Chapter 4 — Practice-on-your-own solutions

The family Copilot picked. A distribution / correlation chart. Bubble charts encode three continuous variables on one canvas — two on the axes, one on the size — and are pattern-hunting artefacts. They are the right tool when the reader is asking «is there a relationship here?» or «where do the outliers cluster?» They are the wrong tool when the reader already has a hypothesis (churn went up) and wants to know which drivers explain the change.

The family the question calls for. A ranked contribution family — typically a Pareto chart, a ranked bar of decomposed drivers, or a waterfall showing how each candidate driver added to (or reduced) the churn delta. The question «why did churn increase?» is a decomposition question with a known outcome and a set of candidate causes; the chart’s job is to rank the candidates by contribution to the observed change and make the top two or three visually loud.

The one-sentence direction. «Show which drivers contributed most to the Q3 churn increase as a ranked contribution chart — one bar per candidate driver, sorted by contribution to the churn-delta descending, with the top three annotated and the cumulative-share line overlaid.» That direction names the family (ranked contribution), the sort (by delta contribution, descending), the emphasis (top three annotated), and the reading order (Pareto with cumulative overlay so the reader can see when the drivers run out of explanatory power). Copilot will produce a defensible chart from that direction on the first pass; the bubble chart cannot be rescued by tweaking colours.

Why the bubble chart fails on this question specifically. Three continuous variables on one canvas force the reader to hold three axes in mind while eyeballing pattern shape. Even when the pattern is present, the reader gets a «there seems to be some relationship between X and Y and maybe size correlates with X» impression — not a rankable list of drivers with quantified contribution. The support team’s decision (which driver to invest remediation in) needs a rank, not a scatter.

The two candidate families.

1. Sorted horizontal bar of revenue lift by product line, five bars total (all product lines), the three winners accented in a positive colour, the two negative-margin lines annotated with a symbol or footnote indicator. Answers «which product lines did the promo lift?» in three seconds.
2. Paired dot-plot or side-by-side bar of margin change on the two hurt lines, showing pre-promo vs post-promo margin percentage or margin dollars, with the delta annotated. Answers «how much did we pay in margin on the two lines the promo hurt?»

The primary. The revenue-lift ranked bar. The audience’s first decision is whether the promo worked — and the answer is yes for three lines, no for two. The revenue-lift chart puts that finding in the eye’s first landing zone. The margin-impact chart cannot lead because it presupposes a reader who already knows the promo lifted revenue on three lines.

Why the other family is a secondary chart, not a shared one. The clustered column tries to merge revenue lift and margin impact into one visual (five product lines × two metrics = ten bars). The reader has to (a) match colours to legend, (b) mentally split the ten bars into two five-bar sequences, (c) reason about two axes with different units (revenue in dollars, margin in percentage points), and (d) form the argument themselves. That is the Lina failure from Chapter 3 in a different form. The two-chart alternative sequences the decisions instead: revenue-lift chart answers «did it work overall?» first; margin-impact chart answers «what did the two hurt lines cost us?» second. Leadership acts on the first before caring about the second, and the second is meaningful only after the first has established the trade-off. Two audiences would need two separate charts; one audience with two sequential decisions still needs two separate charts, presented in reading order.

The one-sentence Refine direction. «Split the current clustered column into two charts: (1) a sorted horizontal bar of Q3 revenue lift by product line, three winners accented, two negative-margin lines flagged with a footnote symbol; (2) a paired dot-plot of pre-promo vs post-promo margin percentage for the two flagged lines, with the delta annotated.»