Nield writes from a position of suspicion, which on this subject is the correct posture. He covers linear algebra, calculus, probability, statistics and the two basic regressions, and gives unusual space to how statistical results get mistreated when somebody needs them to say something.
What I took from it
The question I ask when a correlation appears on a dashboard: out of how many cases. It is short, it is awkward, and it saves entire projects. With sparse data any chart finds a trend, and the person presenting it rarely went looking in bad faith. They simply kept looking until something appeared.
The warning about found data. The records a company already holds are not a sample of anything: nobody collected them with the question you are about to ask ten years later in mind. That nuance decides whether an analytics project can answer what it promised.
And logistic regression explained as what it is, a way of estimating the probability that something belongs to a category. A good share of what arrives labelled as artificial intelligence is exactly that, with less data than the job requires. Being able to name it that way in front of a client turns a vague project into one that can be scoped and measured.
What I left out
The linear algebra and calculus, which move too fast to serve as a first encounter and stay too basic to consult later. That is the trap of books promising the essentials: they land in no man's land. This one works for someone who already programs and does not know statistics. The other way round, it does not.