The Dashboard That Answers Your Question

Act as the analyst who spent years building beautiful dashboards nobody opened, until you understood that a dashboard does not exist to show data: it exists so that one specific person can make one specific decision without having to call you. Context: I need to build a dashboard. The tool I am going to use is [TOOL - EX: Power BI, Looker Studio, Excel, Google Sheets, Metabase]. The data I have is [WHAT YOU HAVE, WITH THE EXACT COLUMN NAMES - EX: a sales spreadsheet with date, product, channel, salesperson, amount and order status]. It is refreshed [HOW OFTEN - EX: once a day, overnight]. The person who will open this dashboard is [WHO LOOKS AT IT, AND WHERE - EX: the shop owner, on a phone, before going to bed]. The decision that person needs to make is [THE DECISION, STARTING WITH A VERB - EX: decide which channel gets next month's budget]. Today they decide it [HOW IT WORKS NOW, WITHOUT THE DASHBOARD - EX: by asking in the WhatsApp group and looking at the bank statement]. The period that matters is [PERIOD AND COMPARISON - EX: the current month against the same month last year]. My level with the tool is [YOUR LEVEL - EX: I can build a pivot table, I have never written a measure]. Task: design the whole dashboard - what it answers, what goes in each block, which chart each block gets and why - before I open the tool. Rules: 1. Start by writing, in one sentence, the question the dashboard answers. If that sentence does not end in a decision, what I asked for is a report and not a dashboard: say so plainly and rewrite the question with me before going on. 2. Choose ONE headline number, the one the person looks at first and which on its own already changes what they do. Everything else on the dashboard exists to explain that number. 3. Every block is born from a written question. A block that answers no question comes off the dashboard, however good the chart looks. 4. Justify each chart type by the question, never by taste: comparing categories is bars, following something over time is a line, part of a whole with few slices is a single stacked bar. A pie with more than three slices, a 3D chart, a doughnut whose legend is bigger than the drawing, and a speedometer gauge do not go in. 5. Do not invent a column. If a block needs data I did not list, write "[MISSING DATA: ...]" and move on - that is what I have to fix at the source before I build. 6. A number on its own says nothing: each one comes with a reference beside it, and it is one of three - the previous period, the target or the average. Say which of the three fits each block. 7. Write the calculation rule for each metric in plain language first, with the filter and the period built in [EX: sum of amount where status = paid, from the first day of the month until today], and only then the formula in my tool, with the function names in the language it is running in. 8. In a bar chart the vertical axis starts at zero. If any block asks for an exception, write why that exception does not mislead someone glancing at it. 9. Seven blocks at most, and all of them have to fit on one screen of the device I named. A dashboard that needs scrolling is a dashboard nobody scrolls. 10. Write the filters that sit at the top and, for each one, the value that is already selected when the person opens it. A dashboard that opens by asking you to choose is a dashboard people close. Format of the answer: first the dashboard question and the headline number, each on its own line. Then the map of the screen, block by block, in the position each one takes, from top to bottom and left to right; in each block, the question it answers, the chart chosen, the calculation rule in plain language, the formula in my tool and the comparison reference. Then the filters, with the starting value of each. Then the list of "[MISSING DATA]". And at the end, the question whoever opens the dashboard is most likely to ask without finding an answer, which is the block for version two.