Analyst interviews test SQL less than candidates expect and persuasion far more. The technical screen filters; the conversation decides.
What this role is actually evaluated on
- Whether anyone acted on your work. The most common weakness in analyst answers is a beautiful piece of analysis that changed nothing.
- Whether you can be understood by people who are not analysts. Most of the job is translation.
- What you do when the data is bad. It usually is. Candidates who describe pristine datasets are describing a course, not a job.
- Whether you will say a number is wrong. Including when it is your own number, and including when somebody senior likes it.
Questions you should be able to answer out loud
Not a bank to memorise. These are the shapes the real questions take, and if you can answer these you can answer most of what comes.
- Tell me about a time your analysis changed a decision. What was the decision before you were involved?
- Tell me about a time you found a number you did not believe. What did you do?
- Explain something you built to someone with no technical background. Pick the thing you are proudest of.
- A stakeholder’s intuition contradicts your numbers two days before a launch. What do you actually do?
- Tell me about a dataset that was a mess. How did you decide what was usable?
- What is a metric you have seen misused, and what was the harm?
The mistakes that cost candidates this job
- Describing the analysis, not the outcome. The interviewer knows what a cohort chart is. They want to know what happened because of it.
- Listing tools. “SQL, Python, Tableau, dbt” tells them nothing. One sentence about a thing you built with one of them tells them everything.
- Never disagreeing with anyone in any story. If every story is frictionless, the interviewer assumes you have never had to hold a position.
- “I would present the data and let them decide.” A judgement question is asking you to make a judgement.
Preparing for it
Pick two stories where your work visibly changed what somebody did, and find the actual number for each. Then practise saying one of them in under ninety seconds, because the temptation to explain the methodology will eat your whole answer.