The retailer should use testing and assessment centres for some vacancies, but not for every one.
The case for the change is strong where the role justifies it. An interview measures how well someone performs in an interview, which is not the same as how well they would do the job; testing and assessment centres put candidates through tasks resembling the real work, so the business sees evidence rather than claims. That produces better appointments, and better appointments matter because a poor one is expensive — wasted training, lost productivity and the cost of recruiting again. Structured assessment is also more consistent, which reduces the influence of interviewer bias and gives the business a documented, job-related reason for each decision.
The case against applying it universally is equally clear. Assessment centres are slow and expensive: they require designing tasks, training assessors and taking managers away from their work for a day. For a large retailer filling many junior positions, that cost multiplied across every vacancy would be considerable, and for a straightforward role the additional accuracy is small — an interview and a reference may identify a suitable shop assistant perfectly well. A long selection process can also deter good candidates, who may accept a faster offer elsewhere.
The deciding factor is the cost of appointing the wrong person. Where a mistake is expensive and hard to reverse — a store manager, a buyer — the extra accuracy easily repays the effort. Where the role is quickly learned and the consequences of a poor appointment are small, the cost of the process exceeds the benefit.
I therefore reject the proposal as stated, because every vacancy is what makes it wrong rather than the method itself. The retailer should introduce assessment for senior and specialist roles while keeping interviews, supported by references, for high-volume junior recruitment. This assumes the retailer can identify which roles are which; if it could not, applying the method to the highest-paid roles would be a reasonable approximation.