Our approach
Bias is never eliminated, but it can be measured and reduced. We do this in three ways: by limiting the inputs the model can use, by auditing score distributions across roles, and by giving recruiters explanations they can challenge.
Practical safeguards
- Demographic attributes are excluded from inputs.
- Score distributions are reviewed for unexpected gaps.
- Every score has a breakdown a human can interrogate.
- Recruiters can flag a score they believe is unfair, which feeds into our model review process.