Set 18 — dashboards, uncertainty, and the story
After lecture 21 · 11 December · about 40 minutes
No dataset. Bring your churn results from Sets 14 to 17.
1 · One figure, three audiences
Take your single best figure and produce three versions: for the data team, for the marketing director, for the board.
What changed between them? If the answer is only the font size, you have not done the exercise — say what you removed each time and why.
2 · Put the uncertainty in
Your model reports a churn probability per customer. Add an interval, or a calibration plot, or a decile table with observed rates — whichever is honest for your model.
Then answer: what does a manager do differently on seeing that a prediction of 0.7 is really “somewhere between 0.5 and 0.85”?
3 · The number that changes a decision
A dashboard with twelve tiles is a dashboard nobody reads.
Choose the one number that should be largest on the page, and defend it against two rivals you rejected. Tie your choice to an action somebody takes on a Monday.
4 · The sentence under the chart
Write the caption for your main figure: one sentence, stating what the reader should conclude, not what the chart contains.
“Churn by tenure” is a label. “Customers leave in the first six months, and almost nowhere else” is a caption. Write captions.
5 · Defend it
In pairs: present your figure in ninety seconds. Your partner asks the two hardest questions they can, one of which must be “how do you know?”
Write down the question you could not answer well. That is your revision list.
Check yourself on Blackboard
Open the test Set 18 on Blackboard: eight questions on the ideas behind exercises 1–4. Feedback appears right after you submit. Not graded; unlimited attempts.
This set checks: that a chart is an argument, that uncertainty shown is uncertainty managed, and that the caption is the part people remember.