Labour market institutions for the AI era: The need for verified employment records
As artificial intelligence transforms labour markets, verified experience increasingly matters as much as formal credentials. Workers who can demonstrate how they have adapted across roles, applied skills in different contexts, and navigated technological change gain advantage over those with equivalent formal qualifications but no verifiable track record. Yet most economies lack the institutional infrastructure to document, verify, and make portable the employment histories that now determine labour market outcomes.
The result is a paradox: countries leading in AI innovation often possess the weakest institutional capacity to translate productivity gains into inclusive growth. This infrastructure gap poses mounting risks as AI reshapes occupational structures at unprecedented speed.
In a recent report, my co-authors and I run a systematic comparative review of employment record systems across eight advanced and emerging economies – the UK, Germany, Australia, New Zealand, Brazil, Estonia, India, and South Africa – along with EU-level initiatives (Levy Yeyati et al. 2025). Three patterns emerge consistently:
real-time reporting embedded in existing workflows,
stable personal identifiers enabling longitudinal tracking, and
worker access to verified records.
The research also identifies recurring implementation failures where misaligned incentives undermined data quality despite technical sophistication.
The employment data paradox
Modern economies generate vast amounts of employment data through tax systems, social insurance, payroll platforms, and licensing boards. The problem is fragmentation, not scarcity. In the US, for example, wage records flow through state unemployment insurance systems designed for compliance, not workforce intelligence (Groshen et al. 2022). Key variables – occupation codes, hours worked, job tenure, workplace location – are inconsistently collected or absent.
This fragmentation imposes measurable economic costs. Imperfect information, stemming from incomplete skill visibility, inflates hiring frictions and delays labour reallocation (Carranza et al. 2022). Employers rely more heavily on educational pedigree or informal networks as screening mechanisms, disadvantaging mid-career workers whose accumulated skills are less visible in formal records (Stanton and Thomas 2015). Workers consequently face a ‘visibility tax’ – their accumulated skills fail to convert into wage offers when they cannot be credibly verified.
In AI adoption scenarios, informational asymmetries are likely to amplify inequality. Workers with strong, verifiable credentials or reputations capture disproportionate productivity gains, while other workers face longer re-entry spells – not because they lack adaptable skills per se, but because labour markets cannot observe or credibly verify their work histories and capabilities (Bolte et al. 2020, Lukác and Grow 2021).
What works: International evidence
Table 1 presents our comparative overview of national cases.
Table 1 National cases comparison
Source: Levy Yeyati et al. (2025).
Three design principles emerge from this benchmarking exercise.
Real-time reporting embedded in existing business processes minimises compliance burden while improving timeliness. The UK's Real Time Information system requires employers to submit payroll data with each pay cycle through existing software (HM Revenue & Customs 2017), and Australia’s Single Touch Payroll achieved similar results.
Stable identifiers enable longitudinal tracking and cross-system integration. Germany’s Integrated Employment Biographies create daily-level employment histories for 80% of the workforce using social insurance numbers (Schmucker and Vom Berge 2025), and India’s e-Shram portal extends social protection to 300 million informal workers through Aadhaar linkage.
Worker empowerment through data access builds trust and increases accuracy. Brazil’s Digital Labour Card gives workers app-based access to official employment histories, and Estonia’s Employment Register automatically notifies workers when their records update (Kattel and Mergel 2018).
Taken together, international evidence points to a common pattern: the most successful systems set national interoperability standards while allowing local implementation, scale iteratively rather than all at once, embed reporting into payroll systems rather than add compliance layers, protect privacy by design, ensure workers can access their own verified records, and align incentives carefully to avoid misclassification or exclusion.
International experience also reveals recurring implementation failures. South Africa's Sector Education and Training Authorities collect employer-reported skill shortages to guide training investments, but incentive misalignment undermines data quality: firms list vacancies tied to existing training programmes rather than forward-looking needs (Matha and Jahed 2024). Brazil's simplified tax regime for micro-entrepreneurs enabled widespread employee misclassification as contractors (Alvarez 2023).
These cases highlight a critical lesson: technical infrastructure alone does not guarantee success. Effective systems require aligned incentives, continuous stakeholder engagement, and careful sequencing. Germany's decades-long development succeeded because mandatory reporting was embedded in social insurance from inception. Australia's phased rollout allowed iterative refinement.
Privacy and governance present equally complex trade-offs. Employment records enable powerful longitudinal analysis but create surveillance risks if implemented without safeguards. The EU's GDPR and digital wallet architecture demonstrate how interoperability can coexist with privacy-by-design: credentials support selective disclosure, allowing workers to prove attributes without exposing complete histories.
