Data Analyst Interview Simulator — the stakeholder who has been burned by bad analysis
The person asking is a cross-functional lead who once made a decision on a chart that turned out to be wrong, and has never fully recovered from it. So they ask about denominators. They ask which segments you looked at before you concluded anything, and whether the effect survives when you split them. "The data showed" is not a sentence you can finish in this room without being asked how.
The kind of stakeholder asking the questions
Not a statistician, but someone who has learned the failure modes the expensive way: metric definitions and their edge cases, denominators, segmentation and Simpson's paradox, correlation against causation, confounders, sample size and minimum detectable effect, statistical power, p-values and their misuse, holdouts, A/B contamination, seasonality — and the dashboard graveyard, which they will mention with feeling. SQL, dbt, notebooks and BI tools come up in passing, never as a checklist you can recite.
What this room actually tests
Whether a metric-definition dispute you were part of actually got resolved and written down somewhere. How you avoid reading causation into observational data. Experiment design realism — sample size, duration, minimum detectable effect, and what result would have falsified your hypothesis. How you handled a stakeholder who did not like the finding, especially one senior to you. Awareness that most dashboards go unused, and what you specifically did about that. And tooling fluency shown in passing rather than listed.
What will get you pushed on
"The data showed" with no methodology behind it — expect "showed how?" before you finish the sentence. Impact claims where no decision actually changed as a result. Tool lists offered where a business number belongs. And confident causal language applied to observational data, which is the fastest way to lose this room; the follow-up will ask what else could explain the same pattern, and then ask again.
How difficulty changes this particular round
Warm-up lets you describe the analysis end to end before questioning the method. Realistic stops you at the conclusion and asks for the denominator and the segmentation. Nightmare supplies a plausible confounder for your headline finding and asks you to rule it out live, then re-asks about your sample size later to check that your first answer was considered rather than reflexive.
What the Sim Report grades
Methodological rigour, causal caution, stakeholder handling, and decision impact — whether anything actually changed because of your work. Each dimension quotes your exact wording, and the report is specific about which sentence overclaimed and how to phrase the same finding defensibly. Analysts usually gain the most points on the second run, once they stop saying "showed" and start saying "is consistent with".
Frequently asked
›Will it ask me to write SQL?
No — it is a voice round. You'll be asked how you'd structure a query or a join, not to type one out.
›Is there a case study or take-home?
No take-home. The persona will pose one live analytical situation and reason through it with you.
›I work mostly in dashboards, not experiments. Is this still useful?
Yes. The dashboard-adoption question is one of the sharpest in the bank, and it's aimed squarely at that work.
›How much statistics does it expect?
Applied, not academic. Knowing why a small sample makes your result unreliable matters more than naming the test.
›Can I practise for data scientist roles?
Partially — the experimentation and stakeholder questions transfer. Modelling and ML system design are not covered.
Related simulators
Four minutes of being asked for your denominator is cheaper than one bad decision made on your chart.
Run the analyst round free