Amey-Thakur Dashboard building
Dashboards fail by including everything measurable. A wall of charts communicates nothing because nobody knows which number matters, and the ones people check daily are always small and focused.
Method
- Start from the decisions it supports. What action does a viewer take based on this, since a metric that changes nothing is decoration (see dashboard-design).
- Lead with the few numbers that matter. Three to five headline figures at the top, detail below for those who need it.
- Show change, not just level. A number without a comparison to last period or to target carries almost no information.
- Order by importance, not by data source. The layout should follow what people look at first rather than how the data was assembled.
- Choose chart types by the question. Trends over time as lines, comparisons as bars, and pie charts almost never (see data-visualization).
- Make the time range and filters obvious. Misreading a dashboard because of an unnoticed filter is common and consequential.
- Retire panels nobody looks at. Dashboards accumulate, and a periodic cull is what keeps the remainder meaningful.
Boundaries
Dashboards show state and do not explain causes, which needs investigation (see search-analytics). Real-time refresh is rarely necessary and adds load. A dashboard is not monitoring: alerting on conditions belongs elsewhere (see alerting-design).