Check whether the AI treats every group fairly. On real business cases, with fresh numbers every time.

• Work out selection rates by group and the approval gap
• Find the error rate and false positive rate gaps
• Measure the disparate impact ratio