American Express Data Scientist Interview Guide

The American Express Data Scientist interview, as candidates describe it, leans heavily on your own work. Interviewers spent much of their time on resume projects and internships, asking why a model was chosen and what data processing was done. One candidate was asked to explain how a decision tree works as if to a five-year-old. A candidate still in the process said the first round was mostly resume questions, and that listing a tool is not enough; you need to explain why you used it and how it worked.

Technical content varies by interviewer. One candidate was asked for a SQL query to find the second-highest salary in each department. Another described intermediate SQL questions plus query optimization. Statistics and ML questions included the Central Limit Theorem, handling a variable with more than 80% missing values, ensemble models, and overfitting and regularization when one category has little data. One candidate was asked why they used random forest rather than XGBoost, and why boosting is more prone to overfitting.

Cases lean toward the card business. Reported examples include raising the purchasing power of existing customers, a tree-based ML case, a credit risk project walkthrough, and a predictive servicing problem built around attrition risk with an A/B test evaluated using p-values. Puzzles and guesstimates also appear, such as the camel and banana puzzle, a jar-measuring puzzle, or estimating daily flights at Delhi airport. One candidate was asked about the company's revenue streams and the strengths and weaknesses of its business model.

Reported structures range from two to four rounds, often ending with a manager conversation about fit and communication. Outcomes were mixed: one candidate received an offer, while several did not. Practice narrating your projects, and state your assumptions out loud in cases.