Finds patterns and builds predictive models from data — a 2008 job title built on three centuries of counting, testing and visualizing evidence.
Also called: Applied Scientist · Machine Learning Scientist
Reviewed 2026-08·Media credits
Darker cells mean a higher score for this topic on that metric.
LessMore
Each bar is a 0-100 atlas score for this topic, not a timeline.
Pay and automation resistance on the same 0-100 scale for this profession.
Both scores stay visible. Drag the slider to emphasize Pay or Resists AI.
Pay
Resists AI
One point: this profession on the two named axes.
Typical years of training before someone usually works in this role.
About 6 years of training before typical field entry
6Years
Milestones in order. This is history, not a weekly activity grid.
Share this career map with someone weighing study, retraining or a first offer.
Last reviewed Sources & creditsMedia creditsMethodology
Data Scientist: Finds patterns and builds predictive models from data — a 2008 job title built on three centuries of counting, testing and visualizing evidence.
A data scientist extracts patterns, predictions and recommendations from data using statistics, programming and domain knowledge, then turns the result into something a business, government agency or research team can act on. The work spans writing SQL to pull data out of a warehouse, cleaning and reshaping it, fitting a statistical or machine-learning model, and explaining what the result does and does not support.
The job title is far younger than the craft behind it. 'Data scientist' dates to around 2008, when DJ Patil and Jeff Hammerbacher independently settled on it while building data teams at LinkedIn and Facebook, and statistician William S. Cleveland proposed 'data science' as an academic field only in 2001. The underlying discipline is centuries older, tracing through John Tukey's 1960s-70s push for exploratory data analysis back to Ronald Fisher's 1920s statistics and John Graunt's 1662 study of London's mortality records.
A 2012 Harvard Business Review article calling it 'the sexiest job of the 21st century' triggered a decade-long corporate hiring boom and hundreds of new university programs. That boom has since matured: the generalist title has split into analytics engineering, machine learning engineering and product analytics, and AI tools now handle a growing share of the routine querying, cleaning and charting the original job assumed a person would do by hand.
Data science is the craft of turning imperfect records into decisions, where the hard part is rarely fitting a model and usually deciding what the data can honestly support.
Working data scientists query warehouses, reconcile definitions, inspect distributions, clean broken fields and explain uncertainty. Modeling is important, but a result only earns trust when its data lineage, assumptions and alternatives can be examined. The strongest practitioners can tell a stakeholder that a confident-looking chart answers the wrong question, or that an observed association is not evidence of a cause.
A product data scientist may design experiments and measure feature outcomes. An applied scientist may develop predictive models. A decision scientist may focus on operations, finance or policy. In mature organizations, analytics engineering, data engineering and ML engineering absorb parts of the older generalist job. Domain knowledge changes the work as much as technical stack: an experiment on a shopping app is not a study of patient outcomes or fraud.
Statistics, computer science, mathematics, economics and physical sciences all provide routes. A degree helps, but an interview or portfolio usually tests SQL, statistical reasoning, data quality and communication more directly. A credible project does not need a flashy model; it needs a clear question, documented data, an appropriate baseline and conclusions matched to the evidence.
Assistants can draft SQL, charts, summaries and baseline models. That genuinely changes entry-level work. It does not settle measurement definitions, expose a hidden confounder, choose an ethical use for a prediction or accept accountability for a decision. The profession is moving toward problem framing, experimentation, causal reasoning and review of machine-generated analyses.
Five common shapes of the same title — specialty, setting or career path.
Uses metrics and experiments to guide product choices, while defending against misleading short-term optimization.
Builds and evaluates models for ranking, forecasting, detection or personalization, often alongside ML engineers.
Combines data with business or public-policy context to support planning, allocation and risk decisions.
Designs controlled tests and quasi-experiments that distinguish an intervention from correlation.
Works with complex observational or experimental data where reproducibility and methodological rigor outrank rapid product shipping.
Same craft, different gatekeeping, status and daily texture — rewritten for readers in each language.
Technology, finance and health sectors use the title differently, and recruiting often tests SQL, experiments and communication explicitly. High compensation is concentrated in a few cities and employers.
Large consumer platforms, telecoms and conglomerates support analytics work, with Korean-language context often vital for product and customer data. Job titles can blur analyst and scientist roles.
Manufacturing, finance and consumer firms are expanding analytics while maintaining established data-governance practices. International teams may value English and cloud experience alongside local domain knowledge.
Data protection and worker representation shape what data can be collected and used. Automotive, manufacturing and research employers create demand for people who can bridge statistics and real operations.
Finance, consulting and product companies cluster heavily around London, while public research and universities offer different career incentives. Contract roles remain visible in the market.
Banks, government technology and regional headquarters create data roles spanning multiple markets. Regulatory expectations and cross-border data rules are frequent constraints.
Commons CC/PD images self-hosted for this profession.
A model can be statistically correct and still mislead a decision-maker; whether a data scientist says so before the deadline, not just when asked, is where attitude outweighs technique.
An analysis that says a feature everyone wants to ship doesn't move the metric they hoped, or that a beloved growth channel is really just picking up existing demand, is unwelcome and easy to soften. Data scientists who report the inconvenient result anyway are the reason a company's decisions track reality; the ones who shade a finding toward what leadership wants to hear turn analytics into decoration.
