15 In-Demand Skills to Learn in 2026 (Global Data)

What a data scientist does (list)

A data scientist turns messy organisational data into a number, model or recommendation that someone acts on. The BLS lists the core daily duties for the occupation (BLS, 2025), and in practice a week contains most of the following.

• Pull and shape data. Write SQL against a warehouse, join tables that do not quite match, and work out why yesterday's row count changed.
• Clean and check. Handle nulls, duplicates, time zones and definitions that differ between teams. This is usually the largest single block of time and nobody enjoys it.
• Analyse or model. Fit a model in scikit-learn, run a regression, or produce a straightforward cut of the data that answers the actual question asked.
• Design or read an experiment. Set up an A/B test, decide a sample size, and tell a product manager whether a result is real or noise.
• Communicate the result. Build a chart in Tableau or Power BI, write a short memo, and present it to people who will not read the code.
• Maintain what already runs. Monitor a deployed model for drift, retrain it, fix a broken pipeline, answer questions about last quarter's numbers.
• Scope incoming requests. Sit with a stakeholder and convert "can you look at churn" into a question with a defined output and a deadline.

Where the work happens. The BLS publishes the largest employing industries for data scientists (BLS, 2025) — check that list on the Occupational Outlook Handbook page rather than assuming it is all technology companies. Most of the work is desk work, done in an office, at home, or split between the two, and remote arrangements vary sharply by employer and country.

A typical schedule. Full-time weekday hours for most people, with longer days around launches, quarter-end reporting, audits and model incidents. The BLS describes the standard work environment and schedule for the occupation (BLS, 2025).

The honest part. A large share of analysis never gets used. Stakeholders change their minds, a cleaner data source appears, or the decision gets made before your work lands. You will also spend more time on data quality than on modelling, and entry-level applicant pools are crowded because the title is fashionable. If you need your work to visibly ship every month, aim for roles closer to product or engineering.

How to become one (steps)

There is no single licensed route. The BLS lists typical entry-level education and the related work experience usually needed for the occupation (BLS, 2025), along with the common fields of degree people hold (BLS, 2025) — read both on the Occupational Outlook Handbook page before you commit money to anything.

1. Collect ten real job ads for roles you would take, in your country, and list every tool named. Half a day, free. Do this first: an ad asking for dbt and Snowflake is a different job from one asking for PyTorch, and you cannot prepare for both at once.
2. Learn SQL to the level of joins, window functions and CTEs. Six to ten weeks at a few hours a week. Free using PostgreSQL installed locally and public datasets; paid course prices vary by provider and country, so check the provider's current page. You are done when you can answer a question from three joined tables without looking up syntax.
3. Learn the Python data stack — pandas, NumPy, matplotlib, then scikit-learn. Two to four months. Free with the official pandas documentation and open notebooks. Build the habit of writing scripts, not just notebook cells.
4. Work through a statistics and experiment design course. Four to eight weeks. Cover sampling, confidence intervals, regression and A/B testing, and practise in statsmodels. This is the part most self-taught candidates skip and most interviews test.
5. Build two or three portfolio projects on genuinely messy data and publish them with Git and GitHub. Six to twelve weeks. Free. One project should use data you collected or an open government dataset, include the cleaning decisions in writing, and end with a recommendation — not an accuracy score.
6. Earn one vendor certification that matches your target ads, such as the Microsoft Certified: Azure Data Scientist Associate (DP-100) or the Google Cloud Professional Machine Learning Engineer. Four to ten weeks of study. Exam fees are set by the vendor and differ by country — check the vendor's exam page. Do this after the portfolio, not instead of it.
7. Get applied experience inside a real organisation. Three to twelve months. Ask your current manager for an analytics project, volunteer for a local charity's reporting, or take an internship or analyst role. The BLS notes that work experience in a related occupation is typically needed for data scientist roles (BLS, 2025), so this step is usually the difference between an interview and silence.
8. Rewrite your CV around outcomes and rehearse the take-home. Two to four weeks. Each bullet should name the tool and the result. Ask one working data scientist to review it and to tell you which project they would probe in an interview.

Message to ask for internal project work

Hi [name] — I'm working through SQL and Python for analysis and I'd like to practise on something real. Could I take on [project], say [X] hours a week alongside my current work? I'd deliver [specific output] by [date] and you can decide whether it's useful.

