The reskilling fallacy
Internal talent marketplaces are producing hard numbers that reskilling programmes cannot match. The BCG/Harvard jagged frontier study proves that blanket AI training is counterproductive.
Internal talent marketplaces are producing hard numbers
The shift from job-based to task-based work allocation is no longer theoretical. Several Fortune 500 companies have deployed internal talent marketplace platforms — primarily Gloat, Eightfold, and Fuel50 — with measurable outcomes that dwarf conventional reskilling ROI.1Internal talent marketplace platforms match employees to projects, gigs, and roles based on skills rather than job titles. The shift from job-based to task-based allocation allows organisations to redeploy existing capacity rather than hiring externally or running broad reskilling programmes.
Unilever launched FLEX Experiences (Gloat-powered) and redeployed approximately 9,000 employees during COVID-19, avoiding layoffs when 25–30% of the business was disrupted. The platform has unlocked over 700,000 hours of capacity and re-prioritised 500,000 worker hours toward 3,000+ business-critical projects, with a reported 41% productivity improvement.
Company-level results: Schneider Electric, Mastercard, Seagate
Schneider Electric found that over 50% of departing employees cited lack of internal growth as their reason for leaving. After deploying an internal marketplace: $15 million in savings from unlocked hours and reduced recruiting costs, 200,000+ hours of employee capacity unlocked, and a Net Promoter Score of 60 — an all-time high.
Mastercard saved $21 million through internal mobility, unlocked 900,000+ hours of productivity, improved employee satisfaction by over 80%, and saw retention rise 30% within the first year.
Seagate saved $1.4 million within four months and reported that women applied at 1.4× the rate of men, with 1.7× greater conversion to assignments — a measurable DEI outcome that no AI-literacy course has replicated.
A critical finding from Prudential Financial's Chief Talent Officer: the central challenge is not technology but managerial mindset. Managers struggle to fractionalise work because they think in roles, not tasks. Gartner reports that over half of HR leaders plan to invest in internal talent marketplaces within three years, and adoption rose from 25% to 35% between 2024 and 2025. Yet only 8% of employees actively use these platforms when available — a culture and management change gap that dwarfs the technology investment.2External hires cost 18–20% more in salary and perform worse for two years (Wharton School). Internal moves fill 30–40% faster (LinkedIn). Employees with higher internal mobility stay 5.4 years versus 2.9 years at low-mobility organisations. McKinsey finds redeployment with reskilling is 20% more cost-effective than layoff-and-rehire.
The jagged frontier demands job redesign, not universal AI training
The most rigorous evidence on human-AI collaboration comes from a pre-registered randomised controlled trial with 758 BCG consultants conducted by Harvard Business School researchers (Dell'Acqua et al., 2023, published in Organisation Science). The study revealed a "jagged technological frontier": for tasks where AI is capable, consultants using AI completed 12.2% more tasks, 25.1% faster, at 40% higher quality. For tasks outside AI's capability boundary, consultants using AI performed 19 percentage points worse than those without it.3Dell'Acqua et al., 2023, Organisation Science. Pre-registered RCT with 758 BCG consultants. Two distinct collaboration patterns emerged: "Centaurs" who maintain clear human/AI task division, and "Cyborgs" who integrate AI interaction continuously into their workflow.
The same tool, opposite results: AI produced 40% higher quality on tasks within its capability boundary and 19-percentage-point degradation on tasks outside it. Training everyone on the same AI tools is counterproductive because each task within a workflow sits differently on the frontier.
The implication for the reskilling playbook is direct: what organisations need is task-level decomposition and redesign — identifying precisely where AI augments and where it degrades human performance.
Microsoft's 2025 Work Trend Index (31,000 workers across 31 countries) introduces the concept of the "Frontier Firm" — organisations structured around human-agent teams rather than traditional functional hierarchies. 46% of leaders report using AI agents to fully automate workstreams. Workers at Frontier Firms report thriving at 71% versus 37% globally.4Deloitte's 2025 Global Human Capital Trends survey (nearly 10,000 respondents) found that 41% of work people do doesn't add real value and only 6% of workers believe their organisation is making great progress in creating value from AI — suggesting massive latent capacity that job redesign could unlock.
An emerging finding from Stanford's Digital Economy Lab (November 2025): using ADP payroll data covering millions of workers, researchers found substantial employment declines for early-career workers (ages 22–25) in AI-exposed occupations like software development and customer support. IDC reports that 66% of enterprises are reducing entry-level hiring as they deploy AI. If junior roles disappear, how do organisations develop the next generation of experienced workers? This question is unaddressed by the reskilling paradigm.
Shorter work weeks work — but only as hour reductions
The evidence on four-day work weeks is strong, specific, and relevant to AI productivity gains.
| Trial | Scale | Key outcome | Sustained? |
|---|---|---|---|
| Iceland (2015–2019) | 2,500 workers, 66 workplaces | Productivity maintained or improved; 86% of workforce gained shorter hours by 2021 | Labour productivity +1.5%/year — highest among Nordic countries |
| UK pilot (2022) | 61 companies, ~2,900 workers | Turnover dropped 57%; burnout down 67%; revenue flat or +1.4% | 89% still operating four-day week at one-year follow-up |
| Belgium law (2022) | First EU legislation | ~1% uptake | Compressed schedule (4×10h) — not an hour reduction |
Belgium's result is the instructive failure: compressing the same hours into fewer days does not address burnout. Reducing total hours does.515% of UK pilot employees said no amount of money would induce them to return to five-day schedules. Physical and mental health improvements were sustained at the one-year follow-up, and 51% of companies had made the change permanent.
