Skip to content

Hiring a Marketer: Work Samples That Predict Performance

How to hire a growth marketer using work samples instead of interview questions: fair paid trials, an AI-era scoring rubric, and India-specific hiring signals.

26 Aug 202611 min read
  • Hiring

If you want to know how to hire a growth marketer in 2026, stop optimising your interview questions and start designing better work samples. Interview questions are now trivially preparable with an LLM: a candidate can rehearse polished answers to every standard question in an evening, which quietly destroys their predictive signal. A well-designed, fairly-paid work sample is the only screen I have found that reliably separates people who can do the job from people who can talk about it.

I have hired and worked alongside marketers on both the agency and in-house side of Indian startups, including growth work at Masai School. This is the process I actually use, plus the rubric. If you are a candidate rather than a hirer, read it anyway, knowing how you will be evaluated is a legitimate advantage.

Key Takeaways

  • Standard interview questions have lost most of their signal; assume every answer was prepped with AI.
  • Work samples predict performance far better, but only if they mirror the real job.
  • Pay for anything over two hours. Unpaid multi-day tasks select for desperation, not talent.
  • Assume AI was used in take-homes. Evaluate editing and judgment, not raw drafting.
  • Use a written rubric before you see the first submission, or you will rationalise your gut.
  • In India, verify the ubiquitous "I ran ads" claim with specific, unfakeable questions.
  • Agency and in-house backgrounds signal different things; neither is better in the abstract.
You are not evaluating what they can produce. You are evaluating what they choose to ship.

Why the Interview-Question SERP Is Now Wrong

Search "how to hire a growth marketer" and you will get lists of questions: "Tell me about a campaign that failed." "How would you grow our top of funnel?" "Walk me through your attribution model."

These were decent questions in 2021. In 2026 a candidate can generate strong, structured, plausible answers to all of them in about fifteen minutes. That is not cheating. It is what any sensible person does. But it means the interview now measures preparation and articulacy, not capability. Those correlate weakly.

What still works in a live conversation

Not the prepared question: the unprepared follow-up. Ask a normal question, listen to the polished answer, then go three layers deeper into a specific decision inside it.

  • "You said you cut that channel. What was the number that made you decide?"
  • "Who disagreed with you, and what was their argument?"
  • "What would have changed your mind?"
  • "What did you get wrong before you got it right?"

Rehearsed answers are wide and shallow. Real experience is narrow and deep. Three follow-ups usually reveal which one you are talking to.

Designing a Work Sample That Actually Predicts

A good work sample has five properties.

1. It mirrors the real job

If the role is 70% content and community, do not test their media-buying spreadsheet skills. Sounds obvious; violated constantly. Write down what the person will spend their first 90 days doing, then test a slice of that.

2. It is bounded to two to four hours

Longer tasks do not produce more signal. They produce self-selection by who has free weekends: which filters out working parents, people already employed full-time, and anyone with a commute. That is not a talent filter.

3. It uses real context, not sanitised context

Give them your actual situation: real numbers (redact what you must), real constraints, the real awkward bit. Sanitised briefs produce sanitised answers.

4. It asks for reasoning, not just output

Require a short rationale section. This is where the actual signal lives now. Output is cheap; reasoning is not.

5. It is paid

Anything beyond about two hours should be paid at a fair rate. This is a fairness issue and a quality one, paid tasks get better effort and let you ask for real work rather than a token exercise.

Three work samples I have found predictive

The channel-diagnosis task. Hand over three months of real (anonymised) channel data and ask: what would you kill, what would you double, what would you need to know before deciding? Tests judgment under ambiguity, the core growth skill.

The edit test. Give them an AI-generated draft of something in your category and ask them to make it publishable. This is the single best sample for 2026 hiring because it isolates exactly the skill that matters now: knowing what is wrong with competent output.

The 30-day plan. "You joined Monday. Budget is X. Here is what we know. What do you do in your first 30 days and what will you have proven by day 30?" Tests sequencing, prioritisation and realism about what can be learned quickly.

Write the rubric before you read the first submission. Otherwise you will just rationalise your gut.

