How to Become a Data Scientist in 2025
The advice I want to give when someone asks me what they can do to become a data scientist, but I’m too “nice” to tell them one-on-one.
First, you must actually know statistics.
Too many would-be data scientists overlook acquiring baseline knowledge. They read blogs on interview tips and tricks, study practice questions for technical interviews, and reach out to a million current data scientists on LinkedIn, but it doesn’t help them because they don’t know statistics well enough.
The first step to being a data scientist is to know statistics. You’d be amazed at how easy technical interview questions become if you know statistics.
Of course, it’s harder to learn and know statistics from first principles and be comfortable using it in new problems you haven’t seen before than to read articles and advice telling you some “tactic” or selling you some “preparation” and a battery of practice questions.
For concrete recommendations on how to learn statistics (these are not affiliate links or anything, just textbooks I recommend):
Learn everything in Casella and Berger (Amazon). That’s just a textbook I know. Any book or course that covers the same topics works.
Learn standard estimators and statistical strategies by reading, for example, Bruce Hansen’s Econometrics (Amazon). This book has a lot of great, practical info. Also, it has an appendix on common Matrix Algebra rules that I refer to a hilarious amount more than a decade after I first read it.
Practice proving theorems and deriving estimators. That’ll teach you the math’s mechanics and give you an intuition for how statistics works. This is key because you’ll often run into questions you don’t know the answer to or face a problem where there isn’t an off-the-shelf estimator for the most natural model of the data, but if you know statistics from the ground up, you can figure it out.
If you do these things, you will pass technical interviews. This level of knowledge is way above the current bar.
The other advantage of this level of familiarity with statistics is that — because you actually know statistics instead of a cookbook of statistical procedures — you’ll find it easy to discuss statistics with less technical stakeholders. If you fully grok what a method does, it is easier to explain it in plain language. So, you’ll do better on stakeholder interviews as well.
The second step to becoming a data scientist is to learn to code. You should actually know Python (I am an unabashed R apologist, but the market wants what the market wants). Not just standard libraries used by data science. Know the full language because then you will be more useful. E.g., if the team can’t get engineering resources for a project, it’s okay. You can sub in to make a quick app. This adds to your value. Many great courses on Udemy can get you there (it’s my honest recommendation. No one’s paying me for it — sadly—and all opinions are my own, but as a disclosure, I work for Udemy).
With programming, there’s no way to truly learn it except by doing it. So, write a bunch of code. Build little apps. They don’t even have to have anything to do with statistics. Just get familiar enough that when someone says, “we want to automate this”, the coding part is not a barrier. You know how to do it. The problem is just designing the solution to work best for the team.
Of course, you should also know SQL, but that’s not much of a time investment.
So, those are my two pieces of advice on how to become a data scientist:
Know statistics.
Know programming.
It’s the time of year when people consider changing careers and resolve to get the skills they need to do so. Hopefully, this post will be useful for people looking to change careers to (or start their careers in) data science.
This post deviates from the blog’s usual topic (statistical methodology). If you’d like more career-oriented posts, let me know!
Thanks for reading!
Zach
Connect at LinkedIn: https://linkedin.com/in/zlflynn
If you want my help with any Experimentation, Analytics, etc. problem, click here.