Build a causal ML pipeline end-to-end
Company: Capital One
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: hard
Interview Round: Onsite
You led a causal-inference project and now need to productionize policy targeting.
Tasks:
1) Formulate a DAG for treatment→outcome with key confounders and justify conditional independence assumptions. State your estimand (ATE and ITE/uplift).
2) Choose and justify a method (e.g., doubly robust learner, causal forest, uplift gradient boosting). Describe checks for overlap/positivity and how you’d handle violations.
3) Training/validation: Describe sample splitting for nuisance models, hyperparameter tuning without biasing effect estimates, and how you’d use policy risk/off-policy evaluation (IPW/DR) to compare targeting policies.
4) Diagnostics: Produce uplift curves/Qini, sensitivity to unobserved confounding (e.g., Rosenbaum bounds) and balance checks.
5) Deployment: Translate ITEs into a treatment policy with operational guardrails, fairness constraints across cohorts, and post-deployment monitoring.
Quick Answer: This question evaluates expertise in causal inference and policy learning, covering ATE/CATE estimation, heterogeneous treatment effect methods, off‑policy evaluation, overlap and sensitivity diagnostics, and operational deployment with fairness guardrails in the Analytics & Experimentation domain.
