TL;DR

Retraining is the answer every government reaches for when asked what happens to displaced workers. A new review, written by the independent researcher David Roodman with Anthropic’s Maxim Massenkoff, pooled 56 randomised American trials alongside European experimental work and found the effects real but small. Their conclusion is that these programmes, as they currently exist, would not cope with displacement at scale.

What the trials actually show

Offer someone a training place and, on average, their odds of holding a job improve by between two and three points, worth roughly $1,000 a year in earnings. Each place costs roughly $13,000. Once extra tax receipts and lower benefit payments are counted, the state gets back over half of that, and the programmes come close to paying for themselves. This is a defensible use of public money. It is not a mechanism for absorbing a labour market shock.

One category does much better. Programmes built around a specific high-demand industry, working with employers who then hire the participants directly, deliver gains several times the average. The complication is that attempts to reproduce them have repeatedly failed — meaning the thing that works is the thing nobody has reliably managed to copy.

Why this matters here

Britain’s version of this debate is the Growth and Skills Levy, which rests on the same unexamined premise: that displaced workers can be moved into new roles quickly enough, and that the constraint is funding rather than evidence about what works. This review says the constraint is evidence. The rigorous studies point at employer-linked, sector-specific placement, which is materially harder to administer than classroom provision and does not scale by simply spending more.

The timing is pointed. The ILO put 6.1% of young workers in the roles most exposed to AI this week, and UK conveyancers, insurers and accountants have all in recent days described entry-level tasks as the first thing automated — which is where people traditionally learn a trade before needing retraining at all.

Looking forward

The authors’ recommendation is to run and rigorously measure a rapid expansion of one promising programme now, while displacement is still limited, rather than designing policy during a crisis. That is a sensible ask and an unusual one to hear from a company whose products are the reason the question is being asked.