You have seen 'data scientist' on dozens of job listings and heard it pays well. The median US salary is $112,590, according to the Bureau of Labor Statistics (BLS 2024) -- a genuine, strong number. What we want to be honest about is what that salary actually buys from you: according to multiple industry surveys, 60 to 80 percent of a working data scientist's day goes to collecting and cleaning imperfect data, not to training the models that made the career sound glamorous. If you want the real picture before you commit months and dollars to a pivot, this is that article.
What a data scientist actually does all day
Plain EnglishWhat is Data Scientist?
A professional who uses statistics, programming, and domain knowledge to extract insights from data and communicate them to decision-makers. Think of them as a translator between raw numbers and business decisions -- someone who can run a statistical test, interpret what it means, and explain it clearly to a VP who has never written a line of code.
A data scientist's core job is to answer business questions with data. That sounds clean. In practice, most projects begin by discovering that the data needed is incomplete, mislabeled, stored across three different systems, and missing values in exactly the columns that matter. Before any analysis begins, data has to be collected, merged, cleaned, and validated. Industry surveys consistently place this phase at 60 to 80 percent of working hours (KDNuggets 2024) -- not because the tools are bad, but because real-world data reflects real-world messiness.
The other 20 to 40 percent is more interesting: exploratory analysis to spot patterns, hypothesis testing to check whether those patterns are statistically real, building predictive models when prediction is what the business needs, and communicating findings to stakeholders who do not think in terms of p-values or confidence intervals. This last part -- the communication -- is where careers are made and lost. A data scientist who cannot explain a model's output to a non-technical executive will never see that model deployed.
On any given week, a data scientist might write SQL to pull a dataset from a data warehouse, design and analyze an A/B test to evaluate whether a new product feature improved user retention, build a regression model to forecast next quarter's revenue, and then present all of it in a slide deck to a non-technical audience. The unifying thread is not machine learning -- it is quantitative reasoning applied to business decisions.
Data scientist vs data engineer vs ML engineer: where the real differences are
Plain EnglishWhat is Data Engineer?
A software engineer who builds and maintains the systems that collect, store, and move data reliably at scale. Where a data scientist asks 'what does this data mean?', the data engineer asks 'how do we get the data here reliably?' They are infrastructure builders; data scientists are interpreters of what that infrastructure surfaces.
The distinction is simpler than most job postings suggest. A data engineer builds the pipes; a data scientist reads the water. In practice: if an e-commerce company wants to predict which customers will churn, the data engineer builds the pipeline that collects transaction history and user behavior from a dozen tables into a clean, queryable dataset. The data scientist takes that dataset, builds and validates the churn model, and presents findings to the marketing team. <a href="/learn/what-does-a-data-engineer-do-2026">Our data engineer explainer</a> covers that side of the stack in full detail.
| Feature | Data Scientist | Data Engineer |
|---|---|---|
| Core question | What does this data mean? | How do we collect and store this data reliably? |
| Primary skills | Python, statistics, ML, stakeholder communication | SQL, Spark, dbt, Airflow, cloud platforms |
| Daily output | Insights, models, presentations, dashboards | Pipelines, data warehouses, data quality systems |
| Code style | Jupyter notebooks, analysis scripts, pandas | Production-grade Python or Scala, infrastructure-as-code |
| Closest analogy | Business analyst + statistician + model builder | Software engineer specializing in data infrastructure |
| At smaller companies | Often covers some data engineering too | Often covers some analysis and modeling too |
The third role that creates confusion is the machine learning (ML) engineer. An ML engineer takes the models a data scientist builds and makes them run reliably in production at scale -- thousands or millions of predictions per second, with monitoring, retraining pipelines, and alerting when model performance degrades. The data scientist owns the research and the model hypothesis; the ML engineer owns the deployment and the infrastructure. At large tech companies these are distinct teams with separate interview tracks. At startups with fewer than 50 engineers, one person often covers all three roles under a single job title.
