A fast-scaling product team was losing strong engineers at the test stage — abstract puzzle rounds felt irrelevant, and completion sank below 55%.
They replaced timed brain-teasers with a real-codebase task in a GPT-enabled workspace, so candidates worked the way they actually do on the job.
For high-volume engineering hiring across India — campus drives and lateral pipelines — the assessment stage is where the funnel quietly leaks. When the test doesn't look like the job, strong candidates disengage and drop off.
The problem: a leaking funnel
With thousands of applicants per role, even a small drop in completion compounds into hundreds of lost candidates. Abstract, timed puzzle rounds filtered for test-taking speed, not engineering ability — completion had fallen below 55%.
- Candidates abandoned tests that felt irrelevant
- Strong builders underperformed on artificial constraints
- Recruiters spent days chasing incomplete submissions
The shift: assessments that mirror the job
Candidates opened a real codebase in a GPT-enabled workspace — files, tests, and AI access — and solved a task resembling day-one work.
- Real repositories, not whiteboards
- AI allowed and fully logged, mirroring modern workflows
- Tasks scoped to the actual role
The outcome
Completion climbed from 55% to 88%, assessment-stage drop-off fell 46%, and the team reached a confident shortlist roughly three times faster — without widening the top of the funnel.
When the assessment respects the candidate's time and reflects the real job, more people finish, and the signal you keep is stronger.