Evaluating a Take-Home When They Obviously Used AI

They used AI. Nearly everyone does now, and the ones who did not are often slower without being better. Treat AI use as a given, not a violation.

What you are evaluating is what they did after the model produced its draft.

Signals of good AI use

  • Specificity your model would not know. Real details about your market, your competitors, your actual funnel, this only comes from research they chose to do.
  • Cuts. Strong candidates ship shorter than the model does. Generic sections are removed, not padded.
  • A defensible point of view. LLM output hedges by default. A clear "I would not do X, and here is why" is a human decision.
  • Named uncertainty. "I would need your CAC by channel before committing to this" is a senior move models rarely make unprompted.
  • Voice consistency. They rewrote in their own register rather than shipping model cadence.

Signals of bad AI use

  • Comprehensive coverage of everything with no prioritisation
  • Confident claims about your business they could not possibly know
  • Frameworks recited rather than applied
  • No numbers, or invented ones
  • The distinctive tidy-listicle rhythm, untouched

Ask about it directly

Add one line to your brief: "Use whatever tools you like, including AI. In your rationale, tell us what the tools did and what you changed." Candidates who answer this well are demonstrating exactly the orchestration skill you are hiring for. Candidates who claim they used nothing are usually the weaker hire.

The Scoring Rubric

Fill this in for every candidate, independently, before you discuss them with anyone else. Discussion first is how one confident voice sets the whole panel's opinion: structured, independent scoring is one of the few hiring practices with consistent support in the practitioner literature, including HubSpot's own writing on marketing team hiring.

DimensionWhat you are testingWeak (1–2)Strong (4–5)Weight
PrioritisationCan they choose?Covers everything equallyNames the one thing that matters and defends it25%
Reasoning qualityIs the logic sound?Frameworks recitedArgument builds from your actual constraints20%
NumeracyDo numbers mean anything to them?Numbers absent or decorativeCorrect maths, sensible assumptions, stated ranges15%
Editing / tasteCan they fix competent-but-bad?Ships model output lightly touchedCuts hard, rewrites hooks, kills the generic15%
RealismDoes the plan survive contact?Ignores budget, team, timeScoped to actual resources, sequenced15%
CommunicationCan a founder act on this?Long, unstructuredTight, scannable, decision-ready10%

Reading the scores

  • Below 2.5 average: pass, regardless of how good the interview felt.
  • 2.5 to 3.5, capable executor. Fine for a role with a strong manager above them.
  • 3.5 to 4.2, can own a channel and a number with light supervision.
  • Above 4.2, can own strategy. These are rare; move fast when you find one.

The disagreement rule

If two reviewers score more than 1.5 apart on any dimension, that is worth a conversation: usually you were testing different things without realising. Resolve the definition, not the candidate.

Hiring in India Specifically

Most hiring content is written for US companies. A few things are genuinely different here.

Compensation: honest ranges only

I am not going to publish precise salary tables, because I do not have rigorous survey data and the numbers you see quoted elsewhere are usually recycled guesses. What I can say honestly:

  • City matters a lot. Bengaluru, Mumbai, Gurgaon and Hyderabad sit meaningfully above tier-2 cities for the same role. Remote-first roles have compressed this gap but not closed it.
  • Stage matters as much as city. A funded Series A company and a bootstrapped agency will quote very different numbers for identical job descriptions.
  • The biggest single multiplier is outcome ownership. Marketers who own a number command substantially more than marketers who own a channel's execution.
  • Ranges are wide. For any given title in India, the honest spread between the low and high end is often 2–3x. Treat every band you read, including any I imply here, as approximate and variable.

Benchmark against actual offers your network has seen in the last six months. Anything older is stale.

Agency vs in-house background

BackgroundUsually strong atUsually weak atBest fit
AgencySpeed, breadth, working across categories, client comms, shipping under deadlineLong-horizon ownership, internal politics, retention and lifecycle depthEarly-stage, needs many things tried quickly
In-houseDepth on one funnel, cross-team work, compounding assets, stakeholder managementAdapting to unfamiliar categories, raw shipping speedPost-PMF, needs one channel taken seriously
FreelanceSelf-direction, breadth, commercial realismWorking inside a team structure, longer feedback loopsContract-to-hire, or a first marketing hire with strong founder support

Neither agency nor in-house is better. Match the background to the shape of the problem.

