Intervooh · Interview questions by job
Data Scientist interview questions (2026)
Researched, current questions asked in real data scientist interviews (Data & AI), with what a strong answer actually does. Questions marked 2026 are the newer, AI-era questions employers now ask.
Build my free day-by-day prep plan →
Tell it the company, role and date; it does the rest. Free, no card.
What they assess
- Statistical reasoning
- Modelling & machine learning
- Experimentation & causal inference
- Communicating findings
- Business & product impact
The questions to expect
Your fraud model has 94% accuracy and the fraud team says it's useless. What's going on?
Class imbalance: with rare fraud, accuracy is meaningless. Talk precision, recall and the cost of each error type — in the fraud team's terms.
Customer retention dropped eight points last quarter. How would you use data to understand why?
Clarify how retention is defined here, check for measurement change, then segment and form hypotheses. Structure first, methods second.
Design an A/B test for a change we're considering. How would you size it, and when would you stop?
Cover metric choice, randomisation unit, power and duration — and resist peeking. Mentioning novelty effects marks you as experienced.
How do you choose between a simple model and a complex one for a business problem?
Default to the simplest model that meets the need — interpretability, maintenance and data volume all argue for it. Complexity must pay rent.
Tell me about a time you had to explain a model's limitations to stakeholders who wanted certainty.
Show the translation: what the model can and can't say, in decision language, plus what you offered instead of false confidence.
Walk me through a project where your model reached production. What was your role after deployment?
Post-deployment ownership — monitoring, drift, retraining, measuring realised impact — separates practitioners from notebook scientists.
Describe a time your analysis produced a counterintuitive result. How did you verify it before sharing?
Show healthy self-suspicion: check the pipeline, try to break your own result, seek a second mechanism. Then commit to the finding.
Why data science, rather than analytics or data engineering?
Show you know the difference — you want the modelling and inference layer — and back it with what you've actually chosen to build.
LLMs can now run basic analyses end to end. Where do you add value beyond what an AI produces?2026
Problem framing, causal reasoning, knowing the business context, and catching plausible-but-wrong output. Confidence without defensiveness.
How do you use AI assistants in your own analysis workflow — and where have they led you astray?2026
A real 'led me astray' story is the credibility marker — hallucinated methods, subtly wrong code — plus the checking habit you built.
Walk me through a piece of analysis or a model that changed a business decision.
Name the decision, not just the deliverable. The strongest answers end with what the business did differently and what that was worth.
Tell me about a time a stakeholder challenged your numbers. How did you respond?
Show you checked before defending. Being openly willing to find your own error is what builds trust in your numbers.
Preparation notes
[object Object]
Turn this into a plan
A list of questions is a start; a programme is what changes the outcome. Intervooh builds a day-by-day plan for your exact data scientist interview — company research, story building with an AI coach, spoken practice with delivery feedback, and scored mock interviews.