Interview questions
Data Scientist interview questions
Data Scientist interview questions for each round: modelling, experiments and statistics, with what the hiring manager is scoring.
Answer one Data Scientist question out loud today and time yourself.
How Data Scientist interviews are scored
Data scientist interviews score whether you pick the simplest method that answers the question. Explaining why a baseline was not good enough carries more weight than naming an advanced algorithm.
What hiring managers check in Data Scientist interviews
- Whether you start from a simple baseline, because a model that barely beats it is not worth maintaining.
- Whether you can say what changed in the business because of your model, not only its accuracy.
- Whether you read an experiment honestly, which the A/B test task exposes when the result is close.
Pick the check you are weakest on and prepare one Data Scientist story for it tonight.
Questions Data Scientist candidates get
On top of the questions every role in this field is asked, these come up for this title in particular.
Screening interview questions
Usually a recruiter or talent partner, 15 to 30 minutes, often by phone or video.
- Which of your models has been used in production, and by whom?
- How much of your work is experimentation versus predictive modelling?
- Do you deploy your own models, or hand them to an engineering team?
- What was the business outcome of the last model you shipped?
Hiring manager interview questions
The person you would report to, 45 to 60 minutes, sometimes with a team member.
- Tell me about a model that performed well offline but failed in production.
- How do you explain a p-value or confidence interval to a product manager?
- What baseline did you compare your last model against, and by how much did it win?
- Tell me about an experiment where you stopped early or changed the plan midway.
Final round questions
Senior leaders, a panel, or a case task, 60 minutes to half a day.
- Walk us through how you would predict which customers will cancel next month.
- Here are the results of an A/B test: would you ship the change, and why?
- Our model's accuracy dropped this month. How would you find out why?
- How would you size the sample for a test on a page with 2,000 visitors a week?
Plus the data questions every round asks
- Walk me through your data background in two minutes.
- Which tools do you use most: SQL, Python, R or a BI tool?
- What kind of data and business questions do you work on now?
- Tell me about an analysis that changed a business decision.
- Tell me about a time the data was messy or wrong and what you did.
- How do you explain a technical result to someone non-technical?
- Sign-ups dropped 15 per cent last month: walk us through how you would investigate.
- How would you design an experiment to test a new pricing page?
- How would you set up a reporting layer for a team with no data culture?
Get the data question pack by email
Every question on this page in one email, grouped by round. Plus three things that are not on the page:
- 12 more questions, 4 per round, that only go out by email
- What a strong answer covers in each round, to check yourself against
- A 7-day prep plan that matches where you are in your search
Hiring manager interview
One question, three answers, three scores
The same question answered three ways, scored the way an interviewer would mark it.
“What baseline did you compare your last model against, and by how much did it win?”
“I used a gradient boosted model and got an AUC of 0.91, which is really strong for this kind of problem.”
“I compared it to logistic regression, which got 0.84. The boosted model got 0.89, so we went with it.”
“The baseline was the rule the team already used: flag anyone with no login in 30 days. It caught about 40% of churners. My model caught 65% at the same contact volume, which meant the retention team called 300 fewer wrong customers a month. I kept it as logistic regression because the gain from boosting was small and the team needed to see why someone was flagged.”
Example scores on the 0 to 100 scale CareersAble uses. Your own answer gets its own score.
Reading the questions is not the same as answering them.
CareersAble writes the questions your actual job ad will ask, records your answers, scores each one three times and tells you plainly what to fix.
Free to start. No card needed.