Prove reuse where one kind of work is hired repeatedly, before building the network.
Fromuniversal network→Toone capability family
An open network for human capital can’t be proved all at once, and it shouldn’t be built all at once. Its value comes from reuse, where evidence created in one transaction lowers the cost of the next, and reuse needs several employers who value the same work. So the unit of build is a capability family rather than a population. That means one kind of observable work, a small consortium of employers with paid demand, an accountable intermediary, and measurement in money rather than registrations.
Registrations are not a network
It is tempting to measure an open human-capital network the way platforms are often measured: learners registered, courses listed, profiles created. Those numbers describe activity. The value this architecture bets on is different. It comes from reuse, when evidence produced in one transaction lets a second employer skip part of its own verification (Shift 2). Reuse only happens where several buyers value the same capability and trust the same kind of proof.
A national registry of profiles doesn’t create those conditions. A capability family in which several employers hire repeatedly, and agree on what good work looks like, does. In the examples the review found, coordination among many parties in human capital relies on an anchor, such as a state agency, a national skills body, a large employer or a sector intermediary. The review found no evidence that an open market of this kind starts itself.
The network forms where one kind of work can be proven once and trusted twice.
The capability family is the unit
A capability family is a cluster of related roles that share tasks, tools and a way of judging the work. The review’s candidates for a first market are task families with visible output, such as software quality assurance, data operations and design production.
The evidence points to this unit from three directions. Work samples pay off most when competencies have to be present from the first day and the output can be judged against the job itself [1]. Small credentials add up best inside progression ladders that an industry has defined: Cedefop’s mapping of certifications in manufacturing and retail found some stacked into ladders that lead from entry-level jobs toward technician and engineering roles [2]. And sector-focused training programs produced their largest and most durable earnings gains by concentrating on specific, well-paid sectors with strong employer links [3].
| Criterion | Why it matters |
|---|---|
| Output can be observed and judged | Evidence and net contribution can be measured |
| Competence is needed at entry | This is where work samples earn their cost [1] |
| Several employers hire repeatedly | Reuse needs a second buyer |
| Employers can agree on a rubric | Evidence can be compared across organisations |
| Confidentiality barriers are low | Evidence can leave the employer that produced it |
| The work pays a real premium | The effort is worth it for workers [3] |
| Training and supervision exist locally | Formation is possible, not just assessment |
Start with demand
The first market shouldn’t be built around learners in the hope that jobs arrive later. The review proposes a small employer consortium with real paid projects or vacancies. That changes the design immediately.
- Real demand
- Defined capability requirement
- Worker
- Paid work
- Verified evidence
- Second employer’s decision
Every step is real, so every step can be measured. Without the second employer’s decision, reuse can’t be shown. Without paid work, relevance can’t be shown. And without cost accounting, surplus can’t be shown.
The first market tests economics. It doesn’t collect registrations.
Go deeper
What the first market needs, and what it doesn’t
The review separates the minimum viable architecture into what the first test needs, what may help but is unproven, and what should wait.
| Must have | Useful but unproven | Later |
|---|---|---|
| One capability family with observable output | A limited capability graph for prerequisites and provenance | Autonomous personal agents |
| An employer consortium with paid projects or vacancies | Several independent trainers | Agent-to-agent negotiation |
| Work samples or paid projects that produce evidence | Reputation that is portable across participating employers | Universal interoperability |
| A transparent rubric with independent or cross-employer verification | An employer subscription for verified talent | Continuous discoverability |
| An accountable intermediary for contracts, supervision, IP, worker protection and disputes | Outcome-linked payment to trainers or assessors | Revolving human-capital finance |
| Worker consent, correction, selective disclosure and export | Public first-loss funding or an outcome contract | A cross-sector, cross-border network |
| Human-approved matching with clear payment terms | ||
| Full measurement of employer cost, output, rework, retention, pay and intermediary cost |
Everything in the right-hand column may matter later. None of it is needed to learn whether the mechanism works, and each item adds privacy, governance, legal or financial risk before there is anything to protect.
Who pays in the first market
The employer consortium pays for what it values. Whether it will pay enough to cover evidence and coordination is itself a result to measure. Public or philanthropic money funds the experiment. Workers don’t pay to be tested, and every apprenticeship is paid. The intermediary’s cost sits inside the test rather than being assumed away. Shift 9 follows the money in full.
The metrics that matter
Reuse rate is the share of evidence records that a second employer accepts without repeating its own test. It is the number the whole architecture depends on.
