Hire a data scientist who can say what the model was for
To hire a data scientist, decide first whether you need experimentation, machine learning or forecasting, then source from people who shipped that work and test how they frame a business question. BLS projects 35 percent growth for data scientists from 2025 to 2035, with a 2025 median of $120,230.
Which kind of data scientist do you need?
4 different jobs share this title. A search that does not pick one returns a mix, and an interview that does not pick one tests the wrong skill.
| Kind | Day to day | Test for |
|---|---|---|
| Product and experimentation | A/B tests, metrics, funnel analysis with product managers | Power, stopping rules, metric choice |
| Machine learning modeling | Churn, fraud, ranking or pricing models that run in production | Baselines, leakage, evaluation tied to cost |
| Forecasting and decision science | Demand, capacity and finance forecasts that drive plans | Uncertainty, backtesting, explaining error bands |
| Research or applied science | New methods, often with a PhD and publications | Depth in one method and judgment on when not to use it |
If the work is mostly pipelines and model serving, you may want an ML engineer or a data engineer instead.
Where to find data scientists
Data scientists leave more public work than most engineers: Kaggle notebooks and competition write-ups, GitHub repositories, talks at PyData and similar meetups, blog posts on experiments, and papers on arXiv for the research end. Read those for problem framing, not polish. A clean notebook on a toy dataset tells you little; a write-up that explains why a metric was chosen tells you a lot.
Then widen the titles. The same work is done by people called applied scientist, decision scientist, quantitative analyst, product analyst, statistician and machine learning engineer. Senior analysts who already run experiments are worth a look for product roles, since the core skill is the same and the title is the only gap.
Search strings for a data scientist
Pair the method with the business context. "Python" and "machine learning" alone return every course graduate in the market.
BOOLEAN
("data scientist" OR "product data scientist" OR "decision scientist" OR "product analyst" OR "applied scientist")
AND ("A/B testing" OR experimentation OR "causal inference")
AND (SQL AND (Python OR R))
AND (product OR marketplace OR subscription)
NOT (intern OR student OR recruiter)
GOOGLE X-RAY
site:kaggle.com "notebook" "churn" "lightgbm"
site:github.com "A/B test" "power analysis" notebook
PLAIN-ENGLISH BRIEF
Data scientist with 3+ years running A/B tests on a consumer or subscription product, strong in SQL and Python, who has chosen metrics with product managers and presented results that changed a roadmap. Hybrid, London.
What does it cost to hire a data scientist in 2026?
BLS publishes a dedicated data scientist line, which few tech roles get. These are the national figures.
| Measure | Figure | What it means for you |
|---|---|---|
| Median pay (2025) | $120,230 a year, $57.80 an hour | Half of US data scientists earn more |
| Highest-paying industry listed | Publishing, broadcasting and content providers, $142,240 | Followed by computer systems design at $132,380 |
| Jobs in 2025 | 275,600 | A pool about a seventh the size of software development |
| Projected growth, 2025 to 2035 | 35 percent | Much faster than average, so competition for experience will rise |
| Openings per year | About 24,800 | Replacements included |
| Typical entry education | Bachelor's degree | A PhD is a must-have only for research roles |
Source: BLS Occupational Outlook Handbook, Data Scientists, read 5 October 2026.
From a written brief to people who match it
The brief is read into filters you can see and edit, then searched across ~900M public profiles.
Write the kind of data scientist you need in plain English. The Cognitive judges each person against your must-haves; anyone who misses one, such as experimentation, drops below the people who meet it, and nobody is removed. Each card opens with a "Why this match" line traced to the profile.
What to assess in a data scientist, and what good sounds like
Credentials predict less than you'd think. Ask about a decision their work changed, then test the reasoning underneath it.
Tell me about an experiment you designed. How did you decide how long to run it?
A power calculation behind the sample size and a guard against peeking or a novelty effect before calling the result.
What minimum effect were you trying to detect, and what if the result landed just short of significance?
