Example · Data Science and Statistics Student
Your statistics courses trained the skills that will last. AI now does much of the notebook work.
20 of your 40 hours still need you
- 12 stay yours
- 8 faster with AI
- 20 tools could take
A typical week at the middle pace. Your plan uses your own tasks.
Is data science still a good major with AI?
In the work a data science and statistics student is heading toward, 20 of a typical 40-hour week still need a person, by our task-by-task count. Tools could already take about 20 hours. Writing the query, cleaning the file and fitting the model are the parts of a first data job that a current model does well, and they are a large share of the week a new graduate is hired into. Your statistics courses trained the part that holds.
Read more
It includes deciding whether a design can answer a question, catching the error that looks fine, and saying out loud what a number cannot support.
You are not late, and your major is not a mistake. The New York Fed's data puts recent mathematics graduates at 5.8% unemployment, below computer science (New York Fed, The Labor Market for Recent College Graduates, February 2026), and BLS (the Bureau of Labor Statistics) projects data scientist jobs growing 35% over the next decade (BLS Employment Projections, August 2026). The door has narrowed for people who arrive with competition notebooks. It is wider for people who arrive having checked someone else's analysis and found what was wrong with it. The semesters you have left are for becoming that person before senior-year recruiting starts.
The hard truth
In statistics the bachelor's is often only the first step.
Many of the best jobs this major points to expect a master's degree. Plan for that now, so it is a choice you make on purpose.
Sources (2)
- BLS says statisticians typically need a master's degree, though some entry-level positions accept a bachelor's. BLS Occupational Outlook Handbook, Mathematicians and Statisticians, 2026-08
- 51% of working-age math graduates hold a graduate degree, as of 2024. Federal Reserve Bank of New York, Labor Market for Recent College Graduates, 2026-02
What stays yours
Asking how the sample was collected before you touch the data
Selection problems do not show up in the data file. A model reads the columns, but it does not know the survey only reached people with landlines. It is the cheapest way to prevent the most expensive errors.
Catching a model's confident wrong answer
As analysis gets cheap to produce, the person who can tell whether it is right becomes the constraint. Most students have done this at least once on a homework problem. The plan turns it into a log you can show someone.
Knowing a specific dataset's quirks: the wave with the coding change, the site that entered dates backward
No model has the undocumented knowledge about a dataset. Every lab and every company has it, and the person who holds it gets asked first.
Working inside a study with a protocol, an IRB approval (the research ethics board) and a PI who signs
In regulated analysis a named, qualified person answers for the statistics. You are not that person yet, but seeing how that work is checked puts you ahead of most applicants.
Telling a non-statistician that two groups are not actually different
Trust is built in person. Many students find this part hard, and the plan practices it early.
Which jobs fit data science majors with AI?
Plan A is your best move in the field you are studying for. Most sessions in the 13-week plan build toward it.
- A
Aim for health research and clinical statistics jobs
Why it fits
Why you. If you have worked inside a study with a protocol, a PI (the lead researcher) and an IRB (the board that approves research on people), or you like being sure more than being fast, this is your door. Most applicants from your major have never seen how regulated analysis is checked.
The gap. SAS, the statistics software used in drug trials, is still common in clinical work, and many students have only used R or Python. BLS says statisticians "typically need a master's degree," though some entry roles accept a bachelor's. Nothing on your resume yet says "regulated."
The trade-off. Starting pay is often below tech, so compare offers. Drug company and contract research hiring runs on its own cycle, and academic and federal health research depends on grant funding, which has been unsettled. You still have to decide about the master's later.
Other moves that fit
These are also good bets in this field. Read them if Plan A doesn't suit you.
Take the actuarial exam route
Why it fits
Why you. If probability and mathematical statistics are behind you, so is most of Exam P's content. Actuarial work rewards people who would rather be right than quick.
The gap. You have not passed any exams yet. Many employers want one or two exams before an internship offer, so check the postings you like.
The trade-off. There are about 1,500 openings a year nationally, a much smaller door than data science. Exams cost $275 each and take evenings for years. The entry analyst's spreadsheet work is as exposed as any analyst's. The credential is what holds.
