What does the myth say?
That human screeners are biased and machines are not, so moving screening to an AI makes hiring fairer by construction. The belief is attractive to leaders who want to fix a real problem quickly, and it is dangerous, because it leads to less scrutiny of a process that needs more. It is also the claim most likely to appear in a vendor's pitch and least likely to survive a regulator's question.
The examination is about what a consistent process does and does not fix.
Where does the belief come from?
From the true observation that human screening is inconsistent: the same application scored differently by two people, or by one person on Monday and Friday. From the intuition that a rule applied by a machine is applied without prejudice. And from vendors who sell unbiased as a feature, because it is what buyers want to hear. The belief is a reasonable inference from a real problem, drawn one step too far.
The step too far is assuming that the rule itself is neutral.
What is true in it?
That a Recruiter Digital FTE on ftes.ai applies the scorecard the same way to the first and the three-hundredth applicant, at any hour, without fatigue or mood, and shows the evidence behind every score. That this removes a large source of unfairness, the inconsistency, and makes the remaining sources visible, because every score can be traced to a criterion and a passage. That a process which shows its reasoning can be audited in a way an unrecorded human judgement cannot. Consistency and transparency are real gains, and they are the honest claim.
What is not true is that consistency is the same as fairness.
What does the mechanism actually do?
The hiring manager writes the scorecard; the FTE applies it to every application, scoring each criterion with the passage that supports it, marking hard requirements as evidenced, not evidenced, or awaiting verification, and flagging applicants the scorecard captures poorly rather than ranking them. Shortlists are proposed; a person decides who is interviewed. Rejections and offers are decided and communicated by people. The scorecard is versioned, so an audit can read which criteria applied when. Decisions about people are on the platform's never-automated list at every autonomy level; the only class that can earn autonomy is interview scheduling.
If a criterion is biased, the FTE applies it consistently, and the audit is what catches it.
- Criteria: yours, written, versioned
- Scores: evidence per criterion, challengeable per line
- Decisions: human, always; scheduling is the only autonomous class
Why is the audit the real guarantee?
Because the only way to know whether a process produces unfair outcomes is to look at the outcomes. Each quarter the head of talent samples shortlists, compares who advanced across groups, opens the scorecards for the samples, reads which criteria drove the results, changes the criteria where a pattern appears, and records the audit with the scorecard version. The FTE makes this possible: every score is traceable, the scorecard is versioned, and the audit trail shows every shortlist. A human process without records cannot be audited this way, which is the honest comparison.
The library's piece on Recruiter FTE bias and fairness describes the audit in detail.
How can you check for yourself?
Ask any vendor, including this one, whether their screening is unbiased. Distrust a yes. Ask whether every score shows its evidence, whether the scorecard is yours and versioned, whether shortlists can be challenged per criterion, whether any decision about a candidate is ever automated, and how an audit is run. Then run one: take last quarter's shortlists, compare outcomes across groups, open the scorecards, and see whether a criterion explains a pattern. If you can do that in an afternoon from the record, the process is auditable, which is the property fairness actually depends on.
The GDPR guide in the library lists what the data protection lead will ask alongside this, and the two reviews are best run in the same quarter.