American Express Data Scientist SQL and Coding Interview 2026
TL;DR
American Express data scientist interviews test SQL depth, not breadth. Coding rounds focus on business impact over algorithmic cleverness. The bar is high but not FAANG—expect 4-5 rounds, including a 90-minute SQL deep dive.
Who This Is For
Mid-level data scientists targeting American Express with 3-5 years of SQL experience, comfortable translating business problems into analytical queries. Not for fresh graduates or those weak in window functions and query optimization.
How hard is the American Express data scientist SQL interview?
The difficulty comes from query complexity, not obscurity. In a Q1 2025 debrief, a candidate failed for writing a 3-join solution where a single window function would suffice. The problem isn’t your syntax—it’s your ability to recognize anti-patterns in your own query.
American Express SQL questions mimic real fraud detection and customer segmentation scenarios. Expect nested window functions, rank/dense_rank edge cases, and date partitioning. The not X, but Y: not about memorizing functions, but about structuring queries that scale to 100M+ rows.
Their data model is relational but denormalized in places—you’ll need to handle wide tables with 50+ columns. One hiring manager noted that candidates who ask clarifying questions about schema relationships outperform those who assume standard star schemas.
Do they ask Leetcode-style coding questions?
They ask Leetcode-medium problems, but framed as business cases. A common 2025 prompt: "Given a list of transactions with timestamps, detect potential fraud in O(n) time." The not X, but Y: not about optimal Big-O, but about justifying your approach in business terms.
In one debrief, a candidate solved the problem in O(n log n) but explained how the tradeoff aligned with American Express’s real-time fraud detection latency requirements. They advanced. Another candidate wrote O(n) code but couldn’t articulate why it mattered. They didn’t.
The coding screen is 45 minutes with a senior data scientist. You’ll write Python in a shared doc, not an IDE. Their eval rubric weights correctness (50%), efficiency (30%), and business context (20%). Most candidates fail the last part.
What kind of SQL problems do they ask?
Expect 3-4 SQL questions in 90 minutes, with one requiring query optimization. A 2025 example: "Find the top 5 merchants by transaction volume for high-risk customers, excluding flagged transactions, with a rolling 7-day window." The not X, but Y: not about writing a working query, but about writing one that won’t timeout on their production data.
They test:
- Window functions with PARTITION BY and ORDER BY
- CTEs vs subqueries (and when to use each)
- Date/Time handling (especially with timezone-aware data)
- Handling NULLs in aggregations
A hiring committee debate in Q4 2024 revealed that candidates who use WITH clauses for readability score higher, even if the query is marginally less efficient. Clarity trumps micro-optimizations.
How do they evaluate coding and SQL together?
They want to see you bridge both worlds. In one 2025 final round, a candidate was given a Python function that called a SQL query and asked to optimize the combined runtime. The not X, but Y: not about mastering both languages separately, but about understanding how they interact in production.
Their systems team noted that candidates who ask about indexing or query execution plans stand out. American Express uses a mix of Spark and traditional RDBMS—knowing when to push logic to SQL vs Python is a differentiator.
What’s the interview process and timeline?
4-5 rounds total: recruiter screen (30 min), SQL assessment (90 min), coding screen (45 min), business case (60 min), final round (60 min with hiring manager). The timeline is tight—expect 2 weeks from first contact to offer for strong candidates.
The SQL assessment is the most consistent filter. In 2024, 60% of candidates failed here, most on window function edge cases. The coding screen has a 40% pass rate, with time management being the primary issue.
How much do American Express data scientists make?
2025 offers for mid-level data scientists range from $160K–$190K base, with $30K–$50K bonus and $20K–$40K RSUs. Senior roles (5+ years) see $200K+ total comp. The not X, but Y: not about the headline number, but about the RSU vesting schedule—American Express uses a 3-year cliff, unlike FAANG’s 4-year linear.
Preparation Checklist
- Master window functions: rank, denserank, rownumber, and their edge cases with ties
- Practice query optimization on large datasets (10M+ rows)
- Solve 10-15 Leetcode-medium problems with business context explanations
- Review fraud detection and customer segmentation case studies
- Brush up on date/time handling in SQL, including timezone conversions
- Work through a structured preparation system (the PM Interview Playbook covers Amex’s SQL-to-business translation patterns with real debrief examples)
- Prepare to explain your query decisions in terms of business impact, not just technical correctness
Mistakes to Avoid
- BAD: Writing a query that works but has O(n²) complexity because you didn’t consider the data volume.
- GOOD: Noting upfront, "This query assumes the dataset is under 1M rows; for larger datasets, I’d add a filter on the date range first."
- BAD: Solving a coding problem with optimal Big-O but no business justification.
- GOOD: "I chose this O(n log n) approach because it aligns with our real-time processing constraints, where sub-second latency is critical for fraud detection."
- BAD: Using nested subqueries when a CTE would improve readability.
- GOOD: "I restructured this with a CTE for clarity, as the nested subqueries made the logic harder to debug in production."
FAQ
What SQL dialect does American Express use?
They primarily use Spark SQL and Postgres. Focus on ANSI SQL compliance—dialect-specific functions are rare in interviews.
How many SQL queries will I need to write?
Expect 3-4 in the SQL assessment, with one requiring optimization or debugging. Time management is critical—spend 20-25 minutes per query.
Do they allow Google during the SQL interview?
No. The SQL assessment is in a locked environment. You’ll have access to a basic SQL reference, but not Stack Overflow or documentation.
Ready to build a real interview prep system?
Get the full PM Interview Prep System →
The book is also available on Amazon Kindle.