Large‐Language Models, Small Labor‐Market Effects: What the New Danish Mega‐Study Really Says

Study at a Glance

Key Takeaways

1. Rapid, Firm‑Led Adoption

• 83 % of employees use chatbots when management encourages them (vs 47 % without encouragement).

• 38 % of firms have already rolled out in‑house or customised LLMs, and 30 % offer staff training.

2. Productivity Gains Are Real but Modest

• Average self‑reported time savings: 2.8 % of work hours (peaking at ~7 % for marketing & software teams under active encouragement).

• Benefits—time saved, quality boosts, creativity—are 10–40 % larger in firms that invest in training and bespoke tools.

3. …Yet Paycheques Stay Flat

• Earnings, hourly wages and recorded hours all show “precise zeros”: effects smaller than ±1  % and statistically insignificant across every occupation studied.

• Only 3–7 % of the minutes saved flow through to pay, hinting at weak wage pass‑through in the short run.

4. New Tasks, Not Mass Lay‑offs

• 17 % of users—and 5 % of non‑users in the same firms—report new AI‑related job tasks (integration, policy drafting, quality review, etc.).

• High‑adoption workplaces show no change in total employment or wage bills, debunking imminent displacement fears.

Why This Matters

1. Reality Check on AI Hype
   The findings undercut predictions of an automatic, rapid labour‑market upheaval. Yes, chatbots speed up certain tasks, but aggregate pay‑packets are unchanged so far.

2. Complementary Investments Are Crucial
   Productivity gains are largest where companies supply training, safe‑data sandboxes and clear usage policies—consistent with the classic “Productivity J‑Curve” pattern.

3. Policy & Strategy Implications
   Workers: Build complementary skills (prompt engineering, AI oversight).
   Employers: Focus on workflow redesign, not just licences.
   Policymakers: Target upskilling and diffusion, not panic‑driven job protection.

Link to study: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5219933