The 2026 data scientist job market: honest numbers
The BLS projects roughly 23,400 new data scientist openings per year through 2034, driven by demand across technology, financial services, and healthcare (BLS 2024). Currently, about 245,900 data scientists are employed in the US. The BLS expects that figure to reach 328,300 by 2034 -- a net addition of more than 82,000 roles over a decade. These are employer-survey numbers, not self-reported, making them the most statistically reliable estimate available.
“Data scientists spend 60 to 80 percent of their time collecting, cleaning, and formatting data. The surprising part is not the percentage -- most practitioners already know this. The surprise is that nearly all data science courses skip this phase almost entirely.”
The market is bifurcating in a way that newcomers need to understand. LinkedIn's 2026 Jobs on the Rise report ranked AI Engineer -- the production-adjacent role that data scientists often transition into -- as the fastest-growing title in the US, up 143 percent year over year in 2025 (LinkedIn 2026). Entry-level general analyst postings are thinning as AI tools absorb basic reporting work. Mid-level and senior positions with ML specializations are growing. The career-switcher targeting ML and production AI deployment is entering a growing market; the one aiming for legacy business intelligence reporting is not.
The skills signal in job postings is specific. Machine learning appears in 77 percent of data science listings. Natural language processing demand tripled in a single year -- from 5 percent to 19 percent of listings between 2024 and 2025. Roles explicitly mentioning large language models or MLOps carry a 15 to 25 percent salary premium over comparable general data science titles. Robert Half projects 4.1 percent year-over-year salary growth for data scientists in 2026, the highest rate of any technology specialty in their survey (Robert Half 2026).
What a data scientist actually earns: BLS vs Glassdoor vs Levels.fyi
Three salary sources give you three different numbers, and all three are accurate for different populations. The BLS median of $112,590 (BLS 2024) is the most statistically rigorous: it surveys employers directly, covers all company sizes and industries, and includes government and non-profit roles where pay is lower. The Glassdoor median of $156,356 (Glassdoor 2026) draws on 57,510 self-reported salaries skewed toward tech-company employees more likely to submit data. Levels.fyi puts median total compensation at $177,000 for data scientists (Levels.fyi 2026), but that figure is almost entirely FAANG and high-growth tech -- it represents the ceiling, not the market.
What actually drives the variance is industry and company size. At the BLS entry level, the bottom 10 percent earns below $63,650 -- typically academic research or non-profit positions. The top 10 percent earns above $194,410, almost entirely in big tech or quantitative finance (BLS 2024). The honest first-year salary for a career-switcher landing their first data science role outside FAANG is typically $85,000 to $110,000. Data scientists at JPMorgan, Capital One, or major hedge funds often out-earn those at mid-size software companies because their models have direct, measurable revenue impact.
| Statistics foundations (Coursera, University of Michigan) 2 months at $50/mo -- cancel after completing the specialization | $100 |
| Python for Data Science and ML bootcamp (Udemy) On-sale price -- Jose Portilla's bestselling course, 4M+ students | $15 |
| Machine Learning Specialization (Coursera, Andrew Ng) 3 months at $50/mo -- the canonical ML preparation path | $150 |
| Google Professional ML Engineer exam (optional) Worth adding if targeting GCP-first employers; requires 3+ years industry experience | $200 |
| Kaggle competitions and public datasets Free -- builds a verifiable track record with real, messy datasets | $0 |
| Total | $465 (self-study) vs $12,000-$18,000 (bootcamp) |
The market is real -- 33.5% BLS growth through 2034, 23,400 new openings per year -- but the entry bar is higher than bootcamp marketing implies. With a quantitative background (finance, engineering, biology, economics, or psychology with real statistics coursework), budget 6 to 12 months for Python, ML tooling, and a portfolio of 3 to 5 end-to-end projects. Without that foundation, budget 18 to 24 months. The honest catch: data science involves open-ended questions, data that resists clean interpretation, and stakeholders who change what they wanted to know mid-project. If ambiguity frustrates you, pick a role with clearer success metrics -- data engineering and ML engineering both fit that profile and pay similarly. If open-ended problems energize you, this is one of the most durable tech careers available. The <a href="/careers/data-scientist">full data scientist career roadmap</a> has the step-by-step learning sequence and realistic timelines for each background type.
