Chapter 1 — Practice-on-your-own solutions

Any three of the four below. The exact wording matters less than naming the specific principle-level defect — not just «looks wrong».

1. KPI card «Churn rate: 12.4%» carries no comparison. Same failure mode as Priya’s $42.7M card: a number without a reference frame is not a story. Is 12.4% good, bad, on plan, better or worse than last quarter? The card asserts a number and asks the audience to supply the meaning. Violates Chapter 2 (context): no reference frame (target, prior period, peer, threshold, forecast) named. Fix: add prior-quarter or plan number below with variance.
2. Bar chart of churn by product line has no visual priority. All bars carry equal weight; nothing directs the leader’s eye to the product line where the decision lives. Same failure as Copilot leaving Priya’s regions in alphabetical order with uniform gray bars. Violates Chapter 5 (focus and decluttering): pre-attentive attributes not directed at the decision-relevant category. Fix: highlight the worst-performing product line in a stronger colour and render the others in muted gray; sort by churn descending.
3. Line chart of churn over six months has no threshold reference. The line goes up or down but the audience cannot see whether the current trajectory is acceptable. A reference line at the target churn rate (or last-year average) would let the eye read the argument in half a second. Violates Chapter 2 (context) again, at the chart level: same reference-frame gap as the KPI card. Fix: overlay a target line and annotate the crossing point if there is one.
4. Top-five churned accounts table lists the wrong metric. Ranking accounts by churn count is one question; ranking them by annual recurring revenue (ARR) at risk is a different question, and the leader making a save-or-let-go call cares about the second one. Two accounts on the top-five-by-count list may be small; two accounts off the top-five-by-count list may be the largest revenue hits. Violates Chapter 4 (chart selection), specifically the measure-choice component: the chosen measure does not match the decision the audience needs to make. Fix: add an ARR-at-risk column and rank by that.

Meta-observation. Three of the four failures are the same failure Priya’s dashboard had: missing comparison, missing emphasis, wrong measure. The pattern is not domain-specific; it is what Copilot returns when the analyst has not named the decision the dashboard is supposed to inform. Priya was staring at a sales dashboard; the churn team is staring at a churn dashboard; the principles that catch the defects are the same.

The one question. «What decision is the audience going to make from this dashboard, and who is the audience?»

Why it is the highest-leverage question to ask first. Every principle downstream flows from the decision and the audience. Without knowing them:

• You cannot judge whether the chart type is right (Ch 4) — a bar chart, a line, a table, or a scatter each serves a different question shape.
• You cannot judge whether the sort order or emphasis serves the decision (Ch 5) — the visual should land on the decision-relevant category first, and you cannot know which category that is without the decision.
• You cannot judge whether the comparison is the load-bearing one (Ch 2) — a chart with an on-plan reference tells one story; the same chart with a year-over-year reference tells another; the right choice depends on which comparison the decision needs.
• You cannot judge whether the colour emphasis is on the correct visual (Ch 6) — the accent should sit on the number the audience will act on.

Cold-checking the visual choices without knowing the decision is looking for defects in the wrong place. A chart can pass every visual-craft check and still be the wrong chart for the meeting it will appear in. Priya’s dashboard is the canonical example: no single panel violates a rule of chart craft in a loud way, and yet every panel fails because Copilot was never told what decision the leader was walking in to make.

Bonus signal. If your colleague cannot answer the question, that is itself the highest-leverage finding of the review. The dashboard is not ready to be reviewed for craft; it is ready to be sent back to a BRD (Chapter 2 § The BRD as the contract between the analyst and the audience). The Copilot-generated artifact is a starting point for the analyst’s thinking, not a substitute for it.