Candidate Assessment: What to Measure and How to Score It
Choose job-related candidate assessments, define scoring before testing, check AI outputs and give applicants clear instructions and accommodation routes.

Candidate assessment is the process of collecting and evaluating evidence about whether an applicant can meet a role’s requirements. It can include a work sample, structured interview or standardized test—not just an online assessment sent by a recruiter.
The useful question is not “Which test produces a score?” It is “What hiring decision will this evidence support?” A polished report does not, by itself, establish that someone can do the job.
Match the assessment to the requirement
Start with the work, then choose the method. The U.S. Office of Personnel Management (OPM) recommends designing assessments around critical competencies identified through job analysis. It distinguishes reliability, or consistency of scores, from validity, or the relationship between assessment performance and job performance. A test can be consistent without measuring something useful for the role. OPM assessment strategy guidance
| Method | Evidence it can provide | What to check |
|---|---|---|
| Work sample or simulation | Performance on a task resembling the job | Does it test skills required on entry, or knowledge the employer plans to teach? |
| Structured interview | Past behavior and reasoning about job-related situations | Are questions and rating standards consistent across candidates? |
| Cognitive or job-knowledge test | Reasoning abilities or relevant knowledge | Is the tested ability necessary for this particular role? |
| Personality or integrity assessment | Traits or dispositions intended to inform predictions about behavior | What evidence connects the result to the role and intended use? |
These distinctions follow OPM’s work-sample guidance, its structured-interview guidance and the EEOC’s descriptions of employment tests.
No method is automatically suitable. OPM specifically cautions that work samples may be inappropriate when the tested activities will be taught after selection.
Build a small, traceable assessment plan
1. Define the requirement in observable terms
Replace “excellent communicator” with a task and standard: “Can explain a billing error accurately, acknowledge the customer’s concern and identify the next action.”
Separate essential entry requirements from trainable skills. This helps avoid rejecting applicants for not knowing an internal system they have never used.
2. Collect evidence without creating unnecessary work
For a fictional customer-support role, a compact assessment could combine:
- A response to a simulated billing complaint.
- A structured interview question about handling incomplete information.
Use invented customer details and a supplied policy excerpt. Set a clear scope rather than asking candidates to solve a live business problem.
Each stage should add evidence the previous stage did not collect. Another test measuring the same thing needs a reason, not merely an available vendor feature.
3. Set scoring standards before reviewing candidates
For the billing exercise, an illustrative rubric might assess:
- Accuracy: Uses the supplied policy correctly and makes no unsupported promises.
- Resolution: Identifies the appropriate next step or escalation.
- Communication: Gives a clear, respectful explanation the customer can follow.
Define what insufficient, acceptable and strong performance look like for each criterion. Decide in advance whether a serious accuracy error is disqualifying or instead requires follow-up.
OPM’s structured-interview model uses predetermined questions and the same rating scale and acceptable-answer standards. That structure makes the basis for comparison explicit. OPM structured interviews
4. Record evidence, not just totals
Keep the response, criterion ratings, supporting observations and decision rationale. “Offered a refund the policy prohibits” is more useful than “weak judgment.”
If reviewers disagree, resolve whether they saw different evidence or interpreted the standard differently before averaging their scores. See how to diagnose conflicting candidate ratings.
Review outcomes as well as individual ratings. The EEOC recommends examining equally effective, less discriminatory alternatives when a selection procedure screens out a protected group, and updating assessments when job requirements change. EEOC testing guidance
When AI contributes to the assessment
Hiring technology can score résumés, administer computer-based tests or support video interviews. Those are different uses, not one interchangeable “AI assessment.” DOJ hiring-technology guidance
Ask the employer or vendor to identify:
- Input: Answers, task outputs, transcripts or other data used.
- Output: A summary, competency score, rank or recommendation.
- Consequence: Whether the output informs review or blocks advancement.
- Evidence: Validation for the role and purpose, plus known limitations.
- Review route: Who can investigate missing data, processing errors or disputed results.
Do not interpret an 82/100 score as an 82% chance of job success unless that meaning is explicitly defined and supported. A score’s label, scale and validation determine what it can justify.
Under U.S. federal guidance, vendor documentation does not transfer responsibility away from the employer: the EEOC says employers remain responsible for ensuring their tests are valid under the Uniform Guidelines. EEOC testing guidance
What candidates should confirm before starting
Ask for the expected duration, tested skills, permitted resources—including AI tools—scoring approach, technical-help contact and next steps. Use these seven pre-assessment questions to clarify the instructions.
If the format creates a disability-related barrier, contact the employer about an accommodation before beginning where possible. Under the U.S. ADA, covered employers must provide reasonable accommodations during hiring to otherwise qualified applicants with disabilities unless doing so would create undue hardship. Tests should measure relevant skills rather than unrelated sensory, manual or speaking impairments. DOJ guidance
For example, additional time may not solve an inaccessible interface; an accessible version or another appropriate adjustment may be needed.
Legal references concern U.S. federal requirements, checked October 9, 2026. State, local and non-U.S. rules may add obligations. This is informational, not legal or employment advice; confirm applicable requirements with qualified counsel.