Why did you pick that model for that problem?
A reason tied to the data, such as gradient boosting in XGBoost or LightGBM over a neural net for tabular churn, and the simple baseline it had to beat.
Tell me about a bug in your features that you caught late.
A leakage case, such as a feature computed after the label date, and how much it had inflated offline AUC.
How did you find it, and what check stops it now?
Which metric did you optimize, and why that one?
A metric fitted to the real constraint, for example precision at fixed recall because the fraud team can review only 200 cases a day.
Describe presenting a null or negative result to a product leader.
How they framed it, what they recommended and the decision that followed.
On take-homes
A long take-home costs the strongest candidates a weekend they won't give you. Use a short, scoped dataset exercise, or ask them to walk through a past analysis and defend its choices live.
A data scientist hiring process in 5 steps
- Pick the kind and the decision Name which of the 4 kinds you need and one decision this person's work should change in the first 6 months. "Decide whether the annual plan discount is worth keeping, with a test we trust."
- Source from public work and wider titles Search profiles, Kaggle and GitHub, and include product analysts, decision scientists and applied scientists in the title list.
- Structured first interview Experiments, modeling, data quality, evaluation and communication on one scorecard, each grounded in their own past project.
- Case or analysis walkthrough A short dataset exercise or a live walkthrough of a past analysis with your lead data scientist. Score reasoning, not the chart.
- Meet the stakeholder A product or finance lead they would serve. Ask the candidate to explain one finding to them in plain words, then decide within a week.
How The Cognitive fits a data scientist hire
The report: a 1 to 5 score on each criterion, overall written feedback, a transcript and a recording.
After sourcing, candidates take a live, two-way AI video interview of 10 or 20 minutes at a slot they book. The rubric is fixed for the role and follow-ups adapt to each answer, so a vague experiment story gets the question about minimum detectable effect.
More for hiring data scientists
Data scientist job description · Data scientist interview questions · AI interviewer for data scientists · Hire AI and ML engineers · Hire data engineers · AI sourcing tool
Frequently asked questions
How much does it cost to hire a data scientist?
BLS put the 2025 median for US data scientists at $120,230 a year, with publishing and content companies paying a median of $142,240. Add sourcing tools, interview time and any agency fee. Research roles that require a PhD sit higher and take longer to fill.
Where can I find data scientists to hire?
Kaggle, GitHub, PyData talks, experiment write-ups and arXiv papers show public work. Profile search finds the many who publish nothing. The Cognitive searches ~900M public profiles from a plain-English brief, and shows tenure per role and an "Open to work" badge, so you can see who might move.
How long does it take to hire a data scientist?
No reliable public figure exists for this role alone. The usual delays are an unclear kind of data scientist, long take-homes that strong candidates skip and slow scheduling with busy stakeholders. Picking the kind first and replacing the take-home with a walkthrough removes most of it.
Do data scientists need a PhD?
BLS lists a bachelor's degree as the typical entry education. A PhD matters for research and new-method roles. For product experimentation or applied modeling, shipped work and clear reasoning predict more than the degree.
What is the difference between a data scientist and a machine learning engineer?
A data scientist decides what to model and whether it worked: framing, features, evaluation, communication. A machine learning engineer makes models run reliably in production: serving, pipelines, monitoring. Many teams need both, and the interview should test the one you are hiring.
Can an AI interview assess a data scientist?
It can test reasoning: how they chose a sample size, why they picked a model, how they caught leakage. The Cognitive's live, two-way AI video interview adapts each question to the last answer and scores every criterion 1 to 5. Keep a short analysis walkthrough with your own data lead for finalists.
What does The Cognitive cost?
AI Sourcing and AI Interview each start at $99/month. Plans are monthly and cancel anytime. Each interview plan carries a monthly allowance, and past it you buy an interview top-up.
Explore: AI interviewer for data scientists · Data scientist interview questions · Data scientist job description · Hire analytics engineers · Pricing