Why our bias testing is broad and constant

Our bias testing is broad and continuous.

Why does it matter? While a model may pass all bias tests in its build process, when the same model is used with new candidates (data it has not seen before), it can generate biased scores due to data drift. As highlighted previously, we have bias constraints in the model build process, which prevents any models showing biased outcomes from going live. We don’t stop there, though. Our Discover Insights dashboards – to which customers have constant access – feature continuous bias reporting on live models.

Our bias tests are not limited to candidate scoring. We also test factors such as drop out, time taken, and candidate satisfaction rates across demographic groups to ensure that the experience remains similar for all groups. We conduct comprehensive periodic data reviews with our customers walking through all these stats show how the model is performing.