How to become a data scientist
Far less modelling than the title suggests. Surveys of the field keep landing on the same rough split — most of the time goes on finding, cleaning and reconciling data, and a minority on the model. The larger surprise is that the hardest part is usually deciding whether the question is answerable at all, and saying so when it is not.
Written by the JobStraight team · pay and hiring figures measured 2026-08-22 · page updated 2026-09-01
The realistic ways in
Which people each route suits, and what it honestly costs.
From data analysis
The most reliable route. You have SQL and domain knowledge already; you add statistics and machine learning. Many people make this move inside one company without changing employer.
From a quantitative postgraduate degree
Statistics, physics, economics, operations research. You have the maths and typically not the engineering — expect the gap to be Git, production code and data pipelines rather than theory.
From software engineering
You can build and ship, which is rarer than it should be in this field. The gap is statistical judgement: knowing when a result is noise, which is a habit rather than a formula.
The title you get hired into first
The titles this role is actually hired under:
- Data Scientist
- Junior Data Scientist
- Data Analyst
- Machine Learning Analyst
- Quantitative Analyst
- Research Analyst
125 live data scientist openings are on the site right now, refreshed on every build and linking to the employer's own posting.
What to learn, in order
Ordered by dependency, not by interest. Durations assume eight to ten hours a week and are an estimate, not a measurement.
-
Statistics that you can defend 8 weeks
Distributions, sampling, confidence intervals, and what a p-value does and does not mean. The foundation everything else stands on.
-
Python for data 8 weeks
pandas, numpy, scikit-learn, and enough visualisation to interrogate your own work rather than present it.
-
SQL and getting the data 4 weeks
The step people skip because courses hand them a clean CSV, and the step the job is mostly made of.
-
Machine learning with a critical eye 10–12 weeks
The main model families, and — more importantly — evaluation, leakage, baselines and drift. Beating a baseline is the job; running a fit is not.
The one piece of work that changes the conversation
One project where you state a hypothesis, test it, and report a negative or ambiguous result honestly. Everyone's portfolio predicts something successfully; almost nobody's shows the judgement to say "the data does not support this". Hiring managers notice, because that judgement is what they are actually buying.
What it pays, measured
From salary figures on live data scientist postings, not a survey. Half sit between the outer two columns.
| Market | Lower quarter below | Midpoint | Upper quarter above | Postings with a figure |
|---|---|---|---|---|
| India | ₹8.5L | ₹15L | ₹23L | 1,398 |
| the UK | £39k | £65k | at least £70k | 1,235 |
| the US | $121k | at least $140k | at least $140k | 4,999 |
How much of this the source estimated rather than read off a posting: India 0%, the UK 29%, the US 70%. The full note, and what it means for each market, is on the salary page below.
The full distribution for each market, the twelve-month movement and the employers posting most of these roles are on the data scientist salary page.
What the postings actually ask for
Counted, not guessed. We read 2,411 job descriptions for the skill-demand study; 108 of them were data scientist roles. This is how often each skill was named in that group.
| Skill | Share of postings | Count |
|---|---|---|
| SQL | 86% | 93 |
| Python | 84% | 91 |
| A/B testing | 49% | 53 |
| Machine learning | 37% | 40 |
| Degree required | 28% | 30 |
| Spark | 23% | 25 |
| Mentoring | 16% | 17 |
| Stakeholders | 9% | 10 |
| Snowflake | 9% | 10 |
| Looker | 9% | 10 |
Read a low percentage carefully. A skill can be near-universal and rarely written down — a team that would not hire someone unable to use version control often stops thinking to ask for it. An assumed requirement is still a requirement.
Who is hiring, right now
Ranked by how often each appears in data scientist advertisements, measured 2026-08-22. Advertisement frequency, not vacancy count — which is why there is an order here and no number.
- India: GENPACT, HP Textile Mill, Oracle, H & R Johnson, Micron Electricals
- the UK: Capgemini, Amazon, IQVIA, RELX Group, JPMorgan Chase
- the US: Deloitte, Capital One, Meta, Oracle, EY
Appearing high can mean growth, turnover, an agency posting for a client, or a bulk feed repeating one advertisement. Research, not a recommendation.
What gets people rejected
Reporting accuracy on an imbalanced problem without naming the baseline. It signals that you optimised a number without understanding what it measured, which is the specific failure this role exists to prevent.
The full question set this role is screened on, each with the shape of a strong answer, is in the data scientist interview questions. The resume structure it is read against is in the data scientist resume example.
How long it really takes
Twelve to twenty-four months from a non-quantitative start, six to twelve from analysis or a quantitative degree. The statistics cannot be rushed and are the part employers probe hardest.
Who finds this harder than expected. People who want the modelling. Expect to spend most of your week on data quality and stakeholder expectations, and to enjoy that or be unhappy.
Common questions
How long does it take to become a data scientist?
Twelve to twenty-four months from a non-quantitative start, six to twelve from analysis or a quantitative degree. The statistics cannot be rushed and are the part employers probe hardest.
Do you need a degree to become a data scientist?
Not usually a specific one — From data analysis; From a quantitative postgraduate degree; From software engineering are all routes people take here. Where a degree matters it is a filter at large employers rather than something the work needs.
What should be in a data scientist portfolio?
One project where you state a hypothesis, test it, and report a negative or ambiguous result honestly. Everyone's portfolio predicts something successfully; almost nobody's shows the judgement to say "the data does not support this". Hiring managers notice, because that judgement is what they are actually buying.
What gets people rejected for data scientist roles?
Reporting accuracy on an imbalanced problem without naming the baseline. It signals that you optimised a number without understanding what it measured, which is the specific failure this role exists to prevent.
Which job title should you apply to first?
Not data scientist necessarily — Data Scientist, Junior Data Scientist, Data Analyst, Machine Learning Analyst are where people are actually hired in.
Is data scientist the right role for you?
It is harder than the internet implies for one group in particular: people who want the modelling. Expect to spend most of your week on data quality and stakeholder expectations, and to enjoy that or be unhappy.
Where to go next
- Live data scientist openings — current vacancies, linking to the employer's own posting.
- Data Scientist salary in detail — full distribution, twelve-month movement, top employers.
- Data Scientist resume example — section order, bullet formula, and a copyable template.
- Data Scientist interview questions — what each round is testing, and what a complete answer contains.
- Check your resume against a real posting — actual skill overlap and the knockout rules, not a keyword score.
- All career paths — the other roles covered the same way.