It is organised by concept, with key terms in the margin and examples in two languages. It covers exploratory analysis, sampling, significance testing, regression, classification and some machine learning. You use it as a reference rather than read it through.
What I took from it
Resampling. It is the practical way to put an interval around a figure without wrestling with theory: you resample what you have, many times, and watch how much the result moves. Out of that came a rule I apply in every report — no number travels alone. An average with no sense of how much it varies is half a truth, and usually the convenient half.
The chapter on multiple testing is what serves me most in front of dashboards. Look at twenty indicators hunting for the one that changed and something will look significant by pure chance. The practical consequence is not to stop looking: it is to decide in advance which ones get watched and what will be done if they move. A dashboard where the anomaly is hunted after the fact manufactures explanations for noise.
And a third, on regression: coefficients almost never mean what people believe they mean, and using them to talk about causes is the commonest mistake in a management report. The book is explicit about separating predicting from explaining, and that distinction is what prevents absurd recommendations.
What I left out
The machine learning chapters, which come up short next to books devoted to it and feel like they are there for the fashionable word in the title. And one formal annoyance: presenting everything in two programming languages at once doubles the length and breaks the reading. Anyone after applied statistics will read half the book, and rightly.