Hiring decisions that hold up when someone asks why

The question that's coming

At some point, a candidate who wasn't selected, a regulator or your own leadership will ask: why did this person get this result? If the answer is "the tool scored them lower", that won't be enough.

Two things are making this more pressing. The EU AI Act treats AI used in employment as high-risk, with a compliance deadline of December 2027. And Canada's proposed Bill C-36 would let people request an explanation of significant automated decisions. It isn't law yet. Both point the same way.

What a defensible process has

A reason tied to the job. Every thing you assess should connect to what the role needs. Write down the link. If you can't explain why a measure matters, drop it.

Consistency. Every candidate gets the same assessment under the same conditions. That's one of the strongest protections against bias, and it also makes results comparable.

Visible reasoning. Results should show what drove them: which skills, which answers, which evidence. A single opaque score is hard to defend.

A person in the loop. The tool informs. A person decides, can see the reasoning, and can disagree with it.

A record. What was assessed, when, by whom, and what changed. Keep it for as long as the law and your policy require.

Checks for fairness. Look at outcomes across groups on a regular schedule, and investigate where the numbers differ.

What to do this month

1. Map your hiring steps. Which are automated, and which are decided by a person?
2. For each measure, write why it matters for the role. Remove the ones you can't justify.
3. Check what a rejected candidate could be told. Try writing the explanation for a real case.
4. Confirm who can override. And that overrides are recorded.
5. Set a review date. A quarterly look at outcomes by group is a good start.