Frameworks, not scripts. Memorised answers fall apart under a follow-up question; a framework survives one.

Where to actually practise

• DataLemur: free SQL and analytics questions sourced from real interviews at named companies, curated by the author of Ace the Data Science Interview. The best starting point if you want realistic questions, not textbook ones.
• LeetCode SQL 50: a curated problem set organised by concept, useful once you want structured, progressive difficulty rather than a random mix.
• StrataScratch: a large bank of real interview questions if you want volume over guided feedback.
• HackerRank SQL: free, includes a skills certification, and many companies use HackerRank's actual interface for screening, so practising here also builds familiarity with the format itself.

SQL

Expect joins, window functions, and a query-optimisation question at Mid+.

Framework: narrate your approach before writing anything: "I'd start by understanding the grain of this table, then..." Interviewers are scoring how you think, not just the final query.

Real question shapes worth practising, not just one:

• "Find the second-highest salary per department." Talk through why you'd use RANK()/DENSE_RANK() over a naive subquery, and what changes if there are ties.
• "Find customers who made a purchase in every month of the year." A classic use of GROUP BY combined with a count check against the number of distinct months.
• "Find the running total of sales by day." The canonical window function question, tests whether you reach for SUM() OVER (ORDER BY ...) instead of a self-join.
• "Identify duplicate rows in a table." Tests whether you know ROW_NUMBER() partitioned by the duplicate-defining columns, then filter to keep one.

Stats and product sense (DA-focused)

Expect questions like "how would you measure the success of X feature," and experiment-design questions.

Framework: clarify the goal metric first, name the risk of a bad proxy metric, then design the test. Always state an assumption you're making out loud: interviewers read silence on assumptions as not having considered them.

Real question shapes:

• "How would you measure whether a new checkout flow is better?" Tests whether you pick a metric tied to the actual business goal (conversion, revenue per user) rather than a vanity metric (clicks).