Western HR is solving the wrong problem

The mainstream response to AI workforce disruption — reskilling programmes, AI-literacy certifications, and wellbeing apps — is contradicted by a growing body of empirical evidence. Rigorous studies show that individual-level interventions produce negligible outcomes, while structural changes to work design, governance, and the employment relationship itself generate measurable, durable results.

Meanwhile, companies that redesigned work structures — decomposing jobs into tasks, building internal talent marketplaces, creating human-AI collaboration architectures — report 3–5× lower costs than external hiring and 30–57% improvements in retention. The divergence between what Western HR invests in and what the evidence supports is no longer a matter of opinion. It is measurable.

The evidence falls across two angles.

The reskilling fallacy

Internal talent marketplaces at Unilever, Schneider Electric, Mastercard, and Seagate are producing hard numbers — $15–21 million in savings, 30–57% retention improvements, hundreds of thousands of hours unlocked — that no AI-literacy certification programme has demonstrated. The BCG/Harvard "jagged frontier" study (758 consultants, pre-registered RCT) proves that blanket AI training is counterproductive: the same tool that produces 40% quality improvement on one task produces 19-percentage-point degradation on the next. What matters is task-level decomposition, not universal upskilling.

European institutional infrastructure — German Kurzarbeit, IG Metall's 32-hour transformation provisions, Swedish Job Security Councils with their 90% re-employment rate — provides governance mechanisms for workforce transition that market-based systems lack. The four-day work week evidence is strong, but only when designed as hour reductions, not compressed schedules.

The trust deficit

The three most rigorous studies of workplace wellbeing interventions — Fleming's Oxford study (N=46,336), the BJ's/JAMA RCT (160 worksites), and the Illinois Wellness Study (ruling out 84% of prior positive findings) — converge on the same conclusion: none of them work. What produces measurable outcomes is autonomy, psychological safety, manager quality, and non-toxic culture — structural factors that require changing the work, not the worker.

Trust is the binding constraint, and surveillance destroys it. With 42% of monitored employees planning to leave and 75% reporting lost trust, the $1.4 billion monitoring industry is generating negative ROI by its own metrics. Companies that chose the opposite path — Costco (94% retention), Patagonia (6–7% turnover), Mondragon (97% cooperative survival rate) — demonstrate what mutual investment looks like in practice.

The uncomfortable conclusion

The problem is not that workers lack AI skills. The problem is that organisations lack the structural willingness to redesign work, share productivity gains, preserve human agency, and rebuild trust.

The EU is building the first comprehensive regulatory framework for AI in employment — the AI Act, Platform Work Directive, and emerging algorithmic management legislation. Where enforcement exists, measurable protections follow. Where it does not, legislation produces what researchers call null compliance.

The evidence for what works exists. The question is whether institutions will act on it before the window closes.