JobInterview AI

Data Scientist interview questions

Data science interviews look for someone who solves a business problem with a model, not someone who looks for problems to fit a model. Expect questions about how you chose a metric, what happened in production, and when you decided not to use machine learning.

Questions to expect

  1. Tell me about a model you took to production.

    What they listen for

    The problem and the baseline before the algorithm. Then how it was served, monitored, and what it changed for the business.

  2. How did you choose the metric you optimised?

    What they listen for

    The link between the metric and the money or the user. And what the metric hides.

  3. Describe a project where the simple approach won.

    What they listen for

    That you tried the simple thing first and can say why complexity was not worth it.

  4. Tell me about an experiment whose result surprised you.

    What they listen for

    That you checked the experiment before believing the surprise: sample, duration, segments.

  5. How do you explain a model's prediction to someone who must act on it?

    What they listen for

    A concrete case with a sales rep, a doctor or an operations lead, and the words you actually used.

One answer, before and after

Tell me about a model you took to production.

You said

I built a gradient boosting model for lead scoring with about two hundred features, tuned the hyperparameters with Bayesian search and got an AUC of 0.87, which was a big improvement over the previous version.

Try

Sales were calling leads in the order they arrived. I built a score that ranked them by chance of signing; a logistic regression got us most of the way, so I kept it for explainability. Reps called the top third first, and meetings booked per rep went from 11 to 15 a month. I review drift monthly with the sales lead.

The problem and the result for the business come before the algorithm.

What a report scores for this role

Business impact
You state what the model changed, in the business's own unit.
Scientific rigour
You name the baseline, the validation and the limits.
Communication
You explain without jargon and without hiding uncertainty.

Examples: each report picks the competencies that fit the role and the interview.

Rehearse this interview out loud

Reading these questions is a start. Saying your answers to someone who follows up is the rehearsal.

Download on the App Store

Other roles