Cost per qualified hire compares the new workflow with the employer’s usual one, including the cost of producing and verifying evidence.
Net contribution per apprentice is accepted output minus supervision, training, infrastructure, coordination, rework and management (Shift 4).
Time to productive work, worker search hours and worker pay show whether workers gain, not only employers.
Willingness to pay asks whether any participant keeps paying fees after the pilot ends.
Two tests, run separately
The first market carries two different bets. One is that evidence of real work can be reused. The other is that learning inside paid work can cover its own supervision. Both can run in the same capability family, but they should be measured separately, each with its own kill signal, so that a strong result in one doesn’t hide a weak result in the other. The review’s planning sequence looks like this.
| Order | Test | Duration | Planning range (USD) | Decision it informs |
|---|---|---|---|---|
| 1 | Evidence increment and employer willingness to pay | 6–9 months | 250,000–600,000 | Build the evidence product, narrow it, or drop the graph and network claims |
| 1 | Net contribution of knowledge-work apprentices | 6–12 months | 500,000–1.2 million | Keep employer-led apprenticeship, add an intermediary, restrict tasks, or drop distributed apprenticeship |
| 2 | Demand-originated discovery | 6–12 months | 300,000–700,000 | Keep opt-in discovery, limit it to recruiter assistance, or abandon it |
| 2 | Predictive validity of assessment | 12–24 months | 300,000–800,000 | Build validity-linked reputation or keep ordinary assessment |
| 3 | Training-to-work conversion | 12–18 months | 400,000–1 million | Separate a work-linked pathway from generic learning |
| 3 | Network surplus with full cost accounting | 18–24 months | 1–2.5 million | Expand, stay a sector intermediary, or drop the open-network thesis |
| 4 | Progressive financeability | 3–5 years | 1–3 million, including a first-loss reserve | Keep finance external, add a narrow guarantee, or build a finance layer |
These are planning ranges for pilot budgets, not measured prices.
What could go wrong
- A thin market.Too few employers may hire from the family often enough for evidence to find a second buyer.
- The wrong family.Output may turn out to be too ambiguous, too dependent on the team, or too confidential to travel.
- Anchor dependence.The consortium may never become a market, leaving a subsidised program rather than a network.
- Replication.A playbook proven in one capability family may not transfer to the next, with its different rubrics, supervision needs and employers.
- Transfer across economies.Most of the evidence comes from formal labour markets in the United States and Europe. Where much work is informal and employers have no HR systems, evidence may matter more and be harder to verify.
How we’d test it
| The test | What we would measure | What would end the idea |
|---|---|---|
| Family screen | Candidate families scored against the criteria above, with employers, before launch | No candidate family combines repeated hiring, observable output and agreement on a rubric |
| Evidence increment (Experiment 1) | Vacancies randomised to résumé and reference, résumé and work sample, or résumé, work sample and portable verification: shortlist time, interviews, hires, early performance, attrition, repeated testing, cost per hire, fee acceptance | No improvement over the employer’s usual workflow, or employers value the evidence but won’t pay what it costs to produce. The review’s planning threshold is at least 20% lower cost per qualified hire, or a reliable gain in early retention, with willingness to pay at least the cost of the evidence |
| Net contribution (Experiment 2) | Accepted output, supervisor and training hours, rework, coordination, wages, retention and time to independent productivity for paid apprentices | Direct output stays negative and only projected retention makes the return look positive (Shift 4) |
How strong is the evidence?
- Established
- Work samples are most valuable where competencies are needed at entry [1]; sector-focused programs with strong employer links have produced large, lasting earnings gains [3]
- Evidence-supported
- Multi-sided human-capital systems need an anchor to start, and no evidence supports an open market that starts itself
- Hypothesis
- One capability family with an employer consortium and an intermediary is the smallest market in which reuse can be measured
- Unknown
- Whether a first market’s playbook replicates across capability families and across economies
Sources
- [1]U.S. Office of Personnel Management, Work Samples and SimulationsChecked
- [2]Cedefop, Microcredentials for labour market education and training: the added value for end users (2023)Checked (executive summary and chapters 1–3)
- [3]Katz, Roth, Hendra and Schaberg, Why Do Sectoral Employment Programs Work? Lessons from WorkAdvance, Journal of Labor Economics (2022)Checked
Spotted a source we missed or a rule we misread? Tell us what we got wrong.