KPI Dashboard Designer

Design a dashboard around the decisions it should drive: which metrics belong, which are vanity, and what each chart should look like when nothing is wrong.

Act as an analytics lead. Design a dashboard for a specific audience and decision, not a wall of charts.

Context:
- Who reads it and how often: [ROLE, CADENCE]
- The decisions they make with it: [LIST THE ACTUAL DECISIONS]
- The business: [WHAT WE DO, BUSINESS MODEL]
- Data I actually have: [SOURCES AND COLUMNS AVAILABLE]
- Tools: [LOOKER / METABASE / SHEETS / POWER BI / OTHER]

Deliver:
1. The three to five metrics that belong on this dashboard, each with: a precise definition (numerator, denominator, time window, filters), why this audience needs it, and the decision it changes
2. Metrics I might expect but should leave off, with the reason each is a vanity or lagging metric for this audience
3. For each chosen metric: the chart type, the comparison shown (versus last period, versus target, versus segment), and what a normal week looks like so a reader can tell signal from noise
4. The layout in reading order, top to bottom, with the one number that should be largest on the page
5. Thresholds and alerts: what value should make someone act, and who
6. A refresh and ownership plan, including what breaks silently and how we would notice

Be concrete: write the metric definitions so two analysts would compute the same number.

How to use this prompt

• Write the decisions in the input honestly — "so the team feels informed" is not a decision, and it produces a useless dashboard.
• Build only the top three charts first and see if anyone opens them before building the rest.
• Paste your existing dashboard and ask which charts fail the "changes a decision" test.

Why this prompt works

Dashboards sprawl because every metric seems worth adding. Anchoring the design to named decisions gives the model a reason to exclude things, and asking what a normal week looks like builds in the baseline that stops readers reacting to ordinary variance.

Variations to try

• Ask for an executive one-pager version with a maximum of four numbers.
• Add "we have three months of data and heavy seasonality" to get more careful comparison advice.
• Ask for the SQL for each metric definition against your table names.

Common mistakes to avoid

• Shipping twenty charts because the data was available for all of them.
• Defining metrics loosely, so finance and marketing quote different numbers in the same meeting.
• No owner and no alerting, so a broken pipeline shows a flat line for a month.

Works well with