Skills you'll need (list)

The BLS lists the important qualities for the occupation as analytical, computer, communication, logical-thinking, math and problem-solving skills (BLS, 2025). Here is what that means as fifteen learnable things.

Six hard skills to learn first

1. SQL — PostgreSQL, BigQuery or Snowflake. Joins, window functions, aggregation, query performance. Non-negotiable in almost every ad.
2. Python for data — pandas, NumPy, scikit-learn, plus Jupyter for exploration. R is a fine substitute in research-heavy and biostatistics settings.
3. Statistics and experiment design — statsmodels or R. Sampling, regression, confidence intervals, A/B tests, and knowing when a result is noise.
4. Machine learning practice — scikit-learn for tabular problems, PyTorch or TensorFlow if your target ads mention deep learning. Cross-validation and leakage matter more than exotic models.
5. Visualisation and BI — Tableau, Power BI or Looker Studio, plus matplotlib or Plotly for ad-hoc work. A chart someone can read in ten seconds beats a dashboard nobody opens.
6. Cloud and pipelines — one of AWS, Azure or Google Cloud, with Airflow or dbt for scheduling and transformation, and Docker plus Git for reproducibility.

Four more that keep appearing in 2026 job ads

1. Working with large language models — retrieval, evaluation and prompt design, and being able to say honestly when an LLM is the wrong tool.
2. MLOps and monitoring — MLflow or a similar registry, model versioning, drift checks after deployment.
3. Data governance and privacy awareness — knowing what data you may use, and how it should be stored, masked and retained.
4. Domain knowledge — finance, health, logistics, marketing or public sector. The combination of domain plus modelling is rarer than either alone.

Five soft skills, and how to show them

1. Written communication. Show it with a one-page project summary that leads with the recommendation and puts the method underneath.
2. Problem framing. Show it by describing, in an interview, a request you pushed back on and the sharper question you replaced it with.
3. Stakeholder management. Show it with an example of agreeing scope and a deadline in writing before starting work.
4. Scepticism about your own results. Show it by naming the limitation of your portfolio project before the interviewer finds it.
5. Collaboration in code. Show it with a GitHub history that includes readable commits, a README and at least one pull request review.

Pay and outlook (prose)

Pay for data work is usually a base salary, sometimes with a bonus, and in listed technology companies often with equity. Contract and consulting work is billed by day or hour instead. What moves the number is industry first, then location, then seniority, then how close your work sits to money — a model that changes pricing or fraud losses is valued differently from a dashboard.

The BLS publishes the median annual wage for data scientists (May 2025), the spread between the lowest-paid ten per cent and the highest-paid ten per cent (BLS, 2025), and median wages by top-paying industries (May 2025). Read those on the Occupational Outlook Handbook page for data scientists rather than relying on aggregated figures from job boards, which mix titles and countries. If you are outside the United States, check your national statistics office or public employment service for the equivalent occupational data, and treat advertised ranges on job ads as a floor for negotiation rather than a fact.

On demand, the BLS publishes a job outlook for data scientists covering 2025 to 2035, the number of jobs recorded in 2025, projected annual openings over the decade, and the underlying driver of employment growth (BLS, 2025). Check those directly — a projection is not a guarantee, and national figures say nothing about your city.

The honest caveat: demand is strong at the experienced end and crowded at the entry end. Employers hiring their first data scientist usually want someone who has already shipped something. That is why an analyst, engineering or research job first is the common route, and why a portfolio that shows judgement beats one that shows ten tutorials.

Asking about range early in a process

Before we go further, could you share the budgeted range for [role]? I'm targeting around [$X] based on [country/region] market data and my experience with [tool], and I'd rather check we're aligned now than after four interviews.

Career path (timeline)

Titles vary between employers, so treat the years as typical rather than fixed. Progress depends on the size of the team — in a small company you may run the whole function in year two, while in a large one the ladder has more rungs.

Sideways moves are common and useful: into data engineering if you like systems, into product analytics if you like decisions, into research or a specialist scientific role if you like depth. Some of those specialist paths are where the BLS observation that some employers require or prefer a master's or doctoral degree (BLS, 2025) bites hardest.

Frequently asked questions

Sources

1. U.S. Bureau of Labor Statistics — Data Scientists : Occupational Outlook Handbook (2025)