AI can assist coding, exploration and written analysis. Reliable data science depends on defining a useful decision, understanding the data-generating process and evaluating results without misleading shortcuts.
Forecasting, experimentation and machine learning applications differ in their failure modes. A model that performs well on a convenient dataset may fail when the population or operating conditions change.
Compare the duties below with a normal week in your own role. Time
spent, responsibility, employer tools and access to reliable data can
change the picture substantially. Use the personal task audit to
identify those differences.
Task-by-task exposure
We selected 8 of 15 O*NET core
tasks by importance. Importance is an O*NET rating on a 1–5 scale; it
is not the percentage of time spent. The category and explanation are
our interpretation, not an O*NET assessment of AI.
0 routine automation 6 ai-assisted 2 human-led
Reading uncertainty: changing one task by one
category step would put this index at approximately 31–44/100. This is a
sensitivity example, not a statistical confidence interval. Small
score differences are not a sound reason to change careers.
Selected core tasks and our interpretation
Task and source detail
Assessment and reason
Analyze, manipulate, or process large sets of data using statistical software.
O*NET task 21823 · Importance 4.33/5
Task source: 08/2026 · Rating: 08/2026
AI-assisted
Analysis code can be drafted and adapted. The scientist must establish whether the data and statistical choices answer the actual question.
Create graphs, charts, or other visualizations to convey the results of data analysis using specialized software.
O*NET task 21828 · Importance 4.33/5
Task source: 08/2026 · Rating: 08/2026
AI-assisted
Chart production is assistable once inputs are defined. Scales, denominators, missing values and uncertainty determine whether the visualization communicates an honest result.
Test, validate, and reformulate models to ensure accurate prediction of outcomes of interest.
O*NET task 21837 · Importance 4.25/5
Task source: 08/2026 · Rating: 08/2026
AI-assisted
Tools can help implement validation. Choosing independent evaluation data and detecting leakage or instability remain essential scientific decisions.
Deliver oral or written presentations of the results of mathematical modeling and data analysis to management or other end users.
O*NET task 21829 · Importance 4.21/5
Task source: 08/2026 · Rating: 08/2026
AI-assisted
A presentation can be drafted from results. Explaining limitations and answering stakeholder questions require understanding what the analysis does and does not show.
A recommendation connects evidence to consequences for the organization. Predictive performance alone cannot decide acceptable trade-offs or whether an intervention is appropriate.
Identify business problems or management objectives that can be addressed through data analysis.
O*NET task 21831 · Importance 4.13/5
Task source: 08/2026 · Rating: 08/2026
Human-led
Problem definition establishes the objective and what success would mean. A model cannot compensate for an irrelevant target or an unobservable business outcome.
Compare models using statistical performance metrics, such as loss functions or proportion of explained variance.
O*NET task 21827 · Importance 4.04/5
Task source: 08/2026 · Rating: 08/2026
AI-assisted
Metric calculation can be automated. Choosing an evaluation design and interpreting a small difference require attention to uncertainty, data splits and operational usefulness.
Routine automation includes conventional configured software; it does
not imply autonomous AI or adoption by every employer. Human-led
allows supporting tools. Vendor links document a capability, not a
validated rating of this complete task. Older task dates are shown
even though the database release is August 2026.
Task wording and importance: O*NET occupation 15-2051.00,
O*NET 31.0 Database
, U.S. Department of Labor, Employment and Training Administration, CC BY 4.0.
Selected and adapted by Career Risk Score; USDOL/ETA has not approved,
endorsed or tested these changes.
What assistance looks like in practice
Hypothetical workflow to illustrate the boundary; not a reported case
study.
A team builds a churn model. An assistant proposes features and training code. The scientist checks whether any feature contains information that would be unavailable at prediction time and whether the evaluation matches the intended deployment.
What must be checked
Audit leakage, use an appropriate holdout and compare against a simple baseline. Examine subgroup performance and whether acting on the prediction would actually help the business decision.
