How to design a fair take-home hiring assignment

A good take-home assignment is a small simulation of the role, with a bounded time budget and a rubric written before anyone submits. It should answer a question the application could not: can this candidate approach a realistic task, explain choices, and handle constraints? It should not be free production work or a disguised request for a complete marketing plan, feature, or design system.

Short exercises can still be demanding. Budget the work from the candidate's perspective, including reading the brief, producing the artifact, and checking the submission. If the task genuinely takes three hours, do not describe it as 30 minutes. Offer an alternative or accommodation when a candidate needs one, and communicate deadlines clearly.

Write a brief with six parts

Every exercise should state:

1. Situation: What organization or fictional product is involved, who the user is, and what problem matters.
2. Inputs: Any data, code, customer quote, or constraints the candidate needs.
3. Output: Exactly what to submit, in which format, and where.
4. Time: A realistic estimate and the due date.
5. Tools: Whether search, templates, collaborators, and AI tools are allowed; whether use should be disclosed.
6. Evaluation: The criteria and what strong evidence looks like.

Make the context self-contained. If the task asks someone to write copy for an unfamiliar brand, give the target audience, product behavior, and desired action. If those are unknown internally, label the scenario fictional. Do not judge candidates for failing to discover private assumptions.

Example: content and operations intern

Situation: A fictional service helps independent restaurants respond to missed reservation inquiries. Busy owners want more confirmed bookings without extra front-desk work. Your audience is a restaurant owner who has not heard of the service.

Task: In 45 minutes, draft one paid social ad (primary text, headline, CTA, and visual concept), above-the-fold landing page copy (headline, subheadline, CTA, three supporting points), and a short explanation of your chosen customer pain, audience, and one metric you would watch. Submit in a text field or document. AI and research tools are allowed; name any major tool contribution in one sentence.

Evaluation: Customer understanding, clarity of the offer, credible claim, fit between ad and landing page, and the quality of the proposed test. Do not score whether the candidate happened to guess the recruiter's favorite slogan. Do not run the candidate's copy in a real campaign without an explicit separate arrangement.

This task uses a fictional product. If your business is real, replace the claims with an approved product brief so candidates do not invent features.

Example: backend intern

Provide a short function, a failing test, and a reproducible bug report. Ask the candidate to submit a fix, one additional test, and a brief explanation of the cause. State the runtime, allowed dependencies, whether internet or AI tools are permitted, and how to run the test. Score the diagnosis, correctness, test coverage, and explanation separately. Avoid a huge repository that takes most of the time just to install.

For a support role, use a fictional customer ticket and ask for a customer reply plus an internal escalation note. For a product role, use a small anonymized set of feedback and ask for a prioritized question list. The Badal role playbooks provide more starting scenarios, but every brief must be tailored to the job's actual work.

Score the process as well as the artifact

A 0–5 exercise rubric should include anchored criteria. For “diagnosis,” a 0 could mean no diagnosis, 2 an identified symptom without a cause, 3 a plausible cause tied to evidence, and 5 a cause, verification, and meaningful prevention step. For copy, the anchor might distinguish generic benefit claims from a specific customer problem and a testable message. Put these anchors in reviewer instructions before the invitation goes out.

Reviewers should mark missing information rather than infer ability from formatting polish. If AI use is allowed, judge the candidate's reasoning, validation, and ownership. Asking candidates to hide tool use encourages inconsistent disclosure. If tool use is prohibited, explain why that constraint resembles the real job and how you will evaluate compliance fairly.

Decide what advancement means

Before seeing submissions, set a rule for what moves forward, what needs manual review, and what prompts a follow-up question. A candidate with a strong artifact but unclear personal contribution might still deserve an interview. A candidate whose file cannot be opened deserves a request for a readable format, not an automatic judgment about quality. Keep the assignment score distinct from the initial application score; the two assess different evidence.

After the review, send a timely and respectful update. If you decline, avoid a detailed claim about the person that the evidence cannot support. If you advance, explain the next step and expected timing. Our email guide includes templates.

Set this up in Badal

Badal offers assessment templates and a custom assessment specification with title, instructions, and one to eight response questions. Recruiters can define it when creating or updating a role. A candidate receives the exercise version active when they applied, so a later edit does not silently change their task. Qualified candidates receive a private link by email after the application screen, subject to processing capacity and successful delivery. Recruiters see the submitted work and evaluation in the candidate report. See assessment setup and the structured pipeline guide.

Agent-ready brief request

Ask me what this person will do in their first month, who the work serves, and what information a candidate can reasonably know. Propose one exercise that takes no more than [time budget]. Include context, inputs, deliverables, format, deadline, allowed tools/AI, accommodations route, and three to five scoring criteria. Show me the candidate-facing brief and reviewer rubric separately. Ask before attaching it to the role.

The strongest assignment is not the most elaborate. It is the smallest exercise that produces evidence your application questions cannot.

Frequently asked questions

Should candidates be allowed to use AI tools?

Set one clear policy for everyone. If tools are allowed, assess verification, judgment, and the candidate's explanation of choices. If tools are restricted, explain the job-related reason and avoid claiming you can reliably detect hidden use from style alone.

How much feedback should a rejection include?

Give only feedback a reviewer can substantiate and your team can provide consistently. A concise, respectful decision is better than invented detail. If you promise individualized feedback, allocate human review time for it before inviting candidates.