Good fit if
- You like turning vague questions into testable ones.
- You can be patient with bad data, definitions, and revisions.
- You can explain a caveat without sounding defensive.
- You would rather be accurate than impressive.
Career decision guide
Data Careers Decision Guide
This page helps you choose between data analyst, data scientist, analytics engineer, and ML-adjacent paths. The decision is not "do you like data?" It is whether you want to make business questions measurable, test uncertainty with statistics and models, build the trusted data layer underneath both, or move closer to software systems. The rewarding part is finding the truth in messy evidence. The hard part is that the data is often broken, the question is political, and AI can now produce shallow answers faster than humans can verify them.
Career Dish uses O*NET and BLS data as the skeleton, then translates the signals into a decision guide: what the work feels like, what kind of stress it creates, what the path costs, and what should make you pause before committing.
$120KData scientist median
$199KTop 10% signal
34%BLS growth
66/100AI exposure
Verdict
Should you go into data?
Yes if you like the awkward middle between numbers and decisions: defining the metric, finding the broken join, deciding whether a model is useful, and explaining uncertainty to people who want a simple answer. Think twice if the appeal is mostly remote work, dashboards, AI hype, or the idea that data will let you avoid people. Data careers are quieter than sales, but they are not socially empty. The work becomes valuable when it changes a human decision.
Think twice if
- You want clean tutorials to resemble the job.
- You hate when stakeholders challenge your numbers.
- You expect AI to do the hard thinking for you.
- You want high pay without building technical or domain depth.
Before you commit
- Build one messy-data project with a decision memo.
- Compare analyst, scientist, analytics engineer, and ML engineer postings.
- Ask working data people how often their analysis changes decisions.
- Price the path against the first role, not the dream title.
Decision scorecard
This scorecard shows a high-upside career family with strong analytical load, meaningful pay potential, and heavy AI assistance, but the value shifts away from routine output and toward evidence judgment: definitions, data quality, causal thinking, stakeholder trust, and model limits.
FitEvidence
Best if messy numbers make you curious
The work fits when a broken metric, weird outlier, or stakeholder claim makes you want to investigate rather than escape.
PathVariable
Analyst can be shorter; scientist is heavier
Analyst roles can open with SQL, BI, and business context. Scientist roles often need stronger statistics, Python, modeling proof, or graduate-level signal.
Pay$120K
The data scientist median is strong
BLS OEWS May 2025 puts data scientists around $120K median pay, but analyst pay depends heavily on lane, industry, and decision ownership.
AI66/100
Routine output is exposed
AI can draft SQL, charts, summaries, and notebooks. The durable value is knowing whether the output should be trusted.
Choose the right data lane
Do not buy a generic "data career" story. The daily work changes sharply by lane. For the direct analyst/scientist breakdown, use the dedicated data analyst vs data scientist comparison.
Data analyst
Business questions, SQL, metrics, dashboards, cleanup, stakeholder explanation, and deciding what the number can honestly support.
Entry path72/100
Data scientist
Statistics, modeling, experiments, prediction, evaluation, uncertainty, and deciding whether a model should exist at all.
Technical bar86/100
Analytics engineer
Modeled data, transformation layers, metric definitions, warehouse quality, reusable tables, and self-serve analytics foundations.
Systems depth82/100
ML engineer
Production models, pipelines, APIs, monitoring, performance, deployment, and the software layer around machine learning.
Engineering load88/100
Is data work stressful?
The stress is rarely "I cannot write the query." It is that the query returns a number nobody wants, or three systems disagree, or the model looks good until you find leakage, or the analysis becomes a political object. This is a good career if that makes you more careful. It is a bad career if that makes you resent the people asking questions.
The metric is disputed
Revenue, churn, active user, conversion, risk, quality, or success may mean different things to different teams.
The data is not built for the question
Fields are missing, definitions changed, labels are weak, and the system was designed for operations rather than analysis.
The stakeholder wants certainty
The business may want a clean yes/no when the honest answer is a range, caveat, test, or warning.
AI raises the proof bar
When tools can generate a plausible answer quickly, human credibility depends on validation, explanation, and judgment.
Typical day
The common loop is frame the decision, inspect the data, build the evidence, explain the uncertainty, and handle the follow-up.
FrameDefine the decisionWhat is being decided, who will use the answer, and what would change their mind?
InspectFind the data shapeTables, joins, missingness, metric definitions, historical changes, and source trust.
