The Engineering Hiring Process That Actually Predicts Performance

Two years ago, we started tracking the correlation between our hiring process scores and actual job performance at six months. The results were uncomfortable: our whiteboard algorithm questions had essentially zero predictive validity. The engineers who passed them weren’t outperforming those who barely passed them. We were filtering for interview skill, not engineering skill.

What We Replaced It With

Work samples. Real work, paid, time-limited. We give candidates a specific problem that closely mirrors the actual work they’d be doing in the role. They work on it asynchronously over three to four days. We review the output, discuss the decisions they made, and assess how they think about tradeoffs.

This approach is more expensive to design and review. It’s also dramatically more predictive of actual performance.

The Structured Interview

For behavioral assessment, we use structured interviews with consistent questions across all candidates and a rubric for evaluating responses. This reduces interviewer-to-interviewer variance and makes the process more defensible. Unstructured interviews — “tell me about yourself” and wherever the conversation goes from there — are essentially noise.

What We Learned

The engineers who performed best in our old process and the engineers who perform best in our new process are not the same people. That’s not a statement about either group — it’s a statement about how poorly designed most hiring processes are at their actual job.