What is the honest claim about bias?
The honest claim about bias in AI screening is that it cannot be eliminated, only made visible and correctable. Any process that ranks people encodes assumptions, whether those assumptions live in a human recruiter's instincts, in a keyword filter, or in a scorecard.
A vendor claiming an unbiased screening system is claiming something no one can demonstrate. What can be demonstrated is consistency, transparency, and human accountability for every decision, which is what a Recruiter FTE is built around.
Why does scoring against your scorecard matter?
Scoring only against the scorecard you set moves the question of fairness to the criteria, where it can actually be examined. The Recruiter FTE applies your criteria with your weights; it does not bring its own notion of a strong candidate.
That means if a criterion is producing rankings you would not defend, the fix is in the scorecard and it is visible. A criterion that rewards a particular career path or institution can be seen doing so and removed.
How does evidence per score help?
Evidence per score lets a hiring manager challenge a ranking at the level of a single criterion rather than accepting or rejecting the whole thing. Each score shows what in the application produced it, so a disagreement is specific.
This is also what makes the process auditable later. A ranking without reasons cannot be reviewed; a ranking with the evidence per criterion can be reconstructed and questioned months afterwards.
- Criteria and weights: yours, visible, changeable
- Application of criteria: the same for every applicant
- Evidence: shown per score, from the application itself
- Decisions: always human, recorded by name in the audit trail
Why are decisions always human?
Rejections, offers, and anything that changes a person's outcome are permanently gated because the cost of those decisions is asymmetric and because accountability for them must sit with a person. The Recruiter FTE ranks and drafts; a named human decides, and the audit trail records who.
This holds at every autonomy level. Interview scheduling can earn autonomy because it changes no outcome; a decision about a candidate never does.
How should you audit it?
Audit the process by sampling shortlists periodically and comparing outcomes across candidate groups, then looking at which criteria drove any differences. Because every score has evidence and every criterion is yours, the audit is a review of the scorecard rather than an investigation of a black box.
Record the audit and what changed as a result. That record, alongside the named-decision audit trail, is what a fair-process review would ask for, and having it is more defensible than any claim of neutrality.
What does a fairness audit look like in practice?
A fairness audit in practice is a quarterly reading of the record rather than a special investigation. Sample the shortlists from the quarter, compare outcomes across the candidate groups you are obliged or choose to monitor, and, where differences appear, open the scorecards to see which criteria drove them. Because every score carries the evidence it rested on and every criterion is one you wrote, the question becomes whether a criterion is defensible, and that is a question the hiring managers and the people team can answer together.
Record the audit, the criteria examined, and any changes made to the scorecard, and keep it with the named-decision audit trail that shows a human made every rejection and offer. That record is what a fair-process review asks for, and it is more defensible than any claim of neutrality, because it shows the process being examined and corrected rather than asserted to be free of error. A Recruiter FTE makes the audit cheap because the evidence is already attached; the organization still has to do it.
Keep the scorecard itself under version control in the remit, with the date and reason for each change. When an audit finds a criterion that drove an unwanted difference and the scorecard is changed, the record shows what was changed, when, and why, and shortlists before and after the change can be compared separately rather than blended.