A five-day experiment or a thin cohort can produce a number that looks precise and is actually noise, and almost nobody downstream will check the sample size before acting on it. The attitude that separates a trustworthy analyst is refusing to let a number sound more certain than the data underneath it actually is, even when a confident number is the one a stakeholder wants.
A model that denies a loan, ranks a résumé, or flags a patient does something to a person who never sees the notebook it came from. Because the model's author is usually the only one positioned to notice a subtle bias or a leaking feature before deployment, taking that responsibility seriously — rather than treating the model as someone else's problem once it ships — is where the harm actually gets caught.
Five concrete postures the work rewards, not slogans.
Looks at distributions, outliers and missing values with their own eyes before fitting anything, on the assumption that a model will quietly absorb whatever is broken in the data rather than flag it for anyone.
Reports a confidence interval or a caveat alongside a headline number instead of only after it is challenged, so a stakeholder cannot mistake a rough estimate for a precise one simply because the uncertainty was left unsaid.
Treats a surprising, convenient-looking correlation as a problem to investigate rather than a result to present, because the more a finding flatters the team's preferred plan, the more it deserves scrutiny.
Abandons a favorite hypothesis once the evidence stops supporting it, rather than running one more cut of the data hoping for a result that finally survives the test, and says so plainly in the write-up.
Tells a stakeholder plainly that the available data cannot support a confident answer, instead of producing a number anyway simply because a number was requested and a blank slide felt unacceptable to present.
Situations that separate résumé language from how someone actually practices.
Leadership already announced the launch date before the A/B test finished. The results show no real lift. Whether the analyst reports that plainly, or finds a secondary metric that looks better, reveals what the job actually is.
Aggregate accuracy looks strong, but a breakdown by demographic or region reveals the model fails one group disproportionately. Noticing this before deployment, unprompted, separates analysts who check from ones who report the average and move on.
A three-week-old dataset gets asked to justify a firm annual growth projection. Delivering a confident-sounding figure is easy and welcome; explaining why the data can only support a rough range is harder and far less popular.
A widely used metric has a subtle bug that happens to make performance look good, and nobody upstream has complained about it. Flagging it means admitting a mistake and disappointing people who liked the old number.
Companies that market themselves as rigorously data-driven sometimes use that language to pressure analysts into finding support for a decision already made, treating one who reports an inconvenient result as not a "team player." The field's early hype as a glamour job also normalized unpaid overtime dressed as curiosity — staying until midnight for one more cut of a dataset gets praised as passion, not questioned as a workload problem.
High
A meaningful share of routine data-science work — SQL querying, exploratory charting, fitting a standard model with AutoML, writing boilerplate pipeline code — is already automatable or AI-assisted today, and that share is growing quickly. What resists automation is framing an ambiguous business problem as a testable question, judging whether a result is real or a statistical artifact, and being accountable when a model's real-world decision turns out to be wrong.
AI & The Future →Majors on this site that list this profession among graduate paths.
Open major encyclopedia →
⚗️Open major encyclopedia →
📊Open major encyclopedia →
🧮Open major encyclopedia →
⚛️Open major encyclopedia →
📉Open major encyclopedia →
🧫Open major encyclopedia →
💱Open major encyclopedia →
🗣️Open major encyclopedia →
Closest neighbours on the six-score profile — not the same field only.
The keeper of the books: heir to a craft so old it invented writing itself, now negotiating with the software built to automate it.
AI-resistant 35 🗺️Decides what a company should build next, and why — turning customer needs, business goals and engineering limits into one shared plan nobody else fully owns.
AI-resistant 50 💻Writes, tests and maintains the code that runs modern life — and is one of the first professions watching AI automate its own daily work.
AI-resistant 35 📣The professional who creates demand — from Pompeii's painted walls and P&G's 1931 brand-man memo to the auction-driven feeds of the digital era.
AI-resistant 38 📦Builds the systems that train, deploy, monitor and govern machine-learning models in production.
AI-resistant 54 🔎Studies how people use products, turning observed behavior, needs and frustrations into evidence teams can design around.
AI-resistant 69Derives and tests the mathematical laws governing matter, energy, space and time, from a lone chalkboard to a 3,000-author particle-collider paper.
AI-resistant 65 🧬The scientist who studies life itself, from Linnaeus naming species by hand to editing genomes with CRISPR, still testing every idea against a living organism.
AI-resistant 64 🧪The scientist who makes and measures matter itself — from Tapputi's Babylonian perfume still to today's robot laboratories, still the one who decides what the spectrum means.
AI-resistant 70 🔭The scientist who measures the universe, from Babylonian clay tablets to space telescopes, still deciding which flicker in the data is a discovery.
AI-resistant 70 🧮From Babylonian scribes to Fields medalists and AI-assisted proof: the profession that turns hard questions into permanent certainty, one theorem at a time.
AI-resistant 70 🧿Builds, measures and controls devices that exploit quantum states for computing, sensing, communication and materials research.
AI-resistant 78 🧫Uses clinical, trial and health-system data to generate reliable evidence for safer care, research and operational decisions.
AI-resistant 68