Question 1

What questions do you ask the interviewer at the end of a data or AI role interview?

Accepted Answer

The questions you ask reveal how seriously you've thought about the role and what you actually care about in a workplace. Strong candidates ask about data infrastructure reality, how the team measures its own success, and what the first 90 days look like — not about benefits or vacation policy.

Question 2

Walk me through how you structure behavioral answers using the STAR method.

Accepted Answer

STAR stands for Situation, Task, Action, Result — a four-part frame that turns vague career stories into crisp, evidence-backed answers. The Result is the most scrutinized part: interviewers want a quantified outcome, not just 'it went well.' Keep each section tight so the whole answer lands in 90–120 seconds.

Question 3

Tell me about a project you're proud of.

Accepted Answer

This open-ended prompt is an invitation to show your best work on your own terms — pick one project that demonstrates technical depth, clear ownership, and measurable impact relevant to the role you're interviewing for. Do not summarize your entire résumé; go deep on one story.

Question 4

Why did you choose data science (or this data role), and what keeps you motivated in it?

Accepted Answer

This question is about self-awareness and genuine fit — not a test of enthusiasm. The best answers trace a specific intellectual or professional turning point, connect it to what you find durable about the work, and tie it forward to why this role and company in particular appeal to you.

Question 5

Tell me about a time you were wrong about something at work.

Accepted Answer

Being wrong and recognizing it quickly is a signal of strong analytical judgment — not weakness. The best answers name a specific, consequential mistake, explain how you discovered you were wrong, describe what you did about it, and connect it to a habit or process you changed as a result.

Question 6

How would you choose a north-star metric for a product, and what makes a metric a good north-star?

Accepted Answer

A north-star metric must satisfy three properties: it reflects the core value delivered to users, it correlates with long-term business outcomes (retention and revenue), and it is actionable — meaning teams can run experiments that move it. Choosing one requires articulating the product's value exchange and then stress-testing candidate metrics against those three criteria.

Question 7

How would you define and measure the success of a new feature — say, a 'Stories' feed added to a social app?

Accepted Answer

Success definition requires aligning a north-star metric to the feature's goal, pairing it with guardrail metrics that catch side-effects, and deciding on a measurement window before launch. For a Stories feed, adoption rate and daily story views per active user are reasonable primary signals, while core feed engagement and notification opt-out rate serve as guardrails.

Question 8

Tell me about a time you disagreed with a stakeholder or PM about a data or model decision.

Accepted Answer

This question tests whether you advocate for data-driven thinking without becoming combative, and whether you know when to escalate versus adapt. Strong answers show you led with evidence, listened to the stakeholder's actual concern, and reached a resolution that was better for the product than either original position.

Question 9

How do you explain a technical result or model to non-technical stakeholders?

Accepted Answer

The best communicators translate outputs into decisions, not equations. Lead with the business implication, use an analogy for the mechanism, and reserve technical detail for an appendix or follow-up. Calibrate depth to the audience in the room, not to what you find interesting.

Question 10

How do you handle a project with ambiguous or poorly-defined requirements?

Accepted Answer

Ambiguity is the default state of most real data work — the practitioner who panics or stalls without a clear spec is a liability. Strong answers show a repeatable approach: clarify the decision being supported, propose a scoped first step, and deliver something reviewable early rather than disappearing for three weeks.

Question 11

How do you prioritize when you have multiple competing requests or projects?

Accepted Answer

Prioritization for data roles comes down to three inputs: business impact, stakeholder urgency, and the actual effort required. Strong candidates describe a repeatable system — not just 'I talk to my manager' — and give a concrete example of a tradeoff they made and why.

Question 12

How would you measure user engagement for a mobile app — what metrics would you use and how would you structure them?

Accepted Answer

Engagement is multi-dimensional: breadth (how many users engage), depth (how much they do per session), and frequency (how often they return). A robust engagement framework stacks these three layers into a metric hierarchy and links them to retention curves, because engagement that does not predict long-term retention is usually noise.

Question 13

What is the difference between leading and lagging indicators, and how do you use both when building a metrics system?