BuildQuery, model, or dashboardSQL, cleaning, exploratory analysis, baselines, charts, validation, and documentation.
ExplainMake it usableMemo, deck, caveats, recommendation, and a plain-language decision path.
Follow-upKeep it honestRevise assumptions, answer objections, monitor drift, and document the next version.
How hard is the path?
1
Start with the lane
Analyst, data scientist, analytics engineer, and ML engineer require different proof. Pick the first role you can plausibly win, not the title with the best salary screenshot.
2
Build the core stack
Analysts need SQL, spreadsheets, BI, data cleaning, metric definitions, and written business analysis. Data scientists add statistics, Python, modeling, experiments, and evaluation.
3
Create one serious project
The project should show the question, source, assumptions, cleaning, analysis, caveats, recommendation, and what you would do next. A pretty chart is not enough.
4
Use domain leverage
Healthcare, finance, education, logistics, operations, retail, marketing, science, and support experience can become an edge if it helps you ask better questions.
5
Validate the first-role market
Compare your proof to real postings before paying for a bootcamp, master's, or certificate. The market is less impressed by shallow AI-assisted portfolios than it used to be.
Pay and ROI
Data scientist pay is strong in the national data: $120K median and $199K near the top 10%. Data analyst pay is more title-dependent: a reporting analyst, product analyst, BI analyst, and analytics engineer may all be called "data analyst" while living in different salary ladders. The ROI question is whether your training buys a real lane, not whether the word data appears in the title.
Good ROILow-cost training, strong SQL, real business project, domain edge, and a first analyst or analytics engineering target.
Risky ROIExpensive credential, weak proof, generic notebooks, no feedback, and no plan for the first interview loop.
AI risk assessment
Data Scientist has moderate exposure: AI can draft SQL, summarize tables, generate notebook code, build first-pass charts, and propose models, but durable value sits in data quality judgment, causal skepticism, experiment design, business context, deployment, and explaining uncertainty honestly.
More exposed
- SQL drafts, spreadsheet cleanup, notebook scaffolding, chart descriptions, and first-pass dashboard summaries.
- Model suggestions, feature ideas, evaluation boilerplate, statistical explanations, and stakeholder-ready narrative drafts.
- Routine anomaly explanations, report refreshes, documentation, and code translation between Python, R, and BI tools.
More protected
- Knowing whether the data can answer the question, or whether the question is flawed.
- Detecting bad definitions, leakage, missing context, selection bias, and metrics that would mislead the business.
- Explaining uncertainty, tradeoffs, model limits, and decision consequences to people who want a simple number.
The practical AI answer is this: do not compete with AI on first-pass output. Compete on asking the right question, auditing the answer, and knowing what a decision-maker should do with the evidence.
Who should avoid this?
You need clean instructions
Data work often starts with unclear questions and imperfect records. The job is partly making the ambiguity visible.
You hate being challenged
Your number, chart, model, or conclusion may be questioned by people with power, incentives, and incomplete context.
You want tools to be the product
SQL, Python, BI, and AI are instruments. The product is a decision people can trust.
You dislike maintenance
Dashboards, metrics, pipelines, notebooks, and models need upkeep after the exciting first version.
Best alternatives
Business analyst
Better if process, requirements, operations, and stakeholder translation appeal more than SQL-heavy data work.
Analytics engineer
Better if the data modeling, warehouse, dbt, transformation, and metric layer is the satisfying part.
Software developer
Better if you want to build product systems rather than analyze evidence for decisions.
Product manager
Better if the user, roadmap, and decision ownership appeal more than analysis execution.
Actuary or statistician
Better if formal quantitative modeling, risk, and regulated statistical work appeal.
Market research analyst
Better if surveys, customer research, segmentation, and market evidence appeal more than internal data systems.
Deep dives for this career
Use these when you want the narrower answer: what data analysis is actually like, what data science is actually like, how stress and salary differ, whether the switch works at 40, how AI changes entry-level data work, or whether analyst, scientist, analytics engineer, or ML engineer fits better.
Analyst RealityWhat Data Analysis Is Actually Like
The real texture of data analysis: business questions, SQL, broken definitions, dashboards, stakeholder pressure, and evidence.
Analyst StressIs Data Analysis Stressful?A stress map for data analysis: metric ambiguity, dashboard debt, ad-hoc requests, ignored analysis, and AI pressure.
