Take-home alternative
The take-home stopped working. Here is what to use instead.
Updated 26 September 2026.
The short answer
A take-home case study now mostly shows who used an AI well, costs a candidate 8 to 15 hours of unpaid work, and costs you an evening per candidate to mark. The alternative is a short work sample in the browser: one hour of real analyst work where the process is visible and scored, not only the final answer.
Why the take-home broke
- An AI does most of it in minutes. Paste in the brief and the data, and a model returns clean SQL, a chart and a tidy summary. Every submission starts to look the same.
- Your best candidates drop out. People with other offers do not spend a weekend on unpaid work. The ones who finish are not always the ones you wanted.
- You only see the answer. Whether they checked the data, what they tried first and how sure they were is invisible in a finished document.
- Marking takes an evening each. With five finalists, that is a week of evenings.
What to use instead
The research on hiring has said for decades that a sample of the real work predicts performance better than a CV or an unstructured interview. The take-home was a work sample. What broke is that it is done alone, at home, over days. Keep the work sample, and change the conditions:
- Short and timed, so it respects the candidate's time and leaves no room to outsource it.
- In the browser, so you see the steps and not only the result.
- Different data for each candidate, so answers cannot be shared or looked up.
- Scored on how they work: checking the data, knowing when they are unsure, catching a wrong number.
Take-home, live SQL interview, or work simulation
| Take-home | Live SQL interview | Trial Day | |
|---|---|---|---|
| Candidate's time | 8 to 15 hours, unpaid | About an hour, with an interviewer watching | Sixty minutes, in the browser |
| Can an AI do it? | Yes, in minutes | Hard, but it tests nerves as much as skill | A company of their own; pasting blocked |
| What you learn | The final answer | Whether they can write a query under pressure | How they work: five things scored across six decisions |
| Your time | An evening per candidate | An hour per candidate, live | A report you read in two minutes |
If you keep a take-home, make it survive AI
Not everyone can switch at once. These changes make a take-home harder to hand to a model:
- Cap it at 90 minutes and say so.
- Give each candidate slightly different data.
- Include a flawed analysis and ask them to critique it, instead of asking for a fresh one.
- Ask how sure they are of each conclusion, and why.
- Spend the follow-up interview on their choices, not their slides: what did you check first, and what would change your answer?
Questions hiring managers ask
Are take-home assignments still worth it for data analysts?
Only if they are short and redesigned for a world where an AI can do the standard version. A long, open take-home now mostly measures who used an AI well.
How long should a data analyst assessment be?
About an hour. Long enough to see real work, short enough that strong candidates with other offers will do it.
Can candidates use ChatGPT on Trial Day?
Not usefully. Each candidate works on a company of their own inside the product, pasting is blocked, and one question asks them to check an AI's analysis.
Is a work simulation fair to candidates?
It takes an hour rather than a weekend, every candidate gets an equally hard company, and they are told before they start what is recorded.
More for hiring managers: Data analyst assessmentSQL test alternativeTrial Day vs TestGorilla