What it tests

SQL fluency, cohort thinking, and ability to translate raw data into a prioritized growth recommendation

Format

1Candidate receives a sanitized dataset from a consumer product: event logs, user signups, transactions, and support flags — typically 3–6 months of data
2Assignment: identify the top retention problem, quantify its impact, and propose the single most impactful intervention
3Deliverable: SQL queries (commented), a cohort analysis, and a one-page written recommendation with projected impact
4Onsite presentation: 20 minutes to walk the growth team through findings, then 30 minutes of live SQL questions and 'what if you're wrong?' probing

What to look for

Do they start with a hypothesis and validate it with data — or just run queries and report what they find?
Is the cohort analysis structured correctly — do they control for acquisition channel, time period, and user segment?
Is the recommendation specific and tied to a mechanism ('users who don't complete X in 48h churn at 3x rate — fix onboarding step 2') or generic ('improve engagement')?
In the live SQL round, can they write clean, efficient queries under pressure — or do they rely on memorized patterns that break on edge cases?

Adaptation guide

Use a synthetic version of your own retention data. Even a fabricated dataset built around your real drop-off patterns will produce far more signal than a generic e-commerce sample.

Full description

Format:

Candidate receives a sanitized dataset from a consumer product: event logs, user signups, transactions, and support flags — typically 3–6 months of data
Assignment: identify the top retention problem, quantify its impact, and propose the single most impactful intervention
Deliverable: SQL queries (commented), a cohort analysis, and a one-page written recommendation with projected impact
Onsite presentation: 20 minutes to walk the growth team through findings, then 30 minutes of live SQL questions and "what if you're wrong?" probing

Time: 3–5 days (take-home)

What to look for:

Do they start with a hypothesis and validate it with data — or just run queries and report what they find?
Is the cohort analysis structured correctly — do they control for acquisition channel, time period, and user segment?
Is the recommendation specific and tied to a mechanism ("users who don't complete X in 48h churn at 3x rate — fix onboarding step 2") or generic ("improve engagement")?
In the live SQL round, can they write clean, efficient queries under pressure — or do they rely on memorized patterns that break on edge cases?

Adaptation: Use a synthetic version of your own retention data. Even a fabricated dataset built around your real drop-off patterns will produce far more signal than a generic e-commerce sample.

Use this template in NouSpark

Start with this assessment template and customize it for your hiring process.

Create Assessment