LinkedIn rarely talks about books. Yet in my three-year journey from studying political philosophy at Oxford to working in data science and product management at Raise, books have been my primary source of knowledge, inspiration and perspective.

I’ve done plenty of Coursera and DataCamp courses. They’re excellent for practical skills and getting specific tasks done, but they rarely go beyond that. As AI takes over more of the coding and analysis grind, I’ve realised my “unique human contribution” lies less in writing lines of code and more in communicating clearly and making good decisions.

My philosophical training has helped here: it made me an open-minded listener, a critical reader and a clearer writer. On the technical side — whether choosing solutions to business problems, identifying the right performance metrics or selecting models — what matters most is understanding the “why” behind the “how”. And books are where I find that deeper “why”.

So, books. I’ll share titles that have proved both enlightening and enduring in my day-to-day work, books that helped me feel, and become, less of an impostor as a non-STEM graduate doing engineering work.

Naked Statistics by Charles Wheelan

I finished this on train journeys during my first week on the job. I’d studied statistics for political science at university and knew how it underpins most academic work, but I hadn’t yet found its relevance in business and everyday decisions. This book bridges that gap.

It isn’t technical. There’s very little maths. But it does something many technical books don’t: it puts statistics in its proper place. It shows how stats can both empower and mislead, and how much discretion analysts hold to (intentionally or not) create helpful or harmful effects with data. It’s written as much for producers of data as for consumers.

Why it matters in my work

As a data scientist, I feed dashboards and predictions to the company. Being acutely aware of my potential to mislead keeps me alert to metric choice, methodology, accuracy, and implications. As a product manager, I use both quantitative and qualitative feedback to prioritise features. Knowing the pitfalls stops me jumping to conclusions and forces me to ask: to what extent, and in what way, does the number actually support a value proposition?

Here are the takeaways that sit at the top of my mind:

1) Distribution matters.

Business asks for what’s “typical”, not “mean” vs “median”. It’s on the analyst to choose the right centre. Averages are useful, especially if they plug into formulae (e.g. forecasting spend as user counts grow). But outliers quickly pull the mean away from the typical. With Pareto-style heavy tails, medians, box plots and histograms often communicate reality better than a single average.

For example, we have a prospect who has cycled in and out of evaluation with us for over three years; if we optimised for average time-to-close, that single deal would skew the mean so much that sales would be perversely incentivised not to close it. A median or percentile-based target avoids that trap.

2) “So what?”

Statistics in themselves are not valuable. They are valuable only when they contribute to answering meaningful questions. Bad statistics are often bad not because they’re inaccurate but because they don’t address the goal, or worse, they substitute and distort the goal, introducing incentives against rather than towards the goal.

Early on, I built dashboards full of charts simply because I could. Some stuck; most were ignored. I once built a classic “time on platform” metric because every product analytics guide mentioned it. When I went to email the results, I got stuck on the first sentence explaining why it mattered. So what? Our B2B product’s job is to help users finish tasks quickly. Making “engagement time” a core KPI would have actively worked against user value.

3) Don’t follow hype (or fear).

Like Thinking, Fast and Slow, Naked Statistics remind us that our gut is easily swung by whatever just happened. I cannot recall how many times a collegue would tell me they urgently need a particular product feature after a tough customer call, but when we zoom out and check how widespread the issue is and how often it occurs versus the rest of the roadmap, we’d see an edge case next to larger, recurring pain points. Once we review that together, we stick with the strategy.

However, Naked Statistics also reminds us that apparent one-offs aren’t often independent. If several users report crashes on the same day — on a page we’ve just changed — I’d be naive to chalk them up to random browser errors. Signal versus noise is tricky; the right move is to zoom out and dive deep rather than rely on gut feelings or wishful thinking.

The book also tells us that regression to the mean is real. Exceptional spikes or dips tend to drift back to the long-term average unless something fundamental changed. As the dashboard builder, I’m often first to see movement in KPIs; before broadcasting wins (or losses), I look for drivers and corroborating metrics, and sometimes just wait for more data. More than once, “good news” turned out to be a tracking glitch or seasonality rather than genuine progress.

4) Good analysis needs a good team.

Finally, “garbage in, garbage out” has kept me grounded. High-quality, representative data depend on processes and systems across the organisation. Even the best chef is only as good as their ingredients and the logistics that keep them fresh. That perspective makes me care about data governance and share credit: dashboards and models only work because engineering and data teams keep the pipeline clean and healthy. When results land, it’s our team effort — not just my modelling — that deserves the applause.

If you’re early in a data/product role, or moving in from a non-STEM background, Naked Statistics is a gentle, reliable companion for learning to think with data.