Most data scientists get there in one of three ways: they start as data analysts and grow into modeling work, they complete a quantitative master's or PhD, or they are domain experts (in finance, healthcare, marketing, science) who learned enough statistics and programming to answer questions in their own field. What all three have in common is solid statistics, everyday fluency in SQL and Python or R, and the ability to turn a vague business question into an analysis someone acts on.
The job is less glamorous than its reputation and more useful. Here is what it involves, what you need to learn, and the route that fits your starting point.
Pay and outlook
The Bureau of Labor Statistics counts about 275,600 data scientist jobs in 2025, with a median annual wage of $120,230 in May 2025. The lowest 10 percent earned less than $67,240 and the highest 10 percent more than $199,130.
The projected growth is the headline number. The BLS expects employment of data scientists to grow 35 percent from 2025 to 2035, adding about 95,400 jobs, compared with 3 percent for all occupations and 25 percent for mathematical science occupations as a group. It projects about 24,800 openings a year on average. The reasons it gives are more demand for data-driven decisions, more data to work with, and companies continuing to build AI-based systems into their workflows.
Pay varies by industry. Among the top-employing industries the BLS lists for May 2025:
| Industry | Median annual wage |
|---|---|
| Publishing, broadcasting and content providers | $142,240 |
| Computer systems design and related services | $132,380 |
| Credit intermediation and related activities | $129,490 |
| Management of companies and enterprises | $128,050 |
| Insurance carriers and related activities | $108,650 |
A 35 percent growth projection does not mean entry is easy. Fast growth in total jobs and stiff competition for junior roles can both be true at once. Plan for a competitive search.
What the job really is
Strip away the hype and most data science work looks like this: someone asks a question (why did churn rise, which customers should we target, will this change improve conversion, how many units should we stock), and you work out whether the data can answer it, get the data into shape, run an appropriate analysis or model, and explain the result in a way that changes a decision.
The BLS duty list matches that: determine which data are available and useful, collect and clean it, build and validate models, use visualization to present findings, and make recommendations to stakeholders. Cleaning gets its own mention for a reason. Messy, incomplete or badly joined data can take up much of a project, and the people who handle it carefully are the ones whose conclusions hold up.
The BLS also notes that some data scientists focus on a particular area, such as building machine learning systems, research, or business strategy. In practice the role splits roughly three ways:
- Analytics-focused. Experiments, metrics, forecasting, causal questions, dashboards that people actually use. Closest to the business.
- Machine-learning-focused. Building, evaluating and deploying predictive models inside products. Closest to engineering; often overlaps with machine learning engineer roles.
- Research-focused. Methods development, often in large tech companies, labs or academia. Often expects a PhD.
Job titles do not map cleanly to these. Read the posting's description of the work, not the title.
The math, statistics and programming you need
The BLS says data scientists typically need at least a bachelor's degree in mathematics, statistics, computer science or a related field, and that some employers prefer or require a master's or doctorate. Whatever the credential, the skill requirements look like this.
Statistics (the most important and most neglected). Probability, distributions, sampling, hypothesis testing and its limits, confidence intervals, regression (linear and logistic) and how to interpret it, experimental design and A/B testing, and the difference between correlation and a causal effect. Many weak candidates can fit a model but cannot say whether its result means anything.
Math. Linear algebra and calculus to the level of understanding how common models are fit and why they fail. The BLS specifically names linear algebra, calculus, and probability and statistics as preparation. You do not need to derive every proof, but you need to recognize when a method's assumptions are broken.
Programming.
- SQL, to a high level: joins, window functions, aggregations, and the habit of checking row counts after every join. Most business data lives in databases, so expect SQL to be tested.
- Python (pandas, NumPy, scikit-learn, a plotting library) or R. Pick one and get fluent; learn the other later if a job needs it.
- Git, notebooks used sensibly, and enough software practice to write code someone else can rerun.
Communication. The BLS lists communication skills alongside math and computing. A correct analysis that nobody understands has no effect. Practice writing a one-page summary that leads with the answer and states its uncertainty plainly.
Domain knowledge. The BLS gives the example of asset management firms that may want finance experience or coursework. This is common across fields: an analyst who understands insurance claims or clinical data is easier to hire than one who has to learn the domain from scratch.
Common entry routes
| Route | Who it suits | Watch out for |
|---|---|---|
| Analyst first | People with a bachelor's degree and some SQL who need to earn while learning | Analyst roles that are only reporting; look for ones with experiments or modeling |
| Quantitative master's or PhD | Those aiming at ML-heavy or research roles, or who lack a quantitative undergraduate degree | Cost and time; a degree without applied projects still leaves a gap |
| Domain expert transition | People already strong in a field such as finance, healthcare or marketing | Underestimating the statistics; learning tools without the theory |
| Software engineering transition | Developers who like statistics and want to move toward ML | Treating modeling as purely an engineering problem |
The analyst route deserves special attention because it is a practical path for people without a graduate degree. Data analyst, business intelligence and analytics engineering roles teach you SQL, real data quality problems, stakeholder management and how decisions get made. Move toward work with experiments, forecasting or modeling, and the shift to a data scientist title is often an internal promotion rather than a new job search.
A graduate degree helps most when the specific jobs you want list it, so check postings in your target area before enrolling. For analytics-focused roles, demonstrated work can carry as much weight.
Adjacent occupations are worth knowing too. Nearby options include computer and information research scientists, who typically need a master's, and software developers, who build the systems models run in. Database administrators and systems analysts also work close to the data and can move toward analytics.
A portfolio that helps
Kaggle-style projects on famous public datasets (Titanic survival, iris flowers, house prices) show that you completed a tutorial. They do not show judgment. A useful portfolio shows how you think from question to decision.
Build two or three projects like this:
- Start with a real question, ideally from a field you know. "Which factors predict late deliveries in this public transport data?" beats "I trained five classifiers."
- Get messy data yourself. Public government data, APIs or scraped data (within the site's terms) that needs real cleaning. Document what you fixed and why.
- Choose a simple baseline first, then justify any added complexity with an honest comparison.
- Write up the result as if for a manager: the answer, the confidence, the limitations, what you would do next.
- Make it reproducible. A repository with a README, a requirements file and code that runs from raw data to final chart.
One project that includes a properly designed experiment or a careful causal analysis is especially persuasive, because those skills are hard to fake and sit at the core of analytics-focused work.
How AI tools are changing entry-level work
AI coding assistants and chat-based tools can now draft SQL queries, pandas code, charts and even first-pass analyses from a plain-language prompt. That makes routine query writing faster, and the BLS growth narrative explicitly expects data scientists to help businesses adopt AI. It is reasonable to expect that some routine entry-level tasks, such as writing standard reports or boilerplate data-cleaning code, will take less time than they used to.
What the tools do not do reliably is decide whether an analysis answers the right question, notice that a join silently duplicated rows, recognize a biased sample, or judge whether an effect is real. Those are the skills that separate a data scientist from someone who can run code, and they matter more when code is cheap to produce.
The practical advice: use the tools, but be able to check their output line by line. In interviews and on the job, expect to defend every number you present, regardless of who or what wrote the code.
The next step
If you have a quantitative background, pick one question in a domain you know, find a messy public dataset that addresses it, and produce a reproducible analysis with a one-page write-up this month. If your statistics are shaky, fix that first: work through a solid introductory statistics course, including regression and experimental design, before adding more machine learning. And if you are already in an analyst role, ask for one project involving an experiment or a predictive model. It is the shortest path to the title.