Employment records as experience credentials
The rise of AI fundamentally alters the balance between formal credentials and verified experience in labour markets (Levy Yeyati 2024). Traditional credentials – diplomas, licenses, certificates – signal knowledge acquisition at a point in time. Employment records, by contrast, document how capabilities deploy across real contexts: what workers did, for whom, where, when, and with what outcomes. They capture not static qualification but dynamic adaptability.
Recent analysis shows AI adoption disproportionately affects junior positions, making technological change ‘seniority-biased’ and increasing returns to accumulated, verifiable experience (Lichtinger and Hosseini Maasoum 2025). Workers who can demonstrate successful transitions across roles and sectors – a retail worker who moved to logistics, a manufacturing technician who learned robotics maintenance – possess valuable signals that static credentials cannot capture.
Without verified employment records, this experience remains invisible. A worker with a decade adapting to technological change across three industries cannot prove that adaptability to prospective employers or training providers. By reverting to educational pedigree as a screening mechanism, the market disadvantages workers with demonstrated practical experience and flexibility rather than formal qualification. Employment record infrastructure transforms experience into legible, portable credentials – the institutional foundation for experience-based labour markets.
Employment record infrastructure functions as general-purpose technology for labour markets: foundational architecture enabling multiple applications rather than a solution to specific problems. Like payment systems or communications networks, value scales non-linearly with adoption. Isolated pilots improve matching marginally; interoperable systems transform labour market transparency and training returns.
This infrastructure character carries implications for policy sequencing. Comprehensive systems ultimately require clear governance frameworks, sustained investment, and coordination across jurisdictions. Germany's Federal Employment Agency harmonises data through national protocols while preserving local operational autonomy. The EU's distributed approach achieves alignment through shared standards rather than centralised control.
The COVID-19 crisis exposed the costs of inadequate employment data infrastructure. Governments struggled to identify displaced workers, target support, and track labour market recovery in real time – predictable consequences of fragmented systems designed for compliance rather than crisis response.
As AI diffusion accelerates, this infrastructure deficit becomes more consequential. Workers need verifiable records to demonstrate adaptability; employers need reliable data to identify talent beyond credential proxies; policymakers need real-time visibility to target interventions. Yet the binding constraint is not technical but institutional: modern employment records require shared governance frameworks – interoperability standards, recognised verifiable issuers, and liability rules – rather than centralised databases.
Whether verified employment data becomes the institutional backbone for experience-based labour markets will depend less on technological innovation than on governance capacity. Countries that build this infrastructure may convert technological change into productivity gains and inclusive growth by enabling workers to carry verified experience across jobs, sectors, and careers. Those that maintain fragmented compliance systems risk concentrating AI-driven returns among workers with visible credentials while others face prolonged displacement – not from skill deficits, but from invisible experience.
References
Bolte, J, N Immorlica and M O Jackson (2020), “The Role of Referrals in Inequality, Immobility, and Inefficiency in Labor Markets”, arXiv:2012.15753v1
Carranza, E, C Garlick, K Orkin and N Rankin (2022), “Job Search and Hiring with Limited Information about Workseekers’ Skills”, American Economic Review 112(11): 3547–3583.
Groshen, E, D Nightingale, A Reamer, Y Magdy and M Raju (2022), "Harnessing Employer Records for Enhanced Research, Statistics, and Evaluation", JEDx & U.S. Chamber of Commerce Foundation.
HM Revenue & Customs (2017), Real Time Information Programme: Post Implementation Review Report, UK Government.
Kattel, R and I Mergel (2018), "Estonia's Digital Transformation: Mission Mystique and the Hiding Hand", UCL Institute for Innovation and Public Purpose Working Paper 2018-09.
Levy Yeyati, E (2024), “Why gen AI can’t fully replace us (for now),” 18 December, Brookings Institution.
Levy Yeyati, E, J Camisassa and I Seyal (2025). "What Works for Employment Records: International Practices and Implications for the United States", Brookings Institution.
Hosseini Maasoum, S M and G Lichtinger (2025), “Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data”, available at SSRN.
Lukac, M and A Grow (2021), “Reputation systems and recruitment in online labor markets: insights from an agent-based model”, Journal of Computational Social Science 4: 207–229.
Matha, P and M Jahed (2024), "A Critique of the Techniques Used by the Sectorial Education and Training Authorities to Detect Skills Shortages in South Africa", Administratio Publica 32(4): 156-175.
Mirelle Alvarez, B (2023), "Shifting Paradigms: The Consequences of Misclassifying Employees as Entrepreneurs", Sao Paulo School of Economics, Fundação Getulio Vargas.
Schmucker, A and P Vom Berge (2025), "Sample of Integrated Labour Market Biographies (SIAB) 1975-2023", FDZ-DATENREPORT, Research Data Centre of the Federal Employment Agency.
Stanton, C T and C Thomas (2015), “Landing the First Job: The Value of Intermediaries in Online Hiring”, The Review of Economic Studies 83(2): 810–854.
