Data Scientist

A data scientist uses statistics, programming and machine learning to find patterns in an organisation's data and turn them into something useful - a forecast, a customer segmentation, a fraud-detection model, or simply a clear answer to a business question. They usually sit inside a company (retail, banking, health, government, tech, consultancy) and work alongside engineers, analysts and the teams who will actually act on the findings. Much of the job is defining the question properly and getting messy data into a usable state, not just building models.

Approximate graduate salary

Broadly, graduate and junior data scientist roles in the UK tend to start somewhere around GBP 28,000-40,000, but this varies a great deal. London and finance or big tech sit at or above the top of that range; public sector, charities, smaller regional employers and the NHS tend to sit lower. Roles requiring a PhD or specialist quantitative skills can start higher again. Treat these as rough approximations only and check current advertised salaries.

What you'd actually do

Writing code - usually Python or R - to pull, clean and reshape data, and SQL to query databases. Cleaning and joining data typically takes up more of the week than modelling does.
Exploratory analysis: plotting distributions, checking for missing or duplicated records, spotting that a field changed definition halfway through last year, and working out whether the data can answer the question at all.
Building and testing models - anything from a logistic regression predicting which customers will cancel, to a forecasting model, to a clustering exercise - and validating them on data the model has not seen.
Talking to the people who own the business problem: a marketing lead, a clinician, a policy team, a product manager. Sessions to pin down what they actually want to know, and later to explain what the model does and does not tell them.
Producing outputs other people can use: a written summary, a dashboard, a slide deck with charts, or a model handed to engineers to run in production.
Code review, version control (Git), and documenting work so a colleague can rerun it in six months.
In many teams, some involvement in getting models deployed and monitored - checking a live model has not degraded as the underlying data shifts.

How graduates get in

Graduate schemes - common in banking, insurance, retail, telecoms, consultancy and large tech firms. These often rotate you through analytics, engineering and reporting teams before you specialise.
Direct entry as a junior/graduate data scientist or data analyst at a smaller company or scale-up. Very common, and often the faster route to hands-on modelling work, though with less structured training.
Starting as a data analyst or business analyst and moving across. This is one of the most common real-world routes - you build SQL and stakeholder skills first, add statistics and Python, and move sideways after a year or two.
The Civil Service Fast Stream has a data-focused route, and government departments, the NHS and local authorities also recruit analysts and data scientists directly. There is a separate government-run data science graduate scheme in some years - check current listings rather than assuming it runs.
A postgraduate MSc in data science, statistics, machine learning or a computational subject. Not compulsory, but genuinely helpful if your first degree was not quantitative, and it is the standard conversion route.
Degree apprenticeships and Level 6/7 data apprenticeships, where you work and study at the same time. Less common than the graduate-scheme route but growing, particularly in large employers and the public sector.
PhD holders in physics, statistics, biology, economics and similar fields move into the role regularly, especially in research-heavy or pharma/biotech teams. This is normal but not required.

What employers ask for

A numerate degree is the usual expectation: maths, statistics, computer science, physics, engineering, economics or a quantitative social science. The subject matters more here than in most graduate roles because employers want to see you have done real statistics and coding.
Many employers ask for a 2:1 or above, some for a 2:1 in a quantitative subject specifically. Smaller companies and scale-ups often care far more about a portfolio of work than about classification.
Programming ability in Python or R, and SQL. SQL is asked for almost universally and is often the thing candidates underestimate.
A grounding in statistics - regression, hypothesis testing, sampling, understanding what a confidence interval means. Employers test this because a lot of applicants can fit a model but cannot say whether the result is meaningful.
An MSc or PhD is required at some employers (particularly research-led, pharma and some finance teams) and completely optional at others. This varies a lot - read the advert rather than assuming.
Evidence of applied work: a dissertation with real data, an internship, a Kaggle-style competition, a personal project on GitHub, or analytical work done in a placement year. Employers vary in how much weight they put on this, but for direct-entry roles it often matters more than anything else.
For public sector, defence and some finance roles you may need UK residency history for security clearance. Check before applying.

