A data analyst loop: SQL, statistics and a stakeholder, over video
Sergei Skrylkov, Founder, Trippi · facts checked against the product on 2026-09-30
A data analyst loop tests three different skills in three different registers: writing SQL while explaining it, reasoning about statistics in plain words, and telling a business person something they did not want to hear. Most candidates prepare the first and lose offers on the third. And in a second language, the numbers themselves — percent versus percentage points, “a hundred twenty k” — are where the misunderstandings live.
The loop, round by round
Recruiter screen — 20–30 minutes
Tools you use (SQL dialect, Python or R, the BI tool), salary, notice. Say the tools as nouns, without versions.
SQL screen — 45–60 minutes, an analyst, shared editor
Two to four queries of growing difficulty on a small schema: joins, aggregation, then a window function. You talk while you type — the coding interview page covers that mechanic.
Take-home or case — a dataset and a few days, or 45 minutes live
A messy table and an open question. Graded on the questions you asked of the data and on the write-up. See take-home assignments.
Statistics and product analytics — 45 minutes, a senior analyst or data scientist
A/B tests, metric definitions, “the number went up, do we believe it”.
Stakeholder round — 45 minutes, a PM or a business lead
Explaining a result to someone who does not care about the method. Often includes presenting the take-home.
Hiring manager — 30–45 minutes
Your past analysis that changed a decision, how you prioritise requests, which team you would sit in.
Ten questions, and what each one tests
| Question | What it tests | What a strong answer does |
|---|---|---|
| “Return the second-highest salary in each department.” | Window functions and ties | Asks what happens with ties, then chooses DENSE_RANK or ROW_NUMBER on purpose. |
| “When did a LEFT JOIN give you the wrong number?” | Join semantics and row duplication | Explains a one-to-many join that inflated a sum, and how you caught it. |
| “How would you find duplicate orders in this table?” | Defining the problem before the query | Asks what counts as a duplicate — same id, or same customer and amount within a minute. |
| “Explain a p-value to our marketing manager.” | Statistics in plain language | One sentence without the word “null hypothesis”, then what it does not mean. |
| “The test shows +3% conversion, p = 0.04. Do we ship?” | Practical versus statistical significance | Checks the test before the result, then talks about effect size and cost of being wrong. |
| “Revenue is up, orders are down. What is going on?” | Metric decomposition | Splits revenue into orders × average order value and asks which products moved. |
| “How would you measure a loyalty programme’s success?” | Counterfactual thinking | Points out that loyal customers were already buying more, and proposes a comparison group. |
| “How do you handle missing data?” | Judgement over technique | Asks why it is missing before choosing to drop, fill or flag. |
| “A stakeholder wants a dashboard with forty metrics.” | Scoping and saying no | Asks what decision the dashboard is for, then proposes five. |
| “Tell me about analysis that changed a decision.” | Impact, not activity | The decision before, the number you found, the decision after. |
Worked example: the A/B test that looks good
The trap in this question is the p-value in it. Candidates hear “0.04” and answer about significance, when the interviewer wants to hear whether you trust the test at all. Here is the kind of structure a card offers — points to speak from, not a script.
Example of a card’s shape — written for this page, not a screenshot
The line it grew from
“The test shows plus three percent conversion with a p-value of point zero four. Would you ship it?”
Points to speak from
- 1Check the test first: were the groups the size you planned, did it run at least a full week, was anything else launched.
- 2Then size it: three percent relative or three points, on what base, and the confidence interval — not only the p-value.
- 3Decide by the cost of being wrong, and say what you would watch in the two weeks after launch.
The card does not know their base conversion rate or how long the test ran. Asking is part of the answer; the numbers they give you arrive on screen as lines of the call.
If English is your second language: the numbers vocabulary
| What you hear | The trap |
|---|---|
| “Up three percent” vs “up three points” | From 10% to 13% is three points, and thirty percent. Interviewers test whether you notice; say which one you mean. |
| “A hundred twenty k”, “two point four mil” | Thousands and millions, spoken fast. Repeat the number back in full if you are not sure. |
| “Point zero four” | Decimals are spoken with “point”, never “comma”. 0.04, not 0,04. |
| “What’s the grain of this table?” | What one row represents — one order, one order line, one customer per day. |
| “Slice it by…”, “cut it by…”, “drill down” | Group the result by a dimension and look inside the groups. |
| “Lift”, “uplift” | The improvement a change caused, usually relative. |
| “Sanity check” | A quick test that a number is plausible before you trust it. |
| “Skewed” | Not “wrong” — a distribution with a long tail, where the mean misleads. |
In the SQL round, table and column names are said out loud, often abbreviated — “the o-id on orders”. If a name is unclear, asking for the spelling is normal and costs nothing.
What Trippi Cue does in a data analyst loop, and what it cannot
Trippi Cue is a Chrome extension. Clicked in the call’s tab in Google Meet, Zoom on the web or Teams on the web, it listens to that tab’s audio, never your microphone, and nothing joins the call. What the interviewer says appears line by line with a translation under it if you want one, so “three points on a base of ten” is on your screen as text. When a question is addressed to you, a short card follows with a structure to answer from. You can also type your own question mid-call — “difference between RANK and DENSE_RANK” — and it answers with the conversation as context.
Limits, plainly: it does not see the shared SQL editor, the schema, the notebook or the dashboard. It cannot run a query and it cannot check yours. It hears speech, not your screen, and not your own voice. Browser tabs only — not desktop apps, not phones.
FAQ
Data analyst, specifically.
What SQL is asked in a data analyst interview?+
Joins, GROUP BY with HAVING, and at least one window function — ranking or a running total. Date handling comes up often. Explaining the query matters as much as the query.
How much statistics do I need?+
Enough to explain an A/B test result and its limits in plain language: significance, effect size, sample size, and why a test can be wrong.
Is a take-home normal for analyst roles?+
Yes, very common. The write-up and the questions you asked of the data are graded more than the code.
Can Trippi Cue read the schema in the shared editor?+
No. It hears the call’s audio only. Table names said out loud reach it; what is typed on screen does not.
Does it help with numbers said too fast?+
Every line appears as recognised text, with a translation if you choose one, so a number said quickly can be read a second later.
Read next
Percent or points — read it, then answer it.
Trippi Cue is in the Chrome Web Store. One click on the icon when the call begins, and nothing before that.