Aim for data cleanup and reporting jobs in one field
Why it fits
Why you. If reconciling messy files and tracking down why two numbers disagree is the part of research you secretly enjoy, this is that job.
The gap. Version control, a transformation tool such as dbt, and one project where you defined and tested a metric for someone else.
The trade-off. It follows the tech hiring cycle, and some of its scaffolding work is where coding agents improve fastest. Treat it as a backup option.
What's a plan B for statistics students?
Plan B is a job AI is creating that these skills already fit. One session a week in the plan trains you for it, so it is ready if your job shrinks or you want a change.
BPlan B, picked for this role
AI evaluation specialist
Catching a model's confident wrong answer is the core of evaluation work, and your statistics courses trained you to look for it.
Other options
LLM evaluation data scientist
Start in an analyst seat and own evaluation for one AI feature, since asking how the sample was collected is half that job.
AI governance analyst
Working inside a study with a protocol and IRB approval prepares you to document how AI systems use data responsibly.
In your own plan you can switch to any of the 47 roles.
The first two weeks
The first 2 of 13 weeks. Most weeks have two 45-minute sessions and one 90-minute session, plus one 40-minute Plan B session (week 4's runs 45 minutes). Weeks 9 and 10 are finals and drop to one 90-minute session plus Plan B, and weeks 11 to 13 use winter break. The plan finishes before the January deadlines many summer programs use. By week 13 you'll have a reanalysis of a published dataset with a methods memo in a public repository, an error log of AI analysis mistakes you caught, and about 15 summer applications sent.
- Week 1: See where your time goes
What you do
- AList a typical week of your classes, research, projects and any other work. Next to each item, note which parts AI already helps with, using the fourteen tasks on this page as a guide.
- APick one public dataset that has a published analysis, such as a CDC survey or a journal's replication files. Save the citation.
- AAsk an AI model to reproduce the published main result. Don't fix anything yet.
- BTake Anthropic's free AI Fluency: Framework and Foundations course. Use the dataset you chose this week in every exercise.
- Week 2: Start your error log
What you do
- AStart a file called errors.md. Log one line per AI mistake, tagged wrong test, leakage (test data seeping into training), ignored design, wrong denominator or overclaim.
- ARead the dataset's notes on how the sample was collected. List what the AI ignored.
- ARead the causal diagram chapters of The Effect, a free causal inference textbook. Draw the diagram for the dataset's main claim.
- BTake Anthropic's AI Capabilities and Limitations course. Note which failure types could show up in data analysis.
Three courses from the plan
Checked 2026-10-13.
- The Effect: An Introduction to Research Design and CausalityNick Huntington-Klein · Free online ("will continue to be free")
Yes, Weeks 2 and 4.
- Causal Inference: The Mixtape (online edition, "The Remix")Scott Cunningham · Free to read online; print from Yale University Press
Later, alongside a causal inference course.
- An Introduction to Statistical Learning with PythonJames, Witten, Hastie, Tibshirani, Taylor · Free PDF
Use it as a reference for model fitting and validation. Keep it out of your portfolio.
Risks beyond AI: moderatePublic-sector and visa risk
Government has long been a steady employer of statisticians, and the federal statistical agencies have lost a large share of their staff since January 2025. If you are an international student, a proposed OPT fee could change what an early job in the US costs your school or employer. Both risks are separate from AI, and neither changes the growth projections for the occupations themselves.
Where this could be wrong
- The destination blend is ours. If most graduates from your program go into product analytics at tech companies, the automatable share rises; if most go into health and government, it falls. Your department's placement data would correct it.
- Mathematics stands in for statistics in the New York Fed data. Statistics and data science majors may do better or worse than that row suggests.
- We scored regulated analysis as holding up because of who signs. If regulators accept AI-assisted outputs with lighter human review, AI takes on part of that work, and the first move becomes about the master's and the signature rather than the programming.
- Federal statistical agencies may rehire. The 2025–26 losses are documented; whether they reverse is a budget question we cannot forecast.
Also matched: data science student, data science major, statistics student. Research checked 2026-10-13. Page updated .