Analyst PayData Analyst Salary RealitySalary reality for data analysts: title spread, BI, product analytics, analytics engineering, first-role proof, and ROI.
Analyst DayDay in the Life of a Data AnalystA day in the life of a data analyst: requests, definitions, SQL, cleanup, dashboards, explanation, and follow-up.
Analyst SwitchCareer Change to Data Analyst at 40A sober career-change guide for data analysis at 40: SQL, BI, proof, business context, AI tools, and first-role strategy.
Science RealityWhat Data Science Is Actually LikeThe real texture of data science: uncertainty, statistics, models, evaluation, weak labels, stakeholder hype, and deployment.
Science StressIs Data Science Stressful?A stress map for data science: model expectations, data leakage, evaluation, stakeholder hype, weak foundations, and AI pressure.
Science PayData Science Salary RealitySalary reality for data scientists: BLS median pay, title inflation, ML engineering overlap, specialization, and ROI.
Science DayDay in the Life of a Data ScientistA day in the life of a data scientist: problem framing, SQL, data cleaning, modeling, evaluation, explanation, and handoff.
Science SwitchCareer Change to Data Scientist at 40A sober career-change guide for data science at 40: statistics, Python, degree filters, portfolio proof, domain leverage, and first-role strategy.
Analyst vs ScientistData Analyst vs Data ScientistCompare data analyst, data scientist, analytics engineer, BI analyst, and machine learning engineer paths.
AI RiskWill AI Replace Data Analysts?A realistic AI-risk guide: what AI automates, what it assists, why junior data work is pressured, and which skills stay durable.
Avery interview: what data work feels like
Avery is an invented guide, not a quoted source. Read this as a practical walkthrough of the situations the role tends to create: the vague request, broken dashboard, model that should not ship, AI-assisted query, stakeholder pressure, salary ladder, and analyst-versus-scientist decision.
Guide profileAvery, analytics lead who has worked analyst, data science, and data platform projects
This section turns the interview-style material into scannable answers so the useful texture is visible without making the reader operate a chat interface.
What do beginners misunderstand?
AveryThey think the job starts when the dataset opens. It usually starts earlier: what decision are we making, what does this metric mean, who logged the event, and what would make this answer dangerous?
What is the stressful part?
AveryBeing the person who knows the number is not as clean as the meeting wants it to be. Sometimes the hard thing is saying, "I can give you a number, but I cannot let you treat it like proof."
How has AI changed the work?
AveryIt makes drafts faster and bad confidence cheaper. I use it constantly, but I trust it the way I trust an intern with no company context: useful, fast, and absolutely in need of review.
Sources and methodology
O*NET Database 30.3Occupation descriptions, alternate titles, work context, work activities, and education signals.
BLS OEWS May 2025National wage estimates, percentile pay, mean pay, and employment estimates by SOC group.
BLS Employment Projections2024 to 2034 projected employment, growth, annual openings, entry education, experience, and training.
BLS OOH profileOfficial Occupational Outlook Handbook context for the matched career family.
Career Dish adds fit scores, workload metrics, AI exposure estimates, and interview-style guide scenes on top of public datasets. Those interpretive layers are meant to make the data scannable, not to replace official licensing or school-specific research.
Career decision FAQ
Should I become a data analyst or data scientist?
Choose data analysis if you want to make business questions measurable with SQL, dashboards, metrics, and stakeholder explanation. Choose data science if you want deeper statistics, modeling, experiments, prediction, and uncertainty work. Consider analytics engineering if the data-quality layer is the part you like most.
Is data analysis still a good career with AI?
It can be, but routine SQL, dashboard summaries, and first-pass reporting are more exposed. The durable analyst understands metric definitions, data quality, business context, and how to explain what the data can and cannot prove.
Is data science still worth it?
Data science can be worth it when the path builds statistics, SQL, Python, model evaluation, domain context, and a credible route into real modeling or decision work. It is weaker when the title is inflated or the training produces shallow notebooks without judgment.
Will AI replace data analysts?
AI will automate or speed up some routine data tasks, but it does not replace judgment about definitions, bad data, causality, stakeholder incentives, model limits, and decision consequences.
What is the best first step for a career changer?
Start with one messy, real project: define a business question, pull or clean data, write assumptions, create a simple analysis, and produce a decision memo. That proves more than a polished dashboard alone.