Find me on
What’s the Impact? Experimentation and Causal Inference Get the Playbook
Free resources

Data science interviews: 4 types of case studies in 2026

The four types of data science case study I saw across 80+ interview rounds at 16 companies in 2026, and how to tell which one you are in.

Data science interviews: 4 types of case studies in 2026, from patterns across 80+ interviews at 16 companies in Spring 2026.
Case study type 1, launch evaluation. You hear "We want to launch X, how would you know it's working?" It tests precise design and analysis of marketplace experiments, and the common trap is jumping to experiment design before discussing downstream effects.
Case study type 2, root cause analysis. You hear "Active minutes dropped 15% starting last Tuesday." It tests whether you can uncover a wide range of hypotheses, and the common trap is settling on the obvious explanation.
Case study type 3, opportunity sizing. You hear "How do we know this feature is worth building?" It tests a defensible value estimate from existing data, and the common trap is proposing an experiment when the real question is whether the build is worth its cost.
Case study type 4, targeting and uplift. You hear "We can only send 100k discounts, who gets them?" It tests who to target with a costly intervention, and the common trap is targeting users likely to convert when many would have converted anyway.
Cheat sheet for telling which of the four case study types you are in from the interviewer's first two sentences, plus a bonus round: any of the four can end with "Is this causal?"

The slides above map the 4 types of data science case study questions I encountered across 80+ interview rounds at 16 companies in April-May 2026. For each type it covers the interviewers’ typical opening, what’s being tested, and the trap I saw most often in my own interviews and in mock interviews with clients. Identifying the type early matters because the correct approach differs for each one.

A few years ago the case study round was close to synonymous with marketplace experimentation. That question still regularly appears, but the same loop can now hand you (1) a launch evaluation, (2) a root cause analysis of a metric that suddenly moved, (3) an opportunity sizing decision about whether a build is worth its cost, or (4) a targeting problem that turns on heterogeneous treatment effects. Any of the four can also end with questions that call for observational causal inference, depending on your level.

← All free resources

Every round in the loop, written up

The Playbook covers each of these formats in detail, along with case studies, statistics, applied machine learning, project deep dives and the behavioral rounds, drawn from the 80+ rounds I went through in Spring 2026.

See what the Playbook covers Other free resources