Image by Author
# Introduction
Everyone knows what comes up in data science interviews: SQL, Python, machine learning models, statistics, sometimes a system design or case study. If this comes up in the interviews, it’s what they test, right? Not quite. I mean, they sure test everything I listed, but they don’t test only that: there’s a hidden layer behind all those technical tasks that the companies are actually evaluating.
Image by Author | Imgflip
It’s almost a distraction: while you think you’re showcasing your coding skills, employers are looking at something else.
That something else is a hidden curriculum — the skills that will actually reveal whether you can succeed in the role and the company.
Image by Author | Napkin AI
# 1. Can You Translate Business to Data (and Back)?
This is one of the biggest skills required of data scientists. Employers want to see if you can take a vague business problem (e.g. “Which customers are most valuable?”), turn it into a data analysis or machine learning model, then flip the insights back into plain language for decision-makers.
What to Expect:
- Case studies framed loosely: For example, “Our app’s daily active users are flat. How would you improve engagement?”
- Follow-up questions that force you to justify your analysis: For example, “What metric would you track to know if engagement is improving?”, “Why did you choose that metric instead of session length or retention?”, “If leadership only cares about revenue, how would you reframe your solution?”
What They’re Really Testing:
Image by Author | Napkin AI
- Clarity: Can you explain your points in plain English without too many technical terms?
- Prioritization: Can you highlight the main insights and explain why they matter?
- Audience awareness: Do you change your language depending on your audience (technical vs. non-technical)?
- Confidence without arrogance: Can you explain your approach clearly, without getting overly defensive?
# 2. Do You Understand Trade-Offs?
At your job, you’ll constantly have to make trade-offs, e.g. accuracy vs. interpretability or bias vs. variance. Employers want to see you do that in interviews, too.
What to Expect:
- Questions like: “Would you use a random forest or logistic regression here?”.
- No correct answer: Scenarios where both answers could be right, but they’re interested in the why of your choice.
What They’re Really Testing:
Image by Author | Napkin AI
- No universally “best” model: Do you understand that?
- Framing trade-offs: Can you do that in plain terms?
- Business alignment: Do you show the awareness to align your model choice with business needs, instead of chasing technical perfection?
# 3. Can You Work with Imperfect Data?
The datasets in interviews are rarely clean. There are usually missing values, duplicates, and other inconsistencies. That’s deliberate to reflect the actual data you’ll have to work with.
What to Expect:
- Imperfect data: Tables with inconsistent formats (e.g. dates show as 2025/09/19 and 19-09-25), duplicates, hidden gaps (e.g. missing values only in certain time ranges, for example, every weekend), edge cases (e.g. negative quantities in an “items sold” column or customers with an age of 200 or 0)
- Analytical reasoning question: Questions about how you’d validate assumptions
What They’re Really Testing:
Image by Author | Napkin AI
- Your instinct for data quality: Do you pause and question the data instead of mindlessly coding?
- Prioritization in data cleaning: Do you know which issues are worth cleaning first and have the biggest impact on your analysis?
- Judgement under ambiguity: Do you make assumptions explicit so your analysis is transparent and you can move forward while acknowledging risks?
# 4. Do You Think in Experiments?
Experimentation is a huge part of data science. Even if the role isn’t explicitly experimental, you’ll have to perform A/B tests, pilots, and validation.
What to Expect:
What They’re Really Testing:
Image by Author | Napkin AI
- Your ability to design experiments: Do you clearly define control vs. treatment, perform randomization, and consider sample size?
- Critical interpretation of results: Do you consider statistical significance vs. practical significance, confidence intervals, and secondary effects when interpreting the experiment’s results?
# 5. Can You Stay Calm Under Ambiguity?
Most interviews are designed to be ambiguous. The interviewers want to see how you operate with imperfect and incomplete information and instructions. Guess what, that’s precisely what you’ll get at your actual job.
What to Expect:
- Vague questions with missing context: For example, “How would you measure customer engagement?”
- Pushing back on your clarifying questions: For example, you might try to clarify the above by asking, “Do we want engagement measured by time spent or number of sessions?”. Then the interviewer could put you on the spot by asking, “What would you pick if leadership doesn’t know?”
What They’re Really Testing:
Image by Author | Napkin AI
- Mindset under uncertainty: Do you freeze, or stay calm and pragmatic?
- Problem structuring: Can you impose order on a vague request?
- Assumption-making: Do you make your assumptions explicit so that they can be challenged and refined in the following analysis iterations?
- Business reasoning: Do you tie your assumptions to business goals or to some arbitrary guesses?
# 6. Do You Know When “Better” Is the Enemy of “Good”?
Employers want you to be pragmatic, meaning: can you give as useful results as quickly and as simply as possible? A candidate who would spend six months improving the model’s accuracy by 1% isn’t exactly what they’re looking for, to put it mildly.
What to Expect:
- Pragmatism question: Can you come up with a simple solution that solves 80% of the problem?
- Probing: An interviewer pushing you to explain why you’d stop there.
What They’re Really Testing:
Image by Author | Napkin AI
- Judgement: Do you know when to stop optimizing?
- Business alignment: Can you connect solutions to business impact?
- Resource-awareness: Do you respect time, cost, and team capacity?
- Iterative mindset: Do you ship something useful now, then improve later, instead of spending too much time devising a “perfect” solution?
# 7. Can You Handle Pushback?
Data science is collaborative, and your ideas will be challenged, so the interviews replicate that.
What to Expect:
- Critical reasoning test: Interviewers trying to provoke you and poke holes in your approach
- Alignment test: Questions like, “What if leadership disagrees?”
What They’re Really Testing:
Image by Author | Napkin AI
- Resilience under scrutiny: Do you stay calm when your approach is challenged?
- Clarity of reasoning: Are your thoughts clear to you, and can you explain them to others?
- Adaptability: If the interviewer exposes a hole in your approach, how do you react? Do you acknowledge it gracefully, or do you get offended and run out of the office crying and screaming expletives?
# Conclusion
You see, technical interviews are not really about what you thought they were. Keep in mind that all that technical screening is essentially about:
- Translating business problems
- Managing trade-offs
- Handling messy, ambiguous data and situations
- Knowing when to optimize and when to stop
- Collaborating under pressure
Nate Rosidi is a data scientist and in product strategy. He’s also an adjunct professor teaching analytics, and is the founder of StrataScratch, a platform helping data scientists prepare for their interviews with real interview questions from top companies. Nate writes on the latest trends in the career market, gives interview advice, shares data science projects, and covers everything SQL.
