Dashboard and Metric Design prompts for Data Analyst
Ready-to-use AI prompts for dashboard and metric design — written for Data Analyst and easy to paste into ChatGPT, Claude or Gemini.
Design a dashboard around a decision
You are an analytics lead designing a dashboard for [audience]. Decisions they make: [decisions]. Data available: [data available]. Tool: [tool].
Start from the decisions, not the data. For each decision, specify the one metric that informs it, its precise definition, the comparison that makes it interpretable (target, prior period, segment), the chart type and why, and the threshold at which the viewer should act.
Then lay out the dashboard: what goes above the fold, what belongs in a drill-down, and what should not be on it at all. Name the metrics people will ask for that you are deliberately excluding and why. Include the refresh cadence and who owns each definition.
Define a metric precisely
Act as an analytics engineer resolving a metric definition dispute. Metric: [metric name]. How different teams currently define it: [current definitions]. Underlying data: [underlying data]. What the metric is used to decide: [use].
Write a single canonical definition covering: the numerator and denominator in words, the entity and time grain, inclusions and exclusions (test accounts, internal users, refunds, deleted records), how late-arriving data is handled, and the timezone.
Then show the SQL that implements it, list the edge cases where reasonable people would disagree with a recommendation for each, and quantify roughly how much the number changes between the competing definitions so people understand what is at stake.
Critique an existing dashboard
You are a data visualization reviewer. Here is a description of my dashboard:
<dashboard>
[dashboard]
</dashboard>
Audience: [audience]. What they should do with it: [intended use]. Complaint: [complaint].
Assess: does each chart support a decision or is it there because the data existed? Is the chart type right for the comparison being made? Can a viewer tell in five seconds whether things are good or bad? Are there missing baselines, misleading axes, or metrics that move together and are redundant?
Give a revised layout, the charts to cut, and the one addition most likely to fix the stated complaint. Explain each change in one line.