A less biased system
Unless you're starting from a position in which you have the perfect workforce - both in terms of performance and human representation, an algorithm that prefers those who are similar to "candidates who have already proved to be good at the job" is inherently biased in favour of the status quo. And how do you get the comparator data? Do you re-interview your existing staff for their current positions and threaten to dismiss them if they fail? If not, it's hardly a like-for-like comparison.
It sounds very much like Amazon's "people who bought X bought Y so, having bought X you will like Y, too" - and I'm sure we've all had experience of just how helpful that is.
Sill this is Unilever, so I'm presuming the main criterion for employment is* the inability to tell Stork from Persil.
*Or was, apparently spreads have been sold off for underperforming. Presumably they joined the company before the AI technology was available. Unfortunately, it also robs me of the opportunity to say of the AI - "I can't believe it's not better". Shame...