Cost the coding round first
The case for any technical screening stage is arithmetic, so do the arithmetic before choosing a format. A coding round costs more than its hour. The engineer reads the CV, runs the session, writes feedback, and loses some focus either side. Call it two engineer-hours per candidate as a working assumption, and adjust it to your own loop.
| Scenario, assuming 20 candidates and 5 who would pass | Coding rounds | Screen hours | Engineer-hours total |
|---|
| No screen | 20 | 0 | 40 |
| 30-minute screen that removes 10 who would have failed | 10 | 10 | 30 |
| Same screen, but 2 of the 10 it removes would have passed | 10 | 10 | 30, and only 3 hires possible |
The third row is why the question is harder than it looks. The cost of a screen is not only the hours it takes. It is the people it wrongly turns away, and those never appear in any number you track.
What good technical screening measures
The design that works is a small sample of the real interview. Its strength is that the screen measures the same thing the coding round gates on, at lower fidelity and lower cost. A screen that measures something else is a second filter with its own errors, not a cheaper version of the first.
So derive the screen from the loop. If the coding round asks candidates to change real-ish code with tests, the screen might ask how they would test a function they are describing, or what they would check first when a change breaks an unrelated test. If the loop's design round cares about trade-offs, one screening question can ask for a trade-off the candidate made recently and what it cost. Three or four questions is enough. Each needs a written note of what a passing answer contains, so two different screeners reach the same verdict.
Who runs it matters. A recruiter with a script can confirm hard requirements such as right to work, location and band, but cannot judge whether a technical answer is good or merely fluent. An engineer or a technical hiring manager can follow up, and a follow-up is what tells memorised answers from understood ones. Make sure that whoever runs the screen has had the same calibration as the loop interviewers. The interview loop design question covers that calibration in detail.
The cheap filters that measure nothing
Several technical screening methods save engineer time because they barely measure anything. Filtering on university, previous employer or CV keywords is fast, but none of those is the thing the coding round tests. It also narrows the pipeline in ways you cannot see. Unproctored written questions sent by email were a reasonable filter once. They are now easy to answer with an AI assistant, so a fluent written answer tells you less than it used to, and a screen built on them should be read with that in mind.
Moving the cost onto the candidate
This is where most proposals go wrong, and it is the trap the question is set for. The quickest way to save engineer time is to make the candidate spend theirs: a three-hour take-home or an automated online assessment before any human contact. On the team's spreadsheet that looks free.
It is not free. Candidates with several options are the ones best placed to decline an unpaid exercise from a company that has not yet spoken to them, so the screen's cost falls hardest on the people you most want. You also lose the chance to sell the role: the screen is often the candidate's first conversation with an engineer, and a well-run one is part of how you hire. A screen that saves your engineers one hour by spending three of the candidate's is a transfer, not a saving.
What an algorithm-first hiring manager proposes
A manager whose whole model of hiring is the algorithm round tends to answer this question by moving that round earlier and automating it: send every applicant a timed online puzzle and interview whoever scores highest. It feels rigorous, and it saves engineer hours. But it filters on puzzle practice, which may not be what the loop's coding round measures at all. It puts the cost on the candidate. And it cannot follow up, so it cannot tell understanding from rehearsal. If the coding round itself is practical, the automated puzzle is a screen for a different job.
Knowing whether it works
To judge your technical screening, track two numbers per quarter: the share of screened candidates who reach the coding round, and the share of those who pass it. A good screen raises the second without crushing the first. If the first collapses, the screen is too strict or measuring the wrong thing. Treat any change in pipeline volume alongside the sourcing picture in filling three senior roles in a quarter.
To get partial visibility of false negatives, occasionally let a borderline screen result through to the coding round and record what happens. A handful of borderline passes that go on to clear the loop tells you the screen's line is in the wrong place.
A screen earns its place only if it measures what the coding round measures, more cheaply. Count the candidate's hours and the people you wrongly turn away, or you will save engineer time by buying a worse pipeline.
© 2026 Preptima. Originally published at preptima.com (opens in a new tab).