. First, pair every number with the action it should trigger. Second, show uncertainty instead of hiding it — a fuzzy range people trust beats a precise number they quietly ignore. Third, design the empty and error states first; that’s where data products actually live most of the time.
None of this is about prettier charts. It’s about respecting that a person on the other side has to act on what you show them. Get that right and the visual design almost falls out for free.