5 Interview Techniques That Actually Predict Job Performance

Industrial-organizational psychology has been studying what predicts job performance for over a hundred years. The results are sobering for anyone who has spent time perfecting their interview questions.

General mental ability tests predict job performance at r=0.51. Structured interviews: r=0.51. Work sample tests: r=0.54. Unstructured interviews — the kind most companies conduct: r=0.20. Reference checks: r=0.26. Years of experience: r=0.18. Educational credentials: r=0.10.

The things most hiring processes weight most heavily are, by and large, among the least predictive. Here are the five techniques that actually work.

1. Work Sample Tests

The most predictive single method. Give candidates a realistic sample of the actual work.

For a content strategist role: draft a brief for a real campaign challenge. For an engineer: debug a real piece of code from your codebase (anonymized if necessary). For a sales role: run a mock discovery call with a prepared objection script.

The validity is high because the signal is direct: you're not predicting whether someone can do the job. You're watching them do it.

The caution: work sample tests must be compensated for senior candidates, carefully scoped to take less than two to three hours, and directly relevant to day-one work — not invented exercises that test abstract problem-solving disconnected from the actual role.

2. Structured Behavioral Interviews

Already covered extensively elsewhere on this blog, but worth restating: behavioral interviews that ask about specific past situations, scored against predefined criteria, by multiple independent evaluators, are among the most reliable tools available for roles where you can't easily construct a work sample.

The key is the structure. An unstructured behavioral interview — asking behavioral questions but without scoring rubrics or standardized evaluation — drops from r=0.51 to something much closer to the unstructured mean.

3. General Cognitive Ability Assessments

Intelligence and learning speed predict job performance across virtually all roles and levels. The correlation is strongest for complex roles with significant novel problem-solving requirements.

The practical challenge is implementation: standalone IQ-adjacent tests are increasingly viewed negatively by candidates, carry adverse impact risk across demographic groups, and can create legal exposure if not validated for the specific role. The better approach is using cognitive assessments as one component of a structured evaluation, not as a single filter, and ensuring your assessments have been validated and tested for disparate impact.

4. Structured Reference Checks

Most reference checks are theater. You call names provided by the candidate, ask softball questions, and receive glowing responses from people the candidate has specifically selected. This is not useful information.

Structured reference checking is different. Use a consistent question set that asks for specific behavioral examples — not "Was Rajan a good leader?" but "Can you describe a specific situation where Rajan had to navigate significant disagreement within his team? What did he do?" Ask for quantitative performance context where relevant: "Where would you rank Rajan relative to other people at his level you've worked with?"

The most reliable reference is someone not on the candidate's provided list — a manager from a previous role who you identified independently. Many candidates don't push back on a recruiter calling their previous employer directly if asked thoughtfully.

5. Situational Judgment Tests

SJTs present candidates with realistic work scenarios and ask how they would respond. Unlike work sample tests, they don't require the candidate to actually execute — they measure judgment about what should be done.

This makes them particularly useful for roles with significant interpersonal, ethical, or management complexity that's difficult to directly sample. A well-designed SJT for a customer success role might present a scenario where a client is threatening to churn over a product issue that isn't actually the company's fault — and measure the candidate's judgment about how to navigate the situation honestly while preserving the relationship.

SJTs require significant development investment to do well. Off-the-shelf versions are widely available but rarely validated for specific roles. Custom development is worth it for high-volume roles where even small improvements in prediction translate to meaningful outcomes at scale.

---