Can AI actually predict which employees are at risk of leaving, and should we act on it?

Our HRIS vendor is selling us an attrition prediction add-on. It claims to identify employees likely to leave 3–6 months before they hand in their notice, based on engagement scores, promotion history, salary vs. market, and system usage patterns. I'm interested but sceptical — I've seen HR tech overpromise before. What's the real-world accuracy of these models, what data do they actually need to work, and are there ethical concerns I should think through before deploying?