What most articles miss: the skills that actually get you hired
The most underweighted skill in data science education is SQL. Every practical data science role starts with querying a production database -- not a pre-cleaned Kaggle dataset loaded into a Jupyter notebook. The production database has dozens of tables, inconsistent naming conventions, missing values in critical columns, and edge cases that were never documented. Nearly every technical screen at companies that hire data scientists at scale tests SQL: window functions, aggregations across multiple joins, and CTEs for complex queries. Python and scikit-learn rank second.
For machine learning preparation, <a href="https://www.coursera.org/specializations/machine-learning-introduction">Andrew Ng's Machine Learning Specialization on Coursera</a> (Stanford/DeepLearning.AI, 4.3 million enrolled learners) is the most widely used starting point and covers both theory and implementation. However, the statistics prerequisite -- probability, hypothesis testing, regression, correlation vs causation -- should come before that ML content, not from it. The specialization assumes you already understand when a result is statistically significant but practically meaningless.
- SQL -- window functions, CTEs, multi-table aggregations: tested in nearly every data science technical interview at any company above startup size
- Python with Pandas, Matplotlib, and NumPy: the primary analysis and modeling environment for most working data scientists
- Statistics: hypothesis testing, A/B test design, regression, and knowing when a correlation is spurious rather than causal
- Machine learning fundamentals with scikit-learn: model evaluation, cross-validation, and knowing when NOT to use ML at all
- Data storytelling: presenting findings to non-technical stakeholders in decision-relevant terms, not academic paper format
- Domain knowledge: healthcare, finance, and retail data science each require industry-specific context that generic certification programs do not teach
The skill that determines long-term career trajectory more than any technical credential is the ability to translate findings into decisions. A data scientist who builds a 94-percent-accurate churn model but cannot explain to a VP which customers to contact, in what order, with what message, and why the model should be trusted -- that model will not get deployed. This communication skill appears almost nowhere in online courses and disproportionately in the actual job. Our <a href="/learn/day-in-the-life-junior-data-scientist-startup-2026">day in the life of a junior data scientist</a> covers what this gap looks like in practice at an early career level.
Who should (and should not) become a data scientist
Not everyone who is interested in data should pursue a data scientist title. The role filters for a specific combination of statistical thinking, programming fluency, and comfort with ambiguity. Understanding where you fall in that profile before investing a year in preparation is worth more than any single course.
- BLS projects 33.5% job growth through 2034 -- genuine, durable demand across every major industry sector
- $112,590 US median salary (BLS 2024), with senior roles in tech and finance regularly exceeding $194,410
- Remote-friendly: most analysis work is asynchronous and does not require physical presence in an office
- Transferable across sectors -- the same skill set applies in healthcare, finance, retail, government, and technology
- High demand for people who can bridge technical and business contexts, a combination that remains relatively rare
- 60 to 80 percent of the actual job is data cleaning and validation, not model building or high-level strategic analysis
- Entry-level market is thinning as AI tools automate basic analytics tasks; differentiation now requires ML specialization
- The 'data scientist' title means very different things at a startup vs a large enterprise -- read the job requirements, not just the title
- Requires a genuine statistics foundation that most bootcamps and many online programs significantly underteach
- Open-ended, ambiguous problem scopes -- if you need well-defined daily tasks and fast feedback cycles, this role will drain you
One path many career-switchers overlook is starting as a data analyst. The analyst role uses SQL and Python for reporting and dashboard work, involves less statistical rigor than a full data science position, and pays $75,000 to $95,000 at entry level in most US markets. It is a real job with real scope, not a consolation prize -- and it provides 18 to 24 months of production data experience that dramatically strengthens data science applications later. Many practicing data scientists at major tech companies started with an analyst title and moved up from inside. For those targeting the ML engineering track after gaining experience, the <a href="/certifications/google-ml-engineer">Google Professional ML Engineer certification</a> ($200 exam) is the clearest GCP credential, though Google's stated prerequisite is three or more years of industry experience.
