Goodfit

How to hire a Data Scientist

Data scientists build predictive models and statistical analyses that guide product and business decisions. They sit between engineering and the business. They turn raw data into findings people can act on. The role varies a lot between companies. Some data scientists are close to machine learning engineers. Others are statisticians. Others are mostly analysts who use Python instead of Excel.

Screen Data Scientist applicants
Data Scientist

Free · Email code to start · Uses your mic · Up to 3 minutes

On this page

What is a Data Scientist?

A Data Scientist uses statistics and machine learning to answer harder questions. What will happen? What caused what? What should we do differently? The work starts with turning a business problem into something you can measure. Then they gather and test the data, and choose and check a method. They are honest about how sure the result is. Finally, they explain it so a non-specialist can act on it. In most companies, modeling takes a small share of the time. The framing and the data take the rest.

Also called
Applied Scientist, Machine Learning Scientist, Quantitative Analyst
Function
Data science and machine learning
Level
Mid to senior individual contributor, often postgraduate-qualified
Reports to
Head of Data Science, Head of Data, or a product or risk function
Common stack
Python, SQL, pandas, scikit-learn, PyTorch; experimentation and notebook tooling
Next step
Senior or Staff Data Scientist, ML Engineer, or Head of Data Science

The hiring challenge

"Data scientist" means ten different jobs, depending on the company. Some teams need someone to train and test models that run in production. Others need someone to run experiments and explain the results. Others need someone to build dashboards with a bit of Python on top. The most common mistake is not deciding which type you need before you start interviewing. A candidate who is brilliant at experiment design may struggle to ship a model to production. The reverse is also true. Write down the first six months of work before you write the job description.

What to look for in a Data Scientist

Four traits matter. Statistical rigor: do they know when a result is significant and when it is not? Can they tell without a tool telling them? Problem framing: can they turn a vague business question into a hypothesis they can test? Explaining to non-technical people: can they explain a model's output to a business leader who has never heard of a confusion matrix? Practical sense: do they pick the simplest approach that works? Or do they build complex models when a basic one would do?

Also check for comfort with messy, real-world data. Real datasets often have errors in names, addresses, and phone numbers. Check that they know the tools your team uses, such as Python, R, SQL, and your cloud platform. And check that they can work with limited labeled data. Not every company has the large datasets that textbook models assume.

Common mistakes when hiring Data Scientists

Hiring a researcher when you need a practitioner. A candidate with a strong academic background may build elegant models that never reach production. If you need models in production, test practical engineering skills as well as statistics.

Not testing communication. A data scientist who cannot explain their findings to a product manager or a VP will produce work nobody acts on. Always include a round where the candidate explains something technical to a non-technical audience.

Putting too much weight on tools. A candidate who knows TensorFlow but not your framework can learn it. A candidate who cannot frame a problem well will not get better with a new tool. Test thinking first and tools second.

Key skills for a Data Scientist

  • Statistics fundamentals (hypothesis testing, distributions)
  • Python (pandas, scikit-learn, or equivalent)
  • Experiment design and A/B testing
  • Model evaluation and selection
  • Clear communication to non-technical audiences
  • Problem framing from vague business questions
  • SQL for data extraction
  • Pragmatic model selection

What a great interview looks like

Voice

"Explain precision and recall to a product manager who is not technical. When would you care more about one than the other?"

Coding

"Here is a dataset. Build a baseline model and explain why you chose your evaluation metric."

Scenario

"Your A/B test does not have enough users to show a clear result. What do you do?"

Voice

"Tell me about a model you built that changed a business decision. What was the question, and what did you recommend?"

Scenario

"A stakeholder asks you to "predict churn." What questions do you ask before you write any code?"

You can use these questions in Goodfit as they are, or edit the questions and the scoring for your team.

Recommended interview process

  1. 1

    Round 1: AI Voice Interview15 to 20 min

    The candidate frames a problem, walks through a past project, and explains their work. The scorecard covers statistical reasoning and clarity.

  2. 2

    Round 2: Technical Assessment60 min

    Analyze a dataset, build a model, and explain the choice of evaluation metric. Scored on the approach as well as accuracy.

