The book walks through biostatistics in Python: cleaning and describing data, hypothesis tests and effect sizes, predictive models, clinical study design, survival analysis and synthesising evidence across studies. It is written for people in biomedical research, not for systems consulting.
What I took from it
Effect size. In medicine it is not enough for a difference to be statistically detectable: you have to say how large it is and whether it changes anyone's life. The same holds in business projects and almost nobody insists on it. Saying a process improved "significantly" means nothing if the improvement is two minutes a month. Since this stuck, any claim of improvement I write carries its magnitude alongside.
Designing the study before looking at the data. A clinical study states what will be measured, in whom, and how the result will be judged, and states it before starting. Applied to a pilot of anything, it prevents the commonest vice: run the pilot, look at everything, and keep the number that came out flattering.
And survival analysis, which turned out to be exactly what is needed when the question is how long something takes and not every case has finished: customers who have not left yet, invoices not yet paid. There is a correct way to handle those unfinished cases, and it is not to drop them.
What I left out
The entire clinical half, which is most of it: trials, cohorts, synthesis of medical literature. I do not work there and I will not pretend otherwise. As a general statistics book it has far better competition; its value to me was borrowed from another trade, and that is how I report it.