Base de connaissances
Qu'est-ce que l'évaluation basée sur les preuves ?
Évaluer les candidats avec des preuves, pas des impressions.
Définition
Evidence-linked evaluation is a candidate assessment method where every score given to a candidate is directly tied to a specific piece of evidence — typically a quote from the interview transcript accompanied by an explanation of the reasoning behind the score.
Unlike traditional evaluation methods where interviewers assign scores based on general impressions ("the candidate seemed strong"), evidence-linked evaluation requires a concrete citation for every rating. This creates a verifiable, auditable record of how each hiring decision was made.
Comment ça fonctionne
- Competency framework is defined: Before the interview, the role's required competencies are specified with clear scoring criteria.
- Interview is transcribed: The conversation is captured in real time, creating a complete text record.
- Each competency is scored: For every competency, a score (typically 1–5) is assigned based on the candidate's responses.
- Evidence is attached: Each score must link to a specific transcript excerpt that demonstrates the candidate's performance level.
- Reasoning is documented: An explanation accompanies each score, describing why the evidence supports the given rating.
Pourquoi les preuves sont importantes
Research on interviewer reliability shows that without structured evidence requirements, two interviewers evaluating the same candidate can arrive at significantly different conclusions. This "noise" in hiring decisions leads to inconsistent outcomes and legal vulnerability.
Evidence-linked evaluation addresses this by:
- Reducing subjectivity: Scores are anchored to observable candidate behavior, not interviewer feelings.
- Enabling calibration: When interviewers disagree, the evidence can be reviewed to resolve the difference objectively.
- Supporting compliance: Under the EU AI Act, high-risk AI systems in recruitment must provide transparency about how decisions are made. Evidence-linked scoring satisfies this requirement.
- Improving candidate experience: Candidates can receive specific, constructive feedback based on actual interview content rather than vague assessments.
Évaluation basée sur les preuves vs. traditionnelle
| Evidence-Linked | Traditional | |
|---|---|---|
| Score basis | Specific transcript quote | General impression |
| Reasoning | Documented per score | Rarely captured |
| Auditability | Full trail from score to evidence | No verifiable trail |
| Reproducibility | High — same evidence, same score | Low — varies by reviewer |
| Legal defensibility | Strong | Weak |
Rôle de l'IA dans l'évaluation basée sur les preuves
AI can automate evidence extraction and scoring while maintaining the evidence chain. When AI evaluation uses deterministic parameters (temperature 0, fixed seed), the same transcript and competency framework always produce the same scores — making the process both reproducible and auditable.
The key principle is that AI evaluations should be shown alongside human ratings, never replacing them. The human reviewer always has the final say, but the AI provides a consistent evidence-backed starting point.