AI Adoption Depends on Cost, Not Just Capability, Study Finds

The exposure gap

Despite rapid advances in artificial intelligence capabilities, workplace adoption has lagged behind predictions. New research examining nearly 10,000 German workers reveals why: organizations adopt AI based on comparative advantage—whether the technology is worth using relative to human labor—not merely on what AI can technically accomplish.

The study, conducted by researchers Ilse Lindenlaub, Roozbeh Oh, Mariana A. Rodríguez, and Laura Veldkamp, analyzed data from the 2024 Digital Transformation and the Changing World of Work survey linked to administrative employment records. Their findings challenge the standard approach to forecasting AI's labor market impact, which relies heavily on "exposure" measures that catalog which occupations contain AI-compatible tasks.

Exposure alone explains only 25% of observed variation in AI adoption across occupations. When the researchers incorporated user costs—the practical burden of deploying AI in production—and worker productivity relative to wages, their model's explanatory power jumped to roughly 60%.

Why accountants resist and teachers adopt

The gap between capability and adoption creates surprising patterns. Accounting faces enormous technical exposure to AI automation, yet adoption remains tempered by steep verification costs, privacy concerns, and the relatively high productivity of human accountants. Teaching, conversely, shows lower technical exposure but benefits from lower deployment costs, driving unexpectedly robust adoption.

Across the German labor market, exposure measures and comparative-advantage measures point in opposite directions for occupations representing approximately 30% of employment. This divergence has direct policy implications: retraining programs, adjustment assistance, and workforce planning built on exposure data alone may target the wrong workers and industries.

User costs drive future diffusion

The researchers used their framework to project adoption over the next three years. They estimate the share of workers using AI will rise from 44% to roughly 81%, with reductions in user costs playing a larger role than improvements in AI productivity itself.

As verification becomes easier, regulatory frameworks mature, and organizational workflows adapt, AI could spread rapidly even in occupations where the technology's raw capabilities improve only modestly. Growth will be fastest among occupations currently in the middle of the adoption distribution, while falling costs will also bring AI into jobs that appear largely untouched today.

Why it matters

This research reframes the AI adoption debate from a question of technological capability to one of economic implementation. For business leaders, it suggests that investment in reducing deployment friction—streamlining verification, clarifying compliance, redesigning workflows—may accelerate returns more effectively than waiting for the next model release. For policymakers, it implies that labor market disruption will follow patterns distinct from technical exposure maps, requiring different targeting for education, retraining, and safety net programs.

The findings were first reported by the Centre for Economic Policy Research and are detailed in CEPR Discussion Paper 21589.