Practical skills to strengthen
Publish a reproducible analysis with data limitations and a baseline.
Practice explaining uncertainty to someone choosing an action.
Monitor a model or analysis for changes in input quality over time.
Changes worth watching
Watch whether employers emphasize production monitoring and domain knowledge.
Track whether automated model building shifts effort toward evaluation and data quality.
There is no supported date when this occupation becomes “safe” or
disappears. Revisit these observations as your tasks and tools change.
Tools behind these capability examples
These are relevant documented capabilities, not endorsements,
adoption statistics or hands-on product reviews. Features depend on
the product, plan and employer configuration. References checked
September 2026.
Copilot in ExcelSpreadsheet assistance. A plausible formula is not evidence that its assumptions fit the decision.
Copilot in PowerPointPresentation drafting; instructional and factual quality still need evaluation.
Pay, demand and entry context
BLS reports median annual pay of $120,230
in 2025, with 24,800 projected
openings per year on average during 2025–35. Openings include
replacement needs as well as growth; they are not a count of currently
advertised vacancies.
The projected employment change is + 34.6%
over ten years. Employment growth can coexist with automation of particular tasks. It does not make every worker or location equally likely to benefit.
Typical entry education
Bachelor's degree
Related work experience
None
Typical on-the-job training
None
Operations research adds explicit optimization and constraints. Market research introduces sampling and customer research methods that are not guaranteed by machine learning experience.
BLS categories describe typical entry, not a complete qualification
checklist. “None” does not mean no learning is needed. National
medians are not starting salaries; location, sector, experience, hours
and benefits matter. Wage data exclude self-employed earnings. Consult
local job descriptions and relevant credential authorities before
paying for training.
These links identify transferable work and a concrete gap to explore.
They are not guaranteed pathways, “AI-proof” jobs or recommendations
based on a small score difference.
Formulate an operational decision with an objective and constraints, then test feasibility.
BLS typical entry: Bachelor's degree; related experience: None.
9 shared skills among each role's top
12
Mathematics, Critical Thinking, Reading Comprehension, Complex Problem Solving, Speaking, Active Learning, Active Listening, Writing, Judgment and Decision Making.
These broad O*NET skills are selected by importance, with
ties broken by skill ID. Overlap does not measure
proficiency or hiring readiness.
Design a customer study that distinguishes the sample from the population you want to understand.
BLS typical entry: Bachelor's degree; related experience: None.
10 shared skills among each role's top
12
Mathematics, Critical Thinking, Reading Comprehension, Complex Problem Solving, Speaking, Active Learning, Active Listening, Writing, Judgment and Decision Making, Monitoring.
These broad O*NET skills are selected by importance, with
ties broken by skill ID. Overlap does not measure
proficiency or hiring readiness.
Use your own plausible pay figures to explore a move. Both pay fields start
at this occupation's national median so the calculator does not imply a
salary increase. A destination median describes existing workers, not your
likely starting offer.
A fixed-pay illustration in nominal dollars. Unpaid months occur at the
start of preparation; costs are charged at the start. Transition must fit
within the comparison period. Excludes taxes, benefits, raises, inflation,
investment returns and the chance of getting a job. A negative result is
possible. No inputs are saved or sent by this calculator.
Calculation and a worked example
Stay = current pay × years. Move = preparation pay × (preparation years −
unpaid months ÷ 12) + new pay × (years − preparation years) − one-time
cost.
For $60,000 current pay, $70,000 new pay, $50,000 preparation pay, two
years preparing, three unpaid months, $8,000 costs and a ten-year
comparison: stay = $600,000; move = $639,500. The $39,500 difference is a
scenario result, not a forecast. Break-even checks cumulative totals each
month under these same assumptions.
Check the evidence and limits
The index weights these sampled tasks equally. It does not measure
time, adoption, cost savings, unemployment or individual ability. The
examples and transition exercises are editorial inferences. Product
documentation supports the narrower capability described, not the
claim that a whole job can be replaced.
