Companies adopting artificial intelligence in Denmark are beginning to change their workforces through slower recruitment rather than mass redundancies, according to new research offering one of the clearest company-level views yet of how AI is feeding into employment.

Danmarks Nationalbank found that companies which began using AI during 2023 and early 2024 subsequently recorded weaker employment growth than comparable businesses that had not adopted the technology. By late 2025, the difference had widened to around 11%, although the researchers stress that the figure represents lower growth relative to previous trends and a control group rather than an 11% fall in headcount.

The effect was strongest among smaller businesses and came mainly through fewer new hires, particularly in occupations judged to have greater exposure to AI. Younger employees and people with longer periods of education were also disproportionately represented in the parts of the workforce where employment growth slowed.

Wages have not shown the same pattern, while overall Danish employment remains broadly unaffected. Taken together, the findings describe companies changing the number and type of people they recruit as AI enters routine operations, rather than employers rapidly replacing existing staff.

Hiring absorbs the first adjustment

The study combines Statistics Denmark information on company AI adoption with monthly employer and employee register data, allowing researchers to follow businesses before and after they began using the technology. That gives the analysis a firmer basis than surveys asking employers what they expect to do or workers whether they feel threatened by automation.

Labour-market effects can emerge long before an organisation announces a restructuring programme. A business that might previously have added ten employees may add eight instead, leaving its current workforce intact while gradually changing its size and composition. Across thousands of companies, weaker recruitment can become economically significant even if redundancy statistics remain quiet.

Nationalbank Governor Signe Krogstrup said: “The transformation we expected from AI has begun in Denmark.” The central bank nevertheless cautions against treating the company-level results as evidence of a wider employment shock.

A mechanical calculation based on the study’s estimates puts the implied reduction in hiring among the relevant AI-adopting companies at roughly 5,700 people over two years, equivalent to about 0.33% of all new hires in Denmark during the period. Employees who might otherwise have joined those companies appear, so far, to have found work elsewhere.

That movement helps explain why weaker employment growth among adopters can coexist with a relatively steady national labour market. Workers moving between employers, new companies, and continued demand elsewhere in the economy can absorb changes that would look more severe if viewed only through an individual organisation’s staffing plan.

Recruitment becomes part of the AI return

Much of the corporate discussion around AI productivity has concentrated on whether existing employees complete tasks more quickly, but part of the financial return may appear through recruitment that never happens. Where software absorbs research, drafting, analysis, coding, or administrative work that would otherwise have supported another role, productivity gains can emerge gradually through payroll growth that trails revenue or output.

Measuring that effect is difficult inside one organisation because a company has to establish what staffing would have looked like without AI. Licence and infrastructure costs are straightforward to record, while an avoided future hire exists only against a counterfactual. The Danish research addresses part of that problem by comparing adopters with businesses that did not make the same change.

The occupational pattern also complicates assumptions about which workers face the earliest effects. Younger employees, graduates, and people in roles with greater exposure to cognitive automation appear more prominently, suggesting that AI may be affecting some of the entry points through which businesses traditionally build professional talent pipelines.

Reducing repetitive junior work can produce immediate savings, although repeated reductions in early-career hiring create a longer-term organisational question. Experienced employees still have to come from somewhere. If entry-level recruitment contracts while businesses continue to need senior judgement, management experience, and domain knowledge, workforce planning has to account for how those capabilities will be developed.

There is no equivalent evidence in the Danish data of a broad shift towards lower-paid work. Average hourly wages at AI-adopting companies did not show a clear relative change, while national figures do not indicate that AI has produced a wider deterioration in pay or employment.

The study consequently lands between two of the louder claims made about workplace AI. It does not show a labour market being rapidly emptied by automation, but nor does it support the idea that adoption has had no detectable employment effect. Companies using AI are already adjusting differently from their peers, and recruitment is where that difference is appearing first.

As adoption spreads beyond early users, the Danish data should become more informative. Companies that embraced AI first may differ from later adopters in productivity, digital capability, management, or workforce structure, while the technology itself continues to change. What the current evidence establishes more firmly is that employment consequences can begin before they become visible in national job totals.