The gap that matters: LSE research finds AI users save an average of 7.5 hours per week (~one full workday), worth approximately £14,000 per employee annually. But CEPR/VoxEU analysis reveals that employees in AI-exposed occupations are working longer hours and reporting lower job satisfaction. Without deliberate intervention, AI productivity gains default to increased output demands, not reduced hours.
European institutional infrastructure
German Kurzarbeit saved an estimated 2.2 million jobs during COVID. Employment fell over 13% in the US in Q2 2020 versus only 1.4% in Germany, despite a slightly larger output contraction.6IMK/Hans Böckler Foundation. IMF research shows that without Kurzarbeit expansion, German unemployment would have risen by an additional 3 percentage points and the consumption contraction would have been 2–3 times larger.
Kurzarbeit is designed for cyclical crises, not structural transformation. The adaptation for AI is occurring through collective bargaining. IG Metall's 2024 collective agreement includes a landmark provision: working time can be reduced to 32 hours per week for up to 36 months for transformation-related employment problems, with employer-financed partial pay compensation from month 13.
Company-level collective agreements: Mahle, Hauni, Deutsche Telekom
Mahle secured 10,500 workers' jobs until 2025 with no operational layoffs.
Hauni saved 253 jobs with a four-day week option and a €1 million/year innovation fund governed by an equal management-works council steering committee.
Deutsche Telekom's AI Manifesto, signed jointly by the company and its group-level works council, stipulates that workers must know when they are interacting with AI, workers must be protected against surveillance, and humans remain in control of decision-making.
The European Banking Sector AI Declaration (March 2024) — the first comprehensive sectoral AI agreement in the EU — requires that "humans will be kept at the centre of work organisation" and that employees cannot be subject to decisions based "solely and exclusively on automated variables."
Denmark's Hilfr2 Agreement (2024) between cleaning platform Hilfr and trade union 3F represents the frontier of AI-era collective bargaining: all cleaning assistants classified as employees (not contractors), employer responsibility for algorithmic decisions, a "digital clubhouse" for remote worker organising, and explicit AI transparency provisions.7The German Works Constitution Act was amended in 2021 to extend works council consultation rights specifically to AI-based tools. A January 2024 Hamburg Labour Court ruling clarified that works councils cannot block ChatGPT deployment but retain information and consultation rights. Only 20% of EU trade unions report having a collective agreement directly addressing AI, though 42% are in active negotiations (UNI Europa, 2024).
Swedish Job Security Councils: the gold standard
Swedish Trygghetsråden achieve outcomes that outperform any government retraining programme in the Western world: a 90% re-employment rate, with two-thirds of workers landing positions at equal or higher wages. The model is financed by employers at approximately 0.3% of payroll through collective agreements and is jointly governed by employer organisations and trade union federations.
The critical design: personal guidance counsellors, career coaching, education funding, and self-entrepreneurship assistance — all beginning during the notice period, not after separation.
The strongest causal evidence for wage insurance
A 2024 NBER working paper (Hyman, Kovak & Leive) studied the US Trade Adjustment Assistance programme's wage insurance component — which subsidises 50% of the wage gap between old and new jobs for workers over 50. Eligibility increased employment probability by 8–17 percentage points, raised cumulative earnings by $18,000 (26%) over four years, shortened unemployment spells by roughly one calendar quarter, and showed no decline in job quality. The programme pays for itself through increased tax receipts and reduced UI payments.
References
- Gloat, "Unilever FLEX Experiences Case Study," 2023. Redeployment of ~9,000 employees, 700,000+ hours unlocked, 41% productivity improvement.
- Schneider Electric internal marketplace results: $15M savings, 200,000+ hours, NPS 60. Mastercard: $21M savings, 900,000+ hours, 30% retention increase. Seagate: $1.4M savings in four months.
- Dell'Acqua, F. et al., "Navigating the Jagged Technological Frontier," Organisation Science, 2023. Pre-registered RCT, N=758 BCG consultants.
- Microsoft, "2025 Work Trend Index," 31,000 workers across 31 countries. Deloitte, "2025 Global Human Capital Trends," ~10,000 respondents.
- Stanford Digital Economy Lab, November 2025. ADP payroll data analysis of early-career employment in AI-exposed occupations.
- Autonomy, "Going Public: Iceland's Journey to a Shorter Working Week," 2021. 4 Day Week Global, "UK Pilot Results," 2023.
- LSE, "AI Productivity Savings," 2024. CEPR/VoxEU, "AI and Working Hours," 2025–2026. UC Berkeley, AI efficiency and task loading study.
- IMK/Hans Böckler Foundation, Kurzarbeit employment impact analysis. IMF, "Short-Time Work Schemes During COVID-19," 2020.
- IG Metall 2024 collective agreement: 32-hour provision for transformation. UNI Europa, "Trade Unions and AI Collective Bargaining Survey," 2024.
- TRR (Trygghetsrådet), annual re-employment statistics. Hyman, Kovak & Leive, "Wage Insurance," NBER Working Paper, 2024.
Share this post