A data scientist needs programming (Python), data access (SQL), statistics and probability, data cleaning and analysis (Pandas, NumPy), visualisation and storytelling, machine learning with honest evaluation, and enough engineering to share or deploy results. Just as important are the non-technical skills: framing business questions, judging what the data can and can't say, and explaining it clearly. Learn them in that order, proving each with a project.

What does a data scientist actually do?

The job varies by company, but most weeks involve a mix of: turning a vague business question into a measurable one, pulling and cleaning data, exploring it to find patterns and problems, running analyses or experiments, building and evaluating predictive models, and presenting findings that lead to a decision. In smaller companies a data scientist may also build dashboards and deploy models; in larger ones those tasks may belong to analysts, ML engineers or data engineers. For the differences between roles, read data science vs ML vs AI engineering.

What are the core technical skills?

SkillWhat it coversProof you have it
PythonCore language, functions, data structures, scriptsClean, reusable analysis code
SQLJoins, aggregation, window functions, CTEsAnswering business questions from raw tables
StatisticsDistributions, sampling, tests, regression, uncertaintyCorrect A/B test analysis
Data wranglingPandas, NumPy, cleaning, reshapingTurning a messy dataset into a tidy one
VisualisationMatplotlib, Seaborn, dashboardsCharts that make one point clearly
Machine learningRegression, classification, trees, boosting, clustering, evaluationA model that beats a baseline on held-out data
Engineering basicsGit, environments, APIs, simple deploymentA project others can run

What order should you learn them in?

  1. Python fundamentals (see how to learn Python from scratch).
  2. SQL, because most data lives in databases.
  3. Statistics and probability (see statistics for data science).
  4. Pandas, NumPy and exploratory data analysis.
  5. Visualisation and storytelling.
  6. Machine learning with proper validation and metrics.
  7. Deployment and MLOps basics, so your work can be used.
  8. Optional specialisation: deep learning, NLP and LLMs, experimentation, forecasting.
Weeks Program Zero spends on the data-science core (Phases 1, 3 and 6)
Programming fundamentals (Phase 1)9
Databases & SQL (Phase 3)5
Data science & classical ML (Phase 6)6

From the published curriculum: Phase 1 is weeks 6–14, Phase 3 weeks 25–29 and Phase 6 weeks 42–47.

Which non-technical skills matter most?

  • Problem framing: turning "why are sales down?" into specific questions and metrics.
  • Scepticism: checking data quality, spotting leakage, distrusting results that look too good.
  • Communication: leading with the answer, using plain language, showing uncertainty honestly.
  • Domain understanding: knowing how the business works makes analyses relevant.
  • Prioritisation: a simple analysis delivered this week often beats a perfect model next quarter.

How do data scientist skills differ from analyst and ML engineer skills?

Skill emphasisData analystData scientistML engineer
SQL and dashboardsVery highHighMedium
Statistics and experimentsMediumVery highMedium
Machine learning modellingLowHighHigh
Software engineering and deploymentLowMediumVery high
Business communicationHighVery highMedium

Which projects prove these skills?

  1. An end-to-end analysis of a public Indian dataset (crop prices, rainfall, public transport) with a written report of findings.
  2. An A/B test analysis on simulated or public experiment data, with effect sizes and uncertainty.
  3. A predictive model that beats a baseline on held-out data, with error analysis.
  4. A deployed model or dashboard that someone else can use.

Keep each project in a clean GitHub repository with a README that states the question, data, method, results and limitations.

How are AI and LLMs changing the role?

AI assistants now speed up routine coding, query writing and first-draft analysis. That raises the value of the skills they can't replace: framing the right question, knowing whether data can answer it, spotting flawed results, and communicating decisions. Many data scientists also now build with LLMs, such as classifying support tickets or extracting fields from documents, which makes basic NLP and LLM skills a useful addition. Read what is generative AI for the background.

What mistakes slow down aspiring data scientists?

  • Jumping to machine learning before being comfortable with SQL and statistics.
  • Collecting certificates instead of building and explaining projects.
  • Using only clean tutorial datasets and never facing real messy data.
  • Reporting model accuracy without a baseline or held-out evaluation.
  • Neglecting communication: an insight nobody understands doesn't change anything.

How do you practise the communication skill?

After every project, write a one-page summary for a non-technical reader: the question, the answer in one sentence, the two or three charts that support it, what the data can't tell you, and the recommended next step. Then explain it out loud in three minutes to a friend outside tech. If they can repeat your main point, you've done it well. Data scientists who can do this reliably are trusted with bigger questions, which is how careers in the field actually grow.

How much maths does a data scientist really use?

Day to day, applied data scientists rely most on statistics and probability: distributions, sampling, confidence intervals, hypothesis tests, regression and Bayesian reasoning. Linear algebra (vectors, matrices, dot products) and basic calculus (derivatives and gradients) explain how models like regression, PCA and neural networks work, which matters when debugging or choosing methods. You rarely derive formulas by hand at work, but you need enough understanding to know when a result is suspicious, why a model overfits, or why scaling changes an outcome. Learning the maths alongside code, when each concept is needed, works better for most people than studying it in isolation first.

Which tools do data scientists use besides Python?

  • Jupyter notebooks for exploration, and scripts or packages for repeatable pipelines.
  • Git and GitHub for version control and collaboration.
  • SQL databases and warehouses where company data lives.
  • BI tools such as Power BI, Tableau or Looker Studio for dashboards shared with business teams.
  • Cloud notebooks and compute for larger data and model training.
  • Experiment tracking such as MLflow (see MLflow explained).

Tools change quickly; the underlying skills of querying, analysing, modelling and communicating transfer between them.

How do you show these skills to employers?

A strong profile combines three things. A GitHub portfolio with two to four complete projects, each with a clear README covering question, data, method, results and limitations. A short written piece, such as a blog post or LinkedIn article, explaining one analysis in plain language, which proves communication skill. And the ability to walk through a project in an interview, including what went wrong and what you'd change. Certificates help most when they're earned through tests and reviewed projects and link to this evidence. Recruiters skim; make the first screen of each repository answer "what did this person find, and how?" in under a minute. Pin your best projects, and remove half-finished ones that dilute the impression.

Above all, keep learning in public: share what you build and what you learn each month. Consistent, visible progress builds credibility faster than any single credential.

How can you learn these skills in a structured way?

Program Zero's data science course brings together Python, SQL and NoSQL databases, NumPy, Pandas, EDA, statistics and classical machine learning, ending with an end-to-end deployed project; the programme then continues into deep learning, LLMs and MLOps for anyone who wants to go further. See the curriculum for the week-by-week order.