The research on reference checking is unflattering, and most recruiters have never seen it. Knowing how weak the standard version is turns out to be the first step to getting something useful out of it.
Reference checks occupy a strange position on a recruitment desk. Everyone does them, almost nobody believes them, and the research suggests the scepticism is well founded.
In the Schmidt and Hunter meta-analysis, reference checks show a mean validity of about 0.23 for predicting later performance — well below structured interviews and general mental ability, which sit around 0.51.1
That is a weak signal. Not a useless one, but weak enough that a reference should almost never overturn a decision that everything else supports.
Three reasons, and all of them are fixable to a degree.
The candidate chose the referee. You are sampling from a population selected precisely for its willingness to say good things.
The referee is managing risk. In an environment where an unfavourable reference carries perceived legal exposure, the rational move is to confirm dates and say something bland.
The questions invite nothing. "Would you rehire?" has one socially available answer. Asking it produces a yes that means very little.
A reference is not a verification step. It is one more conversation, and it is worth exactly as much as the questions you bring to it.
The move that changes the yield is to stop asking for a verdict and start asking for description. People who will not say "he struggled" will readily describe a situation in which he struggled.
Do it before the offer, not after. A reference taken once the decision is made is a formality, and everyone in the call knows it.
Keep what you learn. The single biggest waste in reference checking is that the content evaporates: the consultant hears something useful about how a person works, and it never reaches the record, so it cannot inform the placement or the next role.
Structured reference capture attached to the candidate, so what a referee said survives the placement.