Marketing Analyst Take-Home Interviews: Three Briefs, One Framework

If you're interviewing for a marketing analyst role at a DTC brand, a SaaS company, or any performance-marketing team, the take-home is probably the round you'll remember. It's usually a 4 to 8 hour exercise. They send you a dataset or a hypothetical scenario; you send back a deck or doc with your analysis and recommendations. It's designed to answer one question the recruiter screen and the behavioral round can't: can you actually do the work?

The problem is that most candidates over-index on the technical part (the SQL query, the chart, the regression) and under-invest in the part that actually wins the offer: *how you frame the decision.*

Here's the pattern that works across almost every marketing-analyst take-home.

The three-part answer structure

Every good take-home response has the same bones:

• Context. What's the actual business question? What are you assuming about the goal, the constraint, and the decision-maker?

• Analysis. What did you measure, how, and what came out? This is the math plus the chart.

• Recommendation. What should the team do next, with what confidence, and what do you want to test or learn next quarter?

A weak candidate spends 80% of the deck on analysis. A strong candidate spends roughly 20% on context, 50% on analysis, 30% on recommendation. And the recommendation has to be specific enough to act on Monday morning.

Let's apply this to three briefs that show up constantly.

Brief 1: "Here are 12 months of transaction data. Tell us about our customers."

This is the cohort-retention flavor. Raw data is usually one row per transaction: customer ID, date, revenue, product, maybe channel. The trap is to over-produce the dashboard: retention heatmap, LTV by cohort, product-category correlations, seven slides of charts.

The stronger move:

• Pick one clear question in context. For example: "Is our 2026 acquisition cohort worth what we're paying to acquire them?" That focuses everything.

• Show three numbers that answer it: 90-day retention rate for the 2026 cohort, projected LTV, and the implied LTV:CAC ratio (or call out in context that you need CAC data to complete the answer).

• Give a single recommendation. Either "raise CAC ceiling to $X", or "this cohort isn't paying back, so cut Meta spend by Y% and reinvest in retention email." With a specific test to run next.

What reviewers score you on: did you ask the right business question, or did you just dump charts?

Brief 2: "Here's our channel mix and spend. Recommend an optimization."

This is the media-mix flavor. Expect a sheet with Meta, Google, TikTok, email, maybe affiliate. Spend, impressions, clicks, conversions, revenue for each.

The weak version: compute ROAS per channel, rank them, tell them to cut the lowest.

The stronger version:

• Call out that ROAS-based ranking is attribution-blind. It systematically penalizes top-of-funnel. Reframe the question as "which channel should we test next quarter, not which one is best today."

• Show contribution alongside efficiency for each channel. Meta has 35% of conversions at $2.40 ROAS; the niche affiliate program has 4% of conversions at $6 ROAS. The right move depends on headroom, not rank.

• Propose one specific test. "Hold Meta flat, scale the affiliate program 2x for 4 weeks, measure incrementally via geo-holdout, decision threshold is 10% lift in net-new customers."

What reviewers score you on: did you recognize that last-click is lying, and did you propose a real measurement plan rather than just shuffling budget?

Brief 3: "Here are the results of an A/B test we ran. Did it win?"

This is the experimentation-review flavor. They'll give you conversion rates, sample sizes, and often a p-value or confidence interval that a senior analyst has already computed. Sometimes deliberately wrong.

The weak version: look at the p-value, say "yes/no it's significant."

The stronger version:

• Ask in context what decision the test was meant to inform. Was this a binary ship / don't-ship, or a learning test? The bar for "winning" is different.

• Recompute the confidence interval from raw conversions and sample size. Flag if sample size is too small, if the test ran too short (day-of-week effects), or if there's a novelty-effect risk.

• Recommend: "Ship with a monitoring plan", or "Don't ship, the effect is within noise", or "Extend the test another 2 weeks to hit power."

What reviewers score you on: do you understand what statistical significance means in practice, and can you translate it into a product decision?

What reviewers are actually looking for

Three things, in rough priority order:

• Business judgment over technical flash. A well-framed question beats a fancy SQL window function.

• Clear, confident recommendations. Hedging is the fastest way to lose a reviewer's confidence.

• Showing your thinking, not just your output. Write down the assumptions you made. Show the intermediate numbers. Reviewers want to see how you reason, not just whether you arrived at the same answer they did.

If you can turn in a deck that makes a senior marketer think "yeah, I'd hire this person to own a channel," you're there. Everything else is table stakes.