MERIT’s small market scan should seed pilots, not forecasts

A useful scan with a small company denominator

The MERIT project’s final market-analysis release summarises four years of work on AI, cybersecurity and internet-of-things skills. The final edition combines responses from 25 companies with a review of more than 70 industry reports. It maps reported needs to 299 occupations using ESCO and the European e-Competence Framework.

The release makes its boundary unusually clear: the findings should be treated as signals rather than statistics. Twenty-five companies cannot estimate population demand across Europe, and organisations that chose to participate may be more digitally engaged than non-participants. Report reviews add breadth but can repeat the same underlying surveys or forecasts.

That does not make the scan unusable. It creates a vocabulary of tasks, roles and potential curriculum areas. Its references to employer intentions over the next one or two years can help identify questions worth testing. They should not become precise headcount, salary or enrolment forecasts.

Broader guidance supports foundations, not a fixed list

Cedefop’s inter-agency report on skills development for AI adoption argues that AI changes the mix of cognitive, socio-emotional, digital and AI-specific capabilities. It treats AI literacy and human agency as foundations and warns against reducing development to narrow technical content.

The report is broader than the MERIT company scan and does not validate its individual demand figures. It provides a useful counterweight: even when market evidence names a tool or specialty, capability design should include judgment, collaboration, ethics and the ability to work across changing systems.

Turn the scan into a hypothesis queue

For every signal, record the source denominator, date, geography, occupation mapping and claimed time horizon. Then translate the label into a local task. “AI skills” is not testable; evaluating model output against a domain standard, tracing data provenance or configuring an access boundary is.

Choose a business decision that the task could improve and define evidence before training starts. Evidence might be a work sample, fewer review defects, faster incident triage or successful performance under a changed tool. Keep salary willingness and future-intention statements separate from observed vacancies, completed hiring or productivity.

Run a small test-before-invest pilot with workers whose jobs actually contain the task. Measure baseline performance, instruction time, supervisor load, transfer to a second case and retention after the tool or prompt changes. Stop when the local task is absent, the evidence does not improve or the operating model cannot provide safe practice.

Use the occupation mapping as an index, not an instruction. A task can cross several roles, and one occupation can need different depth by sector. Compare the result with internal job data, vacancy text, incident records and manager interviews before changing a framework or purchasing a curriculum.

This complements the Estonia OSKA delivery-contract analysis: an external skills signal becomes actionable only when ownership, deliverables and local evidence are explicit.

Decision rule

Do not convert the MERIT scan into a demand forecast or a salary benchmark. Admit a topic to the learning portfolio only when it maps to a current task, an accountable decision and a short pilot with a predeclared stop rule.