Verifying "I ran ads"

Almost every marketing CV in India says this. Most of the time it means they had dashboard access. Ask questions that only a real operator can answer.

  • "What was your monthly spend, roughly, and who approved increases?"
  • "What was your CAC at the start and at the end, and what actually moved it?"
  • "Which creative won, and what was your hypothesis for why?"
  • "Tell me about a week where performance dropped. What did you check first?"
  • "Who did you have to convince to change the budget, and how?"

People who ran ads answer these in specifics within seconds. (If you want a primer on what those metrics should look like before you interview, Search Engine Journal covers paid fundamentals reasonably well.) People who watched someone run ads go abstract immediately. This is not a trick. It is just asking for the texture of real work.

Reference checks still work

Underrated in Indian startup hiring because everyone assumes references are friendly. Ask one question: "If you were starting a company tomorrow, would you hire them again, and for what?" The "for what" is where the honest answer hides.

Independent scores first, discussion second. Reverse that order and you get groupthink.

For Candidates Reading This

If you are on the other side of this process, three things.

Do the work sample properly, but time-box it. Going far beyond the brief does not read as enthusiasm; it reads as poor prioritisation, and prioritisation is 25% of the score.

Lead with the reasoning section. Most candidates treat it as an afterthought. It is where the highest-weighted dimensions are actually assessed.

Say what your tools did. Then say what you changed and why. As of 2026: with Forbes and others covering AI's displacement of production work through August, the ability to direct and edit AI well is a stated requirement in a growing number of Indian marketing roles, not a thing to hide. The related "missing generation" discussion in the Content Marketing Institute and MarTech coverage is essentially about this: the market now needs editors and orchestrators more than drafters.

Decline unfair processes. A four-day unpaid task with no rubric and no feedback tells you something real about how that company will treat you as an employee. It is fine to ask for scope, timeline and whether it is paid.

Frequently Asked Questions

Are interview questions still useful for hiring marketers?

Only as a starting point for follow-ups. Prepared answers to standard questions carry little signal now. The unscripted third and fourth follow-up into a specific decision is where the information is.

Should a marketing work sample be paid?

Yes, for anything over about two hours. Unpaid long tasks filter for free time rather than skill, and paid tasks let you request realistic work instead of a toy exercise.

How long should a marketing take-home be?

Two to four hours of genuine effort. If you cannot get signal in four hours, your task is badly designed, not too short.

What if the candidate clearly used AI on the take-home?

Assume they did and evaluate accordingly. You are testing their editing, prioritisation and judgment, not their typing. Ask them to declare what the tools did and what they changed.

How do I evaluate a growth marketer with no formal experience?

Use the same work sample. A self-taught candidate with a documented self-initiated project frequently outperforms an experienced candidate whose experience was mostly execution. The rubric does not care about years.

What salary should I offer a growth marketer in India?

Ranges vary widely by city, funding stage and whether the role owns a business number: often a 2–3x spread for the same title. Benchmark against recent real offers in your network rather than published tables, and weight outcome ownership heavily.

Is an agency or in-house background better?

Neither in general. Agency backgrounds suit early-stage companies needing speed and breadth; in-house backgrounds suit post-PMF companies needing depth on one funnel.

How do I verify someone actually ran paid campaigns?

Ask for texture: monthly spend, who approved increases, what moved CAC, which creative won and why, what they checked first when performance dropped. Operators answer in specifics; observers go abstract.

How many candidates should reach the work-sample stage?

Three to five. More than that and you will not score carefully; fewer and you have no comparison set to calibrate the rubric against.

Should I share the rubric with candidates?

Sharing the dimensions (not the weights) improves submission quality and signals a serious process. It does not let people game it, you cannot fake prioritisation and numeracy by knowing they are being measured.


I write about growth, hiring and organic marketing for Indian startups at younusfardeen.com: practitioner notes rather than theory, including the hires and processes I got wrong. Worth a look if you are building a team.