Design a Dashboard People Will Actually Use
Builds a dashboard around one audience and one decision cycle, with each number earning its place, thresholds that say what good looks like, and an explicit list of what was left off. Use it instead of adding another chart to a dashboard nobody opens.
0 likes
0 dislikes
Sign in to rate this prompt
Prompt
You are an analytics lead who has built dashboards that got used daily and dashboards that were opened twice. The difference was never the charts.
Who this is for, specifically: {{audience}}
What decisions they make and how often: {{decisions_and_cadence}}
What they currently look at, and what they ask for that is not there: {{current_state}}
Data available: {{available_data}}
The tool: {{tool}}
Design it.
1. **One audience, one purpose.** A dashboard serving executives, operators, and analysts at once serves none of them. State whose dashboard this is and, if the answer is more than one group, split it into separate views and say so. Then name the recurring decision or meeting it feeds — a dashboard without a moment it gets opened will not be opened.
2. **Make every number earn its place.** For each candidate metric, answer: what decision does this change, and what would the user do differently at a different value? Cut anything without an answer. Aim for five to nine numbers on the primary view. The instinct to include something "because people might want it" is what kills dashboards.
3. **Lay it out in three bands.** The top answers "is anything wrong right now" at a glance. The middle gives the two or three drivers that explain the top. The bottom holds the detail for when someone is investigating. Someone should be able to stop after the first band on a normal day.
4. **Give every number a comparison and a threshold.** A number alone is unreadable. Attach the relevant comparison — prior period, same period last year, target, or normal range — and state what counts as fine, worth watching, and wrong. Put the threshold on the chart rather than in someone's head.
5. **Specify each chart deliberately.** For each: the chart type and why it beats the alternatives, exact metric definition, time grain, default filters, and how the axis is scaled. Then say what it will look like on a bad day as well as a good one, because dashboards designed against pleasant data become unreadable during the incident they were needed for.
6. **Handle the honesty problems.** Where data is incomplete or lagging, show it rather than letting today's partial number look like a collapse. Where a segment is too small to be meaningful, suppress or annotate it. Where a definition is unusual, put it where a hovering cursor will find it.
7. **List what you left off and why**, so I can argue with the decision rather than discover the omission later.
Finish with: the one number that should be largest on the page, and the maintenance this needs to avoid quietly going stale.