Skills that matter

SQL and data wrangling

Most of the data you need lives in relational databases and arrives incomplete, inconsistent or wrongly labelled, and you cannot model anything until you have fixed that.

Applied statistics and scepticism about results

The core value you add is knowing when an apparent pattern is real, when it is a sampling artefact, and when the model has quietly learned something it should not have.

Python or R programming, including version control

Analysis has to be reproducible and reviewable by colleagues, so writing clean, tested code in Git matters as much as getting the right answer once.

Framing a vague business question as an answerable analytical one

Stakeholders usually arrive with something like 'why are sales down?' and your first job is to convert that into a question the available data can actually address.

Explaining technical results to non-technical people

A model nobody trusts or understands does not get used, so you spend real time translating outputs into plain language and honest caveats for people who will make decisions on them.

Data visualisation

A well-chosen chart is often the deliverable itself, and choosing the wrong one is a fast way to mislead a decision-maker.

Domain curiosity

Knowing how the business or service actually works is what lets you spot that a spike in the data is a system change rather than a genuine behaviour shift.

Where it leads

Graduate/junior data scientist: working on well-defined pieces of a larger project, with your code and analysis reviewed by someone more senior.
Data scientist: owning projects end to end, from scoping with the business through to a deployed model or a delivered recommendation, and starting to review others' work.
Senior data scientist: setting the technical approach, mentoring juniors, and being the person who pushes back when a request is not answerable with the data available. Timelines to this point vary widely - a few years at a fast-moving scale-up, longer in a large structured organisation.
From senior, the path usually forks. One branch is technical depth: lead/principal data scientist, machine learning engineer, or a specialist role in NLP, forecasting or causal inference. The other is management: data science manager, then head of data science or analytics director.
Sideways moves are extremely common and not a step down - into data engineering, product management, quantitative research in finance, MLOps, or consultancy. Some people move into contracting, where day rates can be high but security is lower.

What people get wrong

“The job is mostly building and tuning machine learning models, especially neural networks.”

For most UK data science jobs, the majority of the work is finding data, cleaning it, checking it, and explaining results. Plenty of high-value projects end in a regression, a well-built dashboard or a clear written recommendation rather than a deep learning model.

“You need a PhD, or at least an MSc, to get in.”

Some employers do require one, particularly research-heavy teams, but many hire straight from a good quantitative undergraduate degree or promote from analyst roles. The requirement varies enormously between employers, so filtering yourself out early is a mistake.

“It is a solitary technical role where you are left alone with the data.”

A large part of the job is negotiation and translation - working out what a stakeholder really needs, managing expectations about what the data can prove, and persuading people to act on a finding they may not like.

“'Data scientist' means roughly the same thing everywhere.”

The title covers wildly different jobs. At one employer it is essentially advanced business analysis in SQL and dashboards; at another it is production machine learning that overlaps heavily with software engineering; at another it is closer to statistical research. Read the responsibilities in the advert, not the title.

“Once a model is built, the work is done.”

Models degrade as customer behaviour, systems and definitions change, so monitoring, retraining and occasionally retiring models is ongoing work in most production teams.

Where this varies

The role differs sharply by sector and employer size. In large banks, insurers and consultancies you are more likely to join a structured graduate scheme, work within heavy governance and model-validation requirements, and specialise gradually. In scale-ups and smaller firms you may be one of very few data people, doing your own data engineering, dashboarding and deployment - broader experience, less mentoring. In the public sector and NHS the emphasis is often on transparency, reproducibility and open-source tooling, with lower pay but clearer structures and often better flexibility. Some teams are effectively analytics functions using the data science label; others sit inside engineering and expect you to ship production code. Geographically, London, Manchester, Edinburgh, Bristol, Leeds and Cambridge have the densest markets, though remote and hybrid working has widened access considerably.

General guidance about the role across the UK market, not about any specific employer. Entry routes and requirements vary — always check the individual job advert.