- If You have a quantitative background (finance, engineering, biology, economics) and want to move into tech → Strong fit. The statistics foundation transfers directly. Budget 6 to 12 months to add Python, ML tooling, and a portfolio of 3 to 5 projects.
- If You are switching from a non-quantitative field and are comfortable learning statistics from scratch → Possible but longer. Budget 18 to 24 months. Start with statistics before Python. Consider a data analyst role as an 18-month stepping stone first.
- If You want to build production software systems and care about infrastructure and scalability → Data engineering or ML engineering fits better. Clearer success metrics, faster feedback loops, and production systems work that suits engineering-minded thinkers.
- If You want to work with AI models but find statistics off-putting or frustrating → Consider AI product management or starting as a data analyst. Data science requires genuine comfort with statistical uncertainty that cannot be shortcut.
For the full compensation picture broken down by city, company tier, and years of experience, see our <a href="/learn/data-scientist-salary-guide-2026">data scientist salary guide</a>. The hourly math, what a $130,000 base actually takes home in New York versus Austin, and the difference between base pay and total compensation for FAANG versus non-FAANG roles are all covered there.
What is the difference between a data scientist and a data analyst?+
A data analyst reports on what happened: querying databases, building dashboards, and tracking business KPIs. A data scientist goes further -- designing experiments to test whether a change actually caused an outcome, building predictive models, and applying statistical rigor to questions where the answer is not already in the data. The roles share SQL and Python but diverge sharply in statistical depth and the complexity of questions they address.
Do I need a master's degree to become a data scientist?+
No. Many working data scientists hold undergraduate degrees or completed structured online programs paired with strong portfolios. A master's in statistics, computer science, or a quantitative field accelerates the path to senior roles at research-heavy organizations, but it is not required for most industry positions. Hiring managers at major tech companies evaluate portfolios and technical screen performance first, not degree pedigree.
Is the data scientist role being automated by AI?+
Partially. AI tools handle low-level tasks such as basic data cleaning, simple exploratory analysis, and standard model selection. What cannot be automated is framing the right business question, validating model assumptions, knowing when ML is the wrong tool, and explaining outputs to non-technical decision-makers in terms they can act on. AI is raising the floor on what a data scientist must do to add value -- the role is evolving, not disappearing.
How long does it realistically take to become a data scientist?+
With a quantitative background, 6 to 12 months of focused preparation -- Python, statistics, ML fundamentals, and a portfolio of 3 to 5 real end-to-end projects. Without a quantitative background, 18 to 24 months is more realistic. The BLS projects 23,400 new openings per year through 2034, so job demand is not the constraint -- preparation quality is.
Is a $15,000 data science bootcamp worth it for career-switchers?+
For most career-switchers, a self-study path costing $465 to $650 produces the same portfolio outcomes as a $15,000 bootcamp -- employers evaluate portfolio quality and technical screen results, not the credential. Bootcamps add genuine value when you need cohort accountability, live mentorship, or career placement services you cannot replicate alone. If you choose a bootcamp, verify it has a publicly auditable job placement rate, not just marketing testimonials.
What is the actual difference between a data scientist and an ML engineer?+
A data scientist asks 'what can we learn from this data and what model might help?' An ML engineer asks 'how do we get this model running reliably in production at scale?' The data scientist owns the research and model architecture; the ML engineer owns the deployment pipeline, monitoring, and retraining infrastructure. At smaller companies one person covers both roles; at large organizations they are separate teams with different interview tracks and different performance metrics.
The data scientist title covers a wide range of actual jobs -- from BI-adjacent analyst work at small companies to production ML research at major tech firms. The most useful question to ask any prospective employer: 'What does a typical project look like end-to-end, and what percentage of my time goes to data preparation versus modeling versus stakeholder communication?' The answer will tell you more about the actual role than any job description ever will.