Below are representative questions by area. Use them to assess readiness, identify gaps, and practice crisp, structured answers with examples from your experience.

Technical / Domain (Statistics & ML)

Expect to explain concepts and apply them to real scenarios.

• Explain bias-variance trade-off and how you detect high variance in practice.
• How do you select evaluation metrics for imbalanced classification?
• Walk through cross-validation strategies for time-series forecasting.
• Describe regularization (L1 vs. L2) and when you’d use each.
• How do you detect and prevent data leakage?

Coding & SQL

Be prepared to write and discuss code for data manipulation and analysis.

• Write a SQL query to compute a 7-day rolling average of daily active users.
• In pandas, convert event-level logs into user-level session features.
• Optimize a slow SQL query that joins large tables with window functions.
• How would you structure a Python module to train and evaluate a model reproducibly?
• Explain vectorization in pandas and when loops are acceptable.

Problem-Solving / Case Studies

We assess how you frame ambiguous business problems and design solutions.

• A client’s churn is rising—how do you diagnose and model the problem?
• Design a lead scoring system: data sources, features, model choice, and metrics.
• The model performs well in offline tests but fails in production—what’s your plan?
• Propose a minimal MVP to forecast weekly demand with limited history.
• You have missing data on a critical feature—how do you proceed and communicate risk?

Experimentation & Causal Inference

Demonstrate test design, metric selection, and interpretation.

• Design an A/B test for a new onboarding flow with a guardrail metric.
• Explain p-value vs. confidence interval to a non-technical stakeholder.
• What do you do if you accidentally peek mid-test and see significance?
• When RCTs aren’t possible, what quasi-experimental options exist?
• Interpret a result where lift is small but statistically significant.

Behavioral / Leadership

Show ownership, collaboration, and communication under ambiguity.

• Tell me about a time you drove a project from ambiguity to impact.
• Describe a conflict with a stakeholder and how you resolved it.
• Share a mistake you made in analysis and how you prevented recurrence.
• How do you prioritize when everything seems important?
• Give an example of simplifying a complex concept for an executive.

Deep Dive into Evaluation Areas

Applied Statistics & Machine Learning

This area measures your ability to select, train, and evaluate models that solve real business problems. Expect to justify modeling choices, articulate assumptions, and choose metrics aligned with objectives (classification vs. regression, ranking vs. forecasting). You will likely be asked to compare approaches and reason about trade-offs under constraints.

Be ready to go over:

• Core statistics: distributions, hypothesis testing, confidence intervals, p-values vs. practical significance
• Model selection & validation: cross-validation, regularization, bias-variance, hyperparameter tuning
• Metrics & diagnostics: ROC-AUC, PR-AUC, F1, RMSE/MAE, lift, calibration, residual analysis
• Advanced concepts (less common): time-series cross-validation, survival analysis, SHAP/feature importance caveats, causal inference basics

Example questions or scenarios:

• "You have imbalanced classes—how do you choose and justify your metric and model?"
• "Walk me through how you would validate a forecasting model for weekly demand."
• "Your XGBoost model outperforms logistic regression on AUC but underperforms on PR-AUC. What do you do and why?"

Coding & Data Manipulation (Python + SQL)

We test your ability to write clean, efficient code to explore data, engineer features, and answer questions. Interviews often blend Python data wrangling and SQL querying to ensure you can build analyses end-to-end.

Be ready to go over:

• Python data stack: pandas/numpy operations, groupby/merge/reshape, handling missing/outliers
• SQL fluency: joins, window functions, aggregations, subqueries, CTEs, performance considerations
• Code quality: readability, modularity, testing simple helpers, reproducibility
• Advanced concepts (less common): vectorization trade-offs, basic complexity, query optimization basics

Example questions or scenarios:

• "Given two tables (transactions, customers), compute 30-day rolling revenue per customer using SQL."
• "In pandas, convert event logs to session-level features; discuss edge cases."
• "Refactor a messy notebook into functions and explain testing strategy."

Experimentation & Causal Reasoning

Many projects require A/B testing or evidence of causal impact. We assess how you design experiments, pick metrics, and interpret results responsibly.

Be ready to go over:

• Test design: randomization, power, sample size, guardrail metrics
• Analysis: difference-in-means, non-parametrics, multiple testing, pitfalls (peeking)
• Alternatives: quasi-experiments when RCTs are infeasible
• Advanced concepts (less common): CUPED, inverse propensity weighting, diff-in-diff assumptions

Example questions or scenarios:

• "Design an A/B test to improve signup conversion; define primary/secondary metrics."
• "Results are not statistically significant but business sees a lift—how do you respond?"
• "Randomization failed post-hoc. How do you analyze and communicate risk?"

Data Wrangling, Pipelines, and Visualization

Strong Data Scientists create reliable pipelines and clear visuals that drive decisions. We evaluate how you clean data, automate steps, and present insights succinctly.

Be ready to go over:

• Data quality: missing data strategies, deduplication, anomaly detection
• Pipelines: reproducibility, version control, environments, scheduling basics
• Visualization & storytelling: chart selection, avoiding misrepresentation, dashboard fundamentals
• Advanced concepts (less common): data contracts, data validation tests (e.g., Great Expectations)

Example questions or scenarios:

• "Build a minimal, reproducible pipeline for a weekly churn score refresh."
• "Which visualization best explains a precision-recall trade-off to non-technical leaders?"
• "You inherit a flaky CSV-based pipeline—what’s your stabilization plan?"

Business Acumen & Communication

We measure how well you translate data into decisions. You must connect analyses to strategy, articulate risks, and influence stakeholders.

Be ready to go over:

• Problem framing: turning an ambiguous prompt into testable hypotheses
• Prioritization: scoping MVPs, choosing high-ROI features, communicating trade-offs
• Stakeholder alignment: tailoring depth/format to the audience
• Advanced concepts (less common): north-star metrics, cost-sensitive modeling, scenario planning

Example questions or scenarios:

• "A stakeholder wants a complex deep model—how do you challenge and reframe to meet the actual goal?"
• "Tell a story of a project that changed course based on your analysis."
• "Explain your model to an executive in 90 seconds—what do you include and exclude?"