It’s hard not to feel a bit sorry for Bill Michael, who recently resigned as UK boss of KPMG. While he could have softened the blow, there’s little to find disagreement with in his words:

‘After every single unconscious bias training that’s ever been done nothing’s ever improved. So unless you care, you actually won’t change.’

Fighting bias is big business. Over the last three years, diversity and inclusion professionals have enjoyed a surge in demand for their services. In the United States alone, organisations now spend billions of dollars a year on diversity programmes. And it’s a practice which is rapidly making its way across the Atlantic.

One of the most prominent manifestations of this trend is unconscious bias training. As you might have guessed, measuring and combating biases people don’t know they have can be challenging. The first step in any such programme is therefore to show the participant just how far from grace they truly are.

Generally, this means taking the Implicit Association Test, which involves first categorising faces into ‘black’ and ‘white’, then concepts into ‘good’ and ‘bad’, before mixing the two together: the idea is that if, for instance, you associate good things with white people, you will respond quicker to categorise items when ‘good’ and ‘white’ share a key than when ‘good’ and ‘black’ share one. Once the test is taken, the participant is debriefed, given some information on unconscious bias theory, and some suggested actions for reducing its effect on their behaviour.

Keir Starmer pledged to introduce a programme like this for himself as well as ‘all’ Labour party staff after referring to last summer’s Black Lives Matter protests as a ‘moment’. But bias training remains deeply contentious: it has found itself at the centre of a tug of war between Republicans and Democrats over federal funding for diversity training. And in Britain, there has been a tussle between the government, which is scrapping such training for civil servants, and the civil servants union, which wants to know how discrimination will be defeated in its absence.

Predictably, a lot of the controversy over implicit bias training shakes out on party lines, with progressives eager to embrace a tool which shows them what they already know – that everyone is deeply tainted by the original sin of racism and must seek absolution – and conservatives equally keen to reject something they see as a tool for advancing a left-wing agenda. A better, bipartisan objection is that it doesn’t actually work in the way it’s intended to.

Oxford’s Somerville College provides a good example of the usual sort of well-intentioned rationale for the use of these training programmes: in an email to students, the college says there is ‘irrefutable evidence’ that the social ills of group disparities ‘grow, at least in part, from individual unconscious biases that many or all of us have’.

Is this right? Or could it be argued that ethnic disparities aren’t the result of unconscious bias but other factors, not least old-fashioned conscious prejudice?

The real objection to such training anyhow is that, even if it were true that our unconscious biases are to blame for societal ills, there is very little evidence that the sort of interventions usually delivered in these programmes ever result in changes in behaviour, at least outside of a small backfiring effect.

This may partly be because the Implicit Association Test – the magic mirror used to display our true selves – was never intended to be anything of the sort. Tony Greenwald, a co-creator of the test, told Vox that it was ‘only good for predicting individual behaviour in the aggregate, and the correlations are small’. He wrote in an academic paper that the structure of the tests make them ‘problematic’ if used ‘to classify persons as likely to engage in discrimination’.

Looking at the meteoric rise of unconscious bias training, it’s hard not to suspect that its ascent was directly related to its relatively low cost. For businesses and other big organisations, having something to point to when a PR response is needed is highly useful, even if that ‘something’ is a brief quiz administered online. At the same time, it averts the need for more costly interventions examining what – if anything – might need to be done in the way of genuine reform.

When Bill Michael says of unconscious bias that he doesn’t ‘buy it’, he’s far from alone. But, as Michael has found out himself, it’s a view that can prove to be costly.