Data analyst video interviews: build one sheet, then practice saying the numbers
Analyst interviews reward preparation you can do on paper. Choose the analyses you will talk about, write down what changed because of each and decide how the numbers will sound out loud.
1. The evidence sheet
Write three to five analyses on a single page with the same five columns. Most of the behavioral questions can be answered from one of the rows. Three invented rows show the level of detail to aim for.
| Analysis | What changed | The number | The caveat | Can answer |
|---|---|---|---|---|
| Signups dropped; one paid channel was the cause | Paused the channel and moved its budget | Weekly signups down 12%; the channel accounted for nearly all of it | Attribution was last-click, so some of the drop may have been shifted, not lost | Questions 2, 8, 9 |
| Two dashboards disagreed on active users | One definition adopted; the other dashboard retired | Gap of about 9% traced to how sessions were deduplicated | Older reports were not restated | Questions 3, 6 |
| A reporting query that took forty minutes | Rewritten and scheduled; the weekly report went out on time | Forty minutes to under two | Relied on a summary table that someone must keep up to date | Question 10 |
| Your analysis here | ||||
| Your analysis here | ||||
The caveat column matters most. An analyst who volunteers what the number cannot show is easier to believe than one who is asked. Question numbers refer to the practice set below.
- 1Walk me through how you approach a new analytical problem from question to insight.
- 2Tell me about a time your analysis led to a business decision that had measurable impact.
- 3How do you handle data quality issues and communicate their impact on your findings?
- 4Describe how you design a dashboard that stakeholders will actually use rather than ignore.
- 5How do you tell a clear story with data to an audience that is skeptical or unfamiliar with the numbers?
- 6Tell me about a time two data sources gave conflicting answers to the same question.
- 7How do you prioritize analytical requests when everything is marked urgent?
- 8Describe a time you found an insight that contradicted what leadership believed to be true.
- 9How do you measure whether your analytical work is having business impact?
- 10Tell me about your approach to SQL and what you do when a query is too slow or too complex to maintain.
2. Saying the numbers
Nothing on the page tells you how to read a statistic aloud and on a call the listener has no chart to look at. Decide in advance how each one will sound.
| On the page | Out loud |
|---|---|
| 12.4% | “About twelve percent.” Round unless the exact figure is the point and say what it is a percent of. |
| +12% WoW | “Up twelve percent on the week before.” Never say the abbreviation. |
| n = 1,284 | “Around thirteen hundred users.” Give the size before the result so the listener can weigh it. |
| p < 0.05 | “If there were no real effect, a gap this large would show up less than one time in twenty.” Say it plainly, once and move on. |
| 95% CI: 8 to 16% | “Somewhere between eight and sixteen percent. I would not plan around the exact middle.” |
| LEFT JOIN orders ON customer_id | “I joined orders to customers, so every row is one customer in one month.” Say what the join does to the rows, not the keyword. |
3. Explaining a query with no screen
If you are asked to walk through a query or a model and cannot share it, use the same four steps every time and the explanation stays short.
- Grain. “The result has one row per customer per month.”
- Filter. “I only kept paid accounts, because trials behave differently.”
- Join. “I joined to orders and checked the row count did not grow.”
- Aggregate. “Then I summed revenue and divided by active days.”
Record one row of the sheet
Take the analysis you know best and answer “Tell me about a time your analysis led to a business decision that had measurable impact.” from it, saying each number the way you planned. In the report, look at your pace on the sentences that carry figures and whether any of them were hard to say.