Data · Interview questions

Data Scientist interview questions

These questions test whether a candidate can build predictive models that work in production, not just in a notebook. Each links to a first-year outcome: a model in use, measured lift over a baseline, monitoring, a well-run experiment and reproducible code.

The questions

Ask every candidate the same questions in the same order, and score each answer against the notes before you talk it over with the panel.

  1. Tell me about a model you built that is used in a real business process. Who uses it, and what do they do differently because of it?

    What it tests
    Model in production and in use
    A strong answer
    They describe the process the model feeds, such as a daily list of at-risk customers for the success team, and how they worked with engineers to deploy it. They know how often it is used.
    Watch for
    Their models stayed in notebooks or slide decks.
  2. How do you decide whether a model is good enough to release?

    What it tests
    Baseline and measured lift
    A strong answer
    They compare against a simple baseline, such as the value from last month or a rule the team already uses, on held-out data. They pick a metric tied to the business cost of errors, not just accuracy.
    Watch for
    They report a high accuracy score with no baseline.
  3. Here is a small customer data set. In 90 minutes, build a churn model, compare it with a baseline and tell me whether you would ship it.

    What it tests
    Modelling and honest evaluation (work sample)
    A strong answer
    They split data correctly, avoid leakage from future information, compare with a baseline and give a clear recommendation with limits. Their code is readable.
    Watch for
    Features that leak the outcome, producing results that look too good.
  4. Tell me about a time a model in production started performing worse. How did you find out and what did you do?

    What it tests
    Automated monitoring
    A strong answer
    They noticed through monitoring of inputs or accuracy, found the cause, such as a change in customer mix or a broken data feed, and retrained or fixed it. They added an alert afterwards if there was none.
    Watch for
    Users told them weeks later.
  5. An A/B test on a new checkout page shows a 4 percent lift after three days with 800 visitors per arm. The product manager wants to ship it. What do you say?

    What it tests
    Experiment analysis (scenario)
    A strong answer
    They check statistical significance and power, point out the short run time and possible weekday effects, and recommend running to the planned sample size before deciding.
    Watch for
    They agree to ship because the number is positive.
  6. Walk me through how you would design an experiment to test whether a new onboarding email increases activation.

    What it tests
    Experiment design
    A strong answer
    They define one primary metric, the minimum effect worth detecting, sample size, randomisation and run time before launching. They say what result would lead to go or no-go.
    Watch for
    They would decide the success metric after seeing the data.
  7. If you left tomorrow, how would a colleague retrain and redeploy your models?

    What it tests
    Version control and reproducibility
    A strong answer
    They describe code in Git, pinned environments, data versioning or fixed snapshots and a documented training script. A colleague has done it before.
    Watch for
    The steps live in their head or in notebooks run out of order.
  8. The head of sales asks you to predict which deals will close this quarter. What questions do you ask before you start?

    What it tests
    Framing the problem with the business (scenario)
    A strong answer
    They ask what decision the prediction supports, what data exists and how reliable the CRM records are, and what simple rule the team uses today. They check whether a model is needed at all.
    Watch for
    They start choosing algorithms before understanding the decision.

New to structured interviews? Read how to run structured interviews and how to build an interview scorecard.

Start from the job description

These questions follow the first-year outcomes in the matching template.

Data Scientist job description template

Similar roles

Your next hire, start to finish, in one place.

Free workspace. No card. Full scoring free for 7 days.