Why Dashboards Are Harder to Build Than They Look
A grid of charts looks simple until you have to keep it true, fast, and useful at the same time.
Every client starts the same way. "We just need a dashboard." A few charts, some numbers up top, maybe a filter or two. It sounds like a weekend project. Then you ask where the numbers come from, and the real project appears.
A dashboard is not a display. It is a promise that the number on screen matches reality at this exact moment, across every filter combination, for every user, forever. That promise is the entire difficulty. The charts are the easy 10 percent.
Most of the hard problems are invisible in the mockup. What happens when two data sources disagree about yesterday's revenue because one is on Pacific time and one is on UTC. What happens when a user filters to a date range with no data, versus a date range that hasn't finished loading. What happens when the aggregation logic that made sense at 10,000 rows falls over at 10 million. None of this shows up in Figma. All of it shows up in production.
Then there is staleness, which is its own discipline. Real-time dashboards feel alive right up until two numbers on the same screen were computed thirty seconds apart and now visibly contradict each other. Users notice this instantly, even if they can't articulate why the page suddenly feels untrustworthy. Solving it means deciding, explicitly, what "as of" means for every metric, and showing your work.
The last trap is scope creep disguised as a small ask. "Can we also break this down by region." "Can we add a comparison to last quarter." Each request is reasonable alone. Together they turn a stateless chart into a query planner, with joins, caching, and permission checks nobody budgeted for. Good dashboard work means designing the data layer for the tenth question, not just the first one somebody asked for.
That is really the whole craft: building something that looks like a static picture but behaves like a live system under constant interrogation. Get the underlying data model right, be honest about latency, and the charts take care of themselves.