The Hiring Tool That Learned Your Company's Old Habits

Imagine you build a hiring tool and train it on ten years of your company's hiring decisions. What will it learn to look for?

Not the best candidates. Your past decisions — including every shortcut, preference, and blind spot your hiring managers ever had. If your company historically hired graduates from the same five universities, the tool learns those universities signal quality. If certain neighborhoods, names, or career gaps correlated with rejection, the tool quietly learns those too.

In plain English: the tool doesn't learn what good looks like. It learns what familiar looks like.

This is how bias gets laundered into objectivity. A hiring manager's gut feeling is visible and arguable. A model's score looks scientific — same bias, better disguise, applied to thousands of applications in seconds.

The safeguard is embarrassingly simple: before you trust any AI hiring aid, test it. Run last year's applicants through it and compare who it advances across groups. If the pattern mirrors your history instead of improving on it, you haven't built a hiring tool. You've built a photocopier for old habits. Audit the pattern before you automate it.

I explain frameworks; I don't give legal counsel.