We Replaced Take-Home Tests With 3-Day Paid Work Trials: 7 Steps and the 2 Mistakes That Cost Us a Senior Hire

We ran take home exercises for years. They were cheap, they scaled, and they were quietly useless for the only decision that mattered: whether this person would be good to work with on a hard week. So we replaced them with a three day paid work trial at final stage. The results were better almost immediately, and then we lost a senior candidate we badly wanted because of two process errors that were entirely ours. This is the method as it stands now, including both mistakes.

Why the Take Home Test Stopped Earning Its Place

The take home has three structural problems and none of them are fixable by writing a better prompt.

It measures solitary work on a clean problem, which is not the job. It is trivially delegated or generated, which in late 2026 means a strong submission tells you progressively less about the submitter. And it is unpaid, which selects for candidates with spare evenings rather than candidates with the best judgement. In a market like Dubai, where a large share of strong engineers are already employed on full schedules, that last point quietly removes your best prospects from the pool before you ever speak to them.

A paid work trial inverts all three. The work is real, collaborative and compensated. It is also considerably more expensive and more intrusive, which is exactly why it belongs at the end of a process rather than the beginning.

Step 1: Decide Whether a Trial Is the Right Instrument

Run a trial when the risk you are buying down is judgement under ambiguity: a senior or lead role, a first engineer, a specialist whose domain you cannot assess internally, or anyone who will own a system for years. Skip it when the risk is competence at a known task, which an interview exercise already covers, and skip it for junior roles where you are hiring potential and a trial just penalises inexperience.

A useful test: write down what you expect to learn that your interviews did not tell you. If you cannot name two specific things, you do not need the trial. If the answer is how they handle a codebase they did not write, how they behave when a requirement is underspecified, or whether they ask for help at the right moment, the trial is the correct instrument.

Step 2: Settle the Permit and Contract Questions Before You Invite Anyone

This is the step most teams discover halfway through, and it is the one with real consequences in the UAE. The question is never whether work trials are permissible in general. It is whether this candidate can lawfully be paid for this work right now. Three situations cover almost everyone:

• Already employed in the UAE. Their existing contract and their employer’s position govern what outside work they can accept. Many contracts restrict it. Ask the candidate directly and let them decide what they are comfortable confirming.
• Holds their own freelance permit. The clean case. They invoice you, you pay the invoice, and the arrangement is a straightforward service engagement.
• Neither of the above. A short written service agreement is the usual route. Take advice rather than assume, because an arrangement that looks like unlicensed employment creates exposure for you, not for the candidate.

Alongside the payment question, two documents must exist before any repository or data access: an IP assignment covering anything produced during the trial, and a confidentiality undertaking. Both are short. Neither is optional, and retrofitting them after someone has already read your code is not a repair. For the longer form of the classification question, see our guide on contractor versus employee classification for UAE developers.

Step 3: Price the Trial at a Real Day Rate

Pay what the person would command as a contractor at their seniority. In Dubai in late 2026 that is broadly 1,200 to 2,000 AED per day for mid level engineers and 2,000 to 3,500 AED per day for senior and specialist profiles, with AI and data specialists at the top of that range or above it. Benchmark against your own band rather than against these figures alone; our method for building that band is in creating a developer salary benchmark for Dubai.

Three rules make the payment work as intended. Pay on completion regardless of outcome, state that in the invitation, and pay the invoice the same week. A trial contingent on a positive result is not a paid trial, and candidates recognise the difference instantly.

The economics are not close. Three senior days at 3,000 AED is 9,000 AED to de risk a decision worth multiples of a monthly salary, against a mis hire at that level which costs you the salary, the visa processing, the onboarding, the team disruption and the months before you admit it. We have never had a finance conversation about trial cost that survived that comparison.

Step 4: Scope a Task That Produces Signal in Three Days

The task has four properties, and the fourth is the one people get wrong:

1. Real, but off the critical path. It should be work you genuinely want done, so the candidate is not solving a puppet problem, but nothing should break if it is never delivered.
2. Inside your actual codebase, or a faithful slice of it. The point is to observe someone navigating code they did not write, with your conventions and your rough edges.
3. Underspecified on purpose, in one named place. Leave one requirement genuinely ambiguous. What the candidate does about it is frequently the single most predictive thing you learn.
4. Impossible to finish perfectly. You are buying observation of choices under constraint, not a completed feature. If a strong candidate can finish it comfortably, you have scoped a test rather than a trial.

Step 5: Write the Scorecard Before the Trial Starts

Agree four to six observable dimensions and write down what strong and weak look like for each. Ours are: navigation of unfamiliar code, handling of the ambiguous requirement, quality and intent of questions asked, judgement about scope, communication of trade offs, and the state in which the work was handed over.

Circulate the scorecard to the panel before day one and have each reviewer score independently before anyone discusses it. Our full rubric design method is in building a technical interview scorecard for UAE AI hiring.

Step 6: Run the Three Days as a Working Relationship

Give the candidate a named point of contact, one scheduled fifteen minute check in per day, and explicit permission to ask questions. Then watch how they use it, because how someone uses access to a colleague is a large part of the signal. The candidate who asks one sharp question on the morning of day one and then works is usually stronger than the candidate who asks nothing for three days and delivers something subtly wrong.

Practical points that matter more than they sound. Have accounts, repository access and a working local environment ready before the candidate starts, because a trial where day one is lost to environment setup measures your onboarding rather than their ability. Tell the team that a trial is running and what the person is working on. And if the candidate is doing this around an existing job, agree the three days as non consecutive or as evenings across a week.

Step 7: Debrief Against the Scorecard Within 24 Hours

Collect written independent scores, then meet. Decide inside a day, because the candidate is almost always in another process and your trial has just made you their most engaged employer; losing that position to a slow debrief is a self inflicted wound.

Then do the two things that cost nothing and that almost nobody does. Pay the invoice that week whatever the outcome. And give the candidate specific feedback against the dimensions, including when the answer is no. Candidates who are turned down well refer other candidates, and in a market this small that compounds faster than any sourcing channel. If you need to handle a competing offer at this point, our method is in handling developer counteroffers in Dubai.

Mistake Two: the Task That Was Secretly on the Critical Path

The second error was subtler and it changed our behaviour as the employer rather than the candidate’s. We scoped a trial around a piece of work that we had quietly started to need. Nobody said so, but the reviewers knew the sprint was depending on it.

The effect was immediate. We stopped assessing judgement and started tracking delivery. When the candidate paused on day two to raise a design concern, which was the correct call and precisely the behaviour we claimed to want, it read to the panel as a delay. We had converted our own assessment into a deadline and then marked the candidate against it.

The rule we now apply without exception: if any part of the team would be disappointed by the task not shipping, it cannot be a trial task. Pick something you want but can lose.

FAQ: Paid Work Trials for Developers in Dubai