Accepted Answer

Lagging indicators (revenue, annual retention, NPS) measure outcomes after they have occurred — they are accurate but slow. Leading indicators (D1 retention, feature adoption rate, time-to-value) correlate with future outcomes and are available faster, making them suitable for early experiment decisions. A robust metrics system pairs both, with the leading metric as the experiment signal and the lagging metric as the validation gate.

Question 14

Estimate the number of Uber rides taken in New York City on a typical weekday.

Accepted Answer

Market-sizing questions test whether you can decompose a complex number into estimable sub-components, make explicit and reasonable assumptions, sanity-check against known benchmarks, and communicate uncertainty without losing structure. The answer itself matters less than the reasoning chain.

Question 15

Tell me about a time you had to make a tradeoff under a tight deadline.

Accepted Answer

Deadline pressure forces explicit tradeoffs that normal project timelines let you avoid — and how you navigate them reveals your judgment. Interviewers want to see that you identified the right thing to cut, communicated the tradeoff clearly rather than silently, and delivered something the team could trust within the constraint.

Question 16

Tell me about a time a model or analysis you built failed or underperformed.

Accepted Answer

Interviewers ask this to test intellectual honesty, ownership, and how you learn from setbacks — not to embarrass you. The strongest answers name a real failure, explain the root cause clearly, describe what you did to fix or contain the damage, and articulate the lasting lesson you carried forward.

Question 17

How do you choose the classification threshold for a model when the goal is a business outcome, not pure accuracy?

Accepted Answer

The default 0.5 threshold optimises for balanced accuracy but is rarely the right choice for business objectives. The correct threshold is found by translating the business cost of false positives and false negatives into a cost matrix, then sweeping the threshold on a held-out set to find the point that minimises expected cost or maximises expected profit. Operational constraints — such as review-team capacity — further bound the feasible region.

Question 18

How would you design a metric to measure the quality of a search feature inside an e-commerce app?

Accepted Answer

Search quality has two sides: relevance (did results match intent?) and utility (did the user accomplish their goal?). A good metric system combines an offline relevance signal — such as NDCG computed against human-labelled queries — with an online behavioural signal — such as click-through rate at rank 1 and zero-result rate — tied to a downstream business outcome like add-to-cart rate.

Question 19

How would you design a metric to evaluate the relevance of a content recommendation feed?

Accepted Answer

Feed relevance has no single ground-truth label, so it requires a tiered metric system: an implicit behavioural signal (long dwell time, saves, shares) as the online primary metric; an explicit user-satisfaction signal (thumbs-up/down, survey) as the periodic validation; and an offline ranking metric (NDCG computed from historical high-engagement items) for fast model iteration. The three tiers must converge to be trusted.

Question 20

30-day retention dropped from 42 % to 31 % over the last two months. How do you diagnose the root cause?

Accepted Answer

A retention drop investigation requires distinguishing between an acquisition-mix shift (newer cohorts are lower quality) and a genuine product regression (existing cohorts are performing worse). The two look identical in aggregate retention but have completely different fixes. Cohort analysis — plotting the D30 survival curve for each weekly acquisition cohort — is the first move.

Question 21

You are asked to 'use ML to improve the user experience on our platform.' How do you approach this completely open-ended problem?

Accepted Answer

Open-ended ML problems require scoping before modelling: translate the vague ask into a measurable business objective, identify which user interaction has the highest improvement potential, formulate it as a concrete ML task with a defined label and evaluation metric, then propose the simplest viable model first. Jumping to model architecture before this scoping is the most common interview failure mode.

Question 22

How would you detect whether a metric is being gamed or is otherwise artificially inflated?

Accepted Answer

Gaming is detectable by looking for statistical signatures: abnormal distribution tails, sudden regime changes that correlate with incentive changes rather than product changes, divergence between the metric and correlated downstream outcomes, and segment-level anomalies that cancel out in the aggregate. The detective work combines anomaly detection with causal reasoning about who benefits from inflating the number.

Question 23

DAU dropped 15 % week-over-week with no planned changes. How do you diagnose it?

Accepted Answer

A metric drop investigation starts by confirming the drop is real — ruling out logging bugs and metric-definition changes — before hypothesising causes. Then segment by platform, geography, user cohort, and funnel step to isolate where the drop is concentrated, which points to the most likely root cause.