  3. 3

    Round 3: Hiring Manager Interview45 min

    A business case discussion, how they work with others, and team fit. Only for candidates who passed Rounds 1-2.

In Goodfit, the AI interview runs the first round for every applicant. Skill tests, coding tests and roleplays can run in Goodfit too. Your team reads the scorecards and joins for the later conversations.

FAQ

Data Scientist hiring questions

What does a data scientist do?
They turn a vague business problem into a question data can answer. Then they answer it in a way they can defend. They explore and check the data and build features. They pick the model that fits the problem, which is often simpler than the fanciest option. They test it on a realistic holdout set. They say how sure they are and explain the result and its limits. Much of the job is designing experiments. That means deciding what to test, how long to run it, and whether a difference is real.
What is the difference between a data scientist and a data analyst?
An analyst explains what happened. A data scientist estimates what will happen or what caused what. Analysts mostly use queries, summaries, and charts. Scientists add statistical inference, experiment design, and machine learning. Companies use the titles loosely, so judge by the work. If the role means defending a claim about cause, or a model's accuracy, it is data science. That holds whatever the job post says.
What is the difference between a data scientist and a machine learning engineer?
A data scientist decides what to model and whether the model is sound. A machine learning engineer makes it run in production without breaking. That covers serving, speed, retraining, watching for drift, and rolling back. The skills overlap in the middle. At the edges they split into research design and systems engineering. Teams often expect one person to excel at both. That is the most common reason ML projects get stuck between a notebook and a live system.
What skills does a data scientist need?
They need a real grasp of statistics. That means sampling, confidence, multiple comparisons, and the gap between correlation and cause. They need Python and SQL. They need experiment design, since most business impact comes from tests more than models. They need careful feature engineering and validation. Above all, they must avoid leakage. Leakage is the most common reason a model looks great and then fails in production. They need curiosity about the business. And they need to communicate, because nobody deploys a model they do not understand.
Do data scientists need a PhD?
For most business roles, no. Many have a postgraduate degree because the statistics training is useful. For research roles, it is close to required. For applied work, other things predict success better. Can the candidate frame a business problem in statistical terms? Do they check their work honestly? Can they explain a result to a doubtful executive? Plenty of strong applied scientists come from engineering, economics, or physics.
What is data leakage and why does it matter so much?
Leakage is when the training data includes information the model would not have at prediction time. One example is a feature built from the outcome itself. Another is a random split that puts the same customer in both training and test data. The model then scores very well in testing and fails on live data. It matters because it is common, easy to miss, and costly. The failure shows up after people already trust the model. Asking how a candidate guards against it is one of the fastest ways to gauge real experience.
How is a data scientist's work measured?
By decisions and outcomes, more than by how many models they build. Fair measures include the business effect of experiments they designed. Another is how their live models perform against the baseline they replaced. Another is the share of their projects that reached production. Another is how well they set up measurement for others. Model accuracy alone is a poor metric. A very accurate model of the wrong target is worth nothing.
What interview questions should I ask a data scientist?
Give them a real business problem. Ask how they would frame it, what they would predict, and with what data. Ask how they would know it worked. Ask how they would detect leakage. Ask them to design an experiment for a proposed change. Include how long it would run and what would make them stop early. Ask about a model that did well offline and failed in production. Then ask them to explain a technical result to a non-technical person, live.
How do you screen data science candidates?
Algorithm trivia and take-homes that reward tuning are poor filters for this role. A structured conversation about framing and validation works better. Ask how they would approach a problem and how they would know they were wrong. Score every candidate on the same rubric. Add a short SQL and Python task to check the basics. What predicts success on the job is judgment about problems. Recalling methods matters much less.

Also hiring?

Data

Data Analyst

How to hire

Data

Data Engineer

How to hire

Engineering

Backend Engineer

How to hire

Looking for Data Scientist jobs instead? Browse open roles on Goodfit Jobs

Try Goodfit with your own team

Goodfit talks with applicants and employees by phone or in a browser. Try it on one of your open roles.

Book a demoStart free
  • 20 free assessments, no card needed
  • Then $5 per assessment
  • No seats or annual contract
Book a demo