What the daily work looks like
Most projects start with a vague question from a product or business lead. Your first job is to pin it down: what decision will this inform, what would count as success, and which metric will measure it. O*NET lists identifying business problems that data can address and interviewing stakeholders among the occupation's tasks. A badly framed question wastes weeks of modeling, so senior data scientists spend a lot of time here.
Next comes data. The Bureau of Labor Statistics describes data scientists deciding which data are available and useful, then collecting and cleaning raw, often unstructured data so software can read it. Then you build and test models, compare them with metrics such as loss functions or explained variance, and retest them as new data arrives. Typical outputs include forecasts, customer segments, recommendation systems, fraud detection and A/B test results.
The last step is explaining. You present findings in charts and plain language to technical and nontechnical audiences and recommend what to do. This is where the work turns into a decision, and a model nobody acts on has little value.
How large language models changed the job
Classical modeling is now one part of a wider job. A June 2026 Databricks article says data scientists are increasingly expected to work with large language models, build generative AI applications, and follow models through to deployment and monitoring. It describes two uses of generative AI: as new projects, such as fine-tuning models, retrieval-augmented generation and agents, and as a productivity tool for writing code and exploring data.
Researchers writing in the Harvard Data Science Review argue that LLMs move data scientists away from hand coding, data wrangling and routine analysis toward assessing and managing analyses that AI systems carry out. In practice that puts evaluation at the center. Someone has to decide whether a model's outputs are good enough, design the test set, pick the metric, and check for failure cases. Those are statistics and experiment-design skills, and they matter more now.
Data scientist versus machine learning engineer
The simplest split is ownership. The data scientist owns the question, the experiment and the interpretation: which model, measured how, and what the result means for the business. The machine learning engineer owns the system that serves the model reliably in production, with its pipelines, latency, scaling and monitoring. Databricks puts it the same way and notes that the roles often overlap on small teams and are usually separate at larger organizations.
Expect the line to blur. BLS notes that data scientists with strong coding or engineering backgrounds may build machine learning algorithms and systems, while others focus on research or business strategy. If you enjoy statistics, experiments and explaining tradeoffs, lean toward data science. If you would rather spend your day on infrastructure and production code, machine learning engineering is a better fit.
Skills that matter most
BLS lists analytical, computer, communication, logical-thinking, math and problem-solving skills. In concrete terms that means Python and SQL, probability and statistical inference, supervised and unsupervised learning, experiment design, and the ability to write a clear one-page summary. O*NET lists tools such as TensorFlow, SAS, MATLAB, Docker, GitHub and Kubernetes among the software used in the occupation. Domain knowledge also counts. BLS notes that some employers, such as asset managers, want industry experience or related coursework.
Pay
USD a year, before tax. US figures from BLS OEWS May 2025 for data scientists, moved to 2026 with the Employment Cost Index. BLS has no AI series, so this nearest series covers every data scientist, including those outside AI teams. Job ad rows are midpoints of US base pay ranges in 63 live AutomationRoles.ai ads from 27 employers on 5 November 2026; they lean toward large AI and tech employers and leave out stock and bonus. Refer to the AI Data Scientists salary guide for figures from live adverts.
What you need
- EducationBLS says data scientists typically need at least a bachelor's degree in mathematics, statistics, computer science or a related field. Some employers require or prefer a master's or doctoral degree.
- No licenseNo license or regulator governs data science work. Certifications are optional and do not replace evidence that you can do the work.
- Statistics and mathYou need probability, inference, linear algebra and experiment design to choose metrics, compare models and know when a result is noise.
- ProgrammingYou will write code to clean data, run analyses and build models. Python and SQL are the usual base.
- Evaluation skillWith LLM work, much of the job is deciding whether outputs are good enough. That means building test sets, picking metrics and finding failure cases.
- CommunicationBLS lists communication as a core quality. You must explain results and tradeoffs to people who will not read your code.
How to get there
- Before first roleBuild the foundationsStudy statistics, probability, linear algebra and programming, through a degree in math, statistics, computer science or a related field.
- Before first roleShow applied workFinish projects that go from a messy dataset to a clear recommendation, with honest evaluation and a written summary.
- Early careerJunior data scientist or analystClean data, run analyses and A/B tests, and build baseline models under review from senior staff.
- Mid careerOwn projects end to endFrame problems with stakeholders, choose metrics, design experiments and evaluate models, including LLM-based features.
- SeniorSenior or staff data scientistSet evaluation standards, guide modeling choices across teams, and decide which questions are worth answering.
Open now on AutomationRoles.ai
Questions people ask
Do data scientists on AI teams still train models?
Yes. BLS lists creating, validating, testing and updating algorithms and models as a core duty. On LLM projects, more of the time goes to evaluating and adapting existing models than training from scratch.
Is a data scientist the same as a machine learning engineer?
No. The data scientist frames the problem, builds and evaluates the model, and interprets results, while the machine learning engineer makes the model run reliably in production. On small teams one person often does both.
How much do data scientists earn in the US?
BLS reports a May 2025 median of $120,230 for all US data scientists. Moved to 2026, the middle half earns roughly $88,300 to $163,900. Ads on AutomationRoles.ai, which lean toward large AI employers, show a median range midpoint of $202,500 in base pay.
Will LLMs replace data scientists?
No source read here predicts that. Researchers in the Harvard Data Science Review argue the role is shifting from hands-on coding and routine analysis toward assessing and managing work done by AI systems. BLS projects 35 percent employment growth from 2025 to 2035.
Do you need a PhD to be a data scientist?
BLS says the typical entry requirement is a bachelor's degree, though some employers require or prefer a master's or doctoral degree. Research-heavy roles are more likely to ask for advanced degrees.
