IBM·Data Scientist·Technical Phone Screen·Junior

Junior

May 2026

Summary

IBM data scientist internship interview for a marketing analytics team. One main question that was really more of a guided case study, asking you to walk through two different modeling approaches on the same project. Felt more like a structured conversation than a typical interview.

Questions Asked (1)

Walk me through a project where you built both a statistical model (like logistic regression or a GLM) and a machine learning model (like a tree-based method or boosting). Cover the business problem, how you defined the target variable, what features you used and why, how you handled stakeholder expectations around interpretability versus performance, and what you'd do differently looking back.

Data ModelingStakeholder ManagementTechnical Trade-offs

Author's notes

This is a lot to hold in your head at once.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Choose a project where you deliberately built both a statistical model (e.g., logistic regression) and a machine learning model (e.g., XGBoost) to compare interpretability and performance. Structure your answer using a clear narrative: business context, target definition, feature engineering, stakeholder alignment, and retrospective lessons. Emphasize how you balanced trade-offs and communicated results to non-technical audiences.

Pro tip: Quantify the performance gap between the models (e.g., 'the ML model improved AUC by 5 points but the logistic regression was within 2 points and far more interpretable') and explain how you used that to drive the final recommendation. This shows you think in terms of business value, not just model metrics.

1. Set the Business Context

Briefly describe the business problem, why it mattered, and what decision the model would inform. Mention the key stakeholders and their primary concerns (e.g., regulators needing explainability, marketing needing lift).

2. Define the Target and Features

Explain how you defined the target variable (e.g., churn in next 90 days) and why. Describe the features you engineered, including any domain-driven transformations, and how you ensured they were available at prediction time.

3. Build and Compare Models

Detail the statistical model (e.g., logistic regression with regularization) and the ML model (e.g., gradient boosting). Highlight how you evaluated both (e.g., AUC, calibration, lift) and any trade-offs in performance vs. interpretability.

4. Manage Stakeholder Expectations

Describe how you communicated the trade-offs, such as using SHAP values to explain the ML model or presenting coefficients from the logistic regression. Explain how you aligned on the final model choice and addressed concerns about complexity or trust.

5. Reflect and Improve

Share what you would do differently, such as starting with a simpler model as a baseline, involving stakeholders earlier, or investing in better feature monitoring. Emphasize lessons learned about balancing rigor and pragmatism.

Key Points to Mention

Target variable definition and alignment with business objective (e.g., precision-recall trade-off based on cost of false positives/negatives)
Feature selection rationale: domain knowledge, data leakage prevention, and handling of categorical variables
Model evaluation metrics beyond accuracy: AUC, lift, calibration, and business KPIs
Interpretability techniques: coefficients, odds ratios, SHAP, partial dependence plots
Stakeholder communication: translating technical trade-offs into business impact and building trust
Retrospective improvements: automating feature pipelines, setting up model monitoring, or conducting A/B tests

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.