Output-Based Hiring
Why Israeli founders pay candidates to do test projects (and you should too)
Why interviews don't work
Decades of research, going back to Schmidt & Hunter's 1998 meta-analysis, are unanimous: unstructured interviews predict job performance at about 0.14 correlation. That's barely above random. The 'gut feel' you have at the end of a candidate interview is almost meaningless data.
Structured interviews do better (0.51 correlation). Cognitive assessments do better (0.51). Reference checks do worse than people assume (0.26). Work samples - actually watching the candidate do work that mirrors the job - score 0.54. The highest of any hiring signal.
Israeli founders learned this not from the research. They learned it from mandatory military service, where everyone gets sorted by demonstrated capability, not by interview charm. The lesson sticks: output beats credentials, every time.
What a paid test project looks like
After the initial screen, you offer the top 2-3 candidates a paid 4-hour test project. The project mirrors actual work the role will do. Not a brain teaser. Not a hypothetical case study. Real, current work that you'd otherwise do internally.
Pay them their normal hourly rate. If you're hiring a $90K/year role, that's roughly $50/hour, so the project costs you about $200. This is the cheapest selection signal you'll ever buy.
Give them clear deliverables, a 1-week deadline, access to whatever Slack channels and documentation they'd have in the role. Let them ask questions. Watch how they ask, what they ask, how they prioritize when they only have 4 hours.
At the end of the project, you have a tangible artifact: the work they actually produced. That's the basis for the hire decision, not the interview vibe.
What to watch for
Quality of output, obviously. But also: how clearly did they ask clarifying questions vs assume? Did they push back on the scope when it was wrong, or just execute the wrong thing? Did they meet the deadline, or ask for an extension?
How they communicate matters as much as what they produce. A great candidate sends a 5-bullet summary of what they did, what they considered, what they would have done with more time. A weak candidate dumps the file and asks 'is this what you wanted?'
Most importantly: did the work get better as you'd expect a hire to get better? In a 4-hour window, you can't see year-3 performance. But you can see: do they ask the right questions, prioritize the right things, communicate clearly under time pressure?
Common objections
'Won't great candidates refuse a paid test project?' Rarely. When you're transparent ('we pay $X for 4 hours, we use this to make a decision based on actual work'), serious candidates appreciate the rigor. Candidates who refuse are usually candidates you don't want.
'Isn't this gaming the system - getting work for cheap?' If you're doing it right, no. The project is artificial enough that the output isn't usable as production work. It mirrors the work but uses safe, non-production data. You're paying for the selection signal, not for free labor.
'What about senior hires?' Especially for senior hires. A senior candidate has the most professional pride; they want to demonstrate. The project is structured differently (e.g., a strategic doc instead of a tactical artifact), but the principle holds.
Key takeaways
- →Unstructured interviews predict performance at 0.14 - barely above random
- →Paid 4-hour test projects score 0.54 - the highest of any hiring signal
- →Pay candidates their normal hourly rate. ~$200 for a $90K role.
- →Watch how they ask clarifying questions, communicate under pressure, deliver against deadlines
- →Skeptical candidates are usually candidates you don't want