The adoption of artificial intelligence inside corporate workflows is expanding, yet the economic payoff remains modest. A comprehensive global survey published by the National Bureau of Economic Research (NBER), led by Nicholas Bloom and Steven J. Davis, presents a sobering picture of corporate reality. Surveying nearly 9,000 CEOs, CFOs, and top managers across seven industrial and emerging economies, including Germany and the United States, the researchers tracked actual adoption patterns. While 78 percent of companies report deploying AI productively and 80 percent of executives use it personally, the past three years yielded an average productivity gain of just 0.2 percent annually.
At the same time, corporate capital commitments are climbing rapidly, though the distribution is starkly asymmetric. US businesses plan to allocate an average of 2,000 dollars per employee to AI solutions in 2026, which translates to an annual aggregate budget of roughly 300 billion dollars. However, capital expenditure is heavily concentrated: a tiny cohort of just two percent of firms accounts for more than 50 percent of all AI spending. Looking ahead, surveyed executives project an acceleration of productivity gains to 0.6 percent annually over the next three years, an improvement that remains measured relative to current financial commitments.
The difficulty of capturing returns is underscored by a recent CFO survey from Deloitte Deutschland, which examined more than 120 finance chiefs at large German corporations. Only 17 percent of surveyed enterprises can currently prove the economic value or measurable return on investment of their AI initiatives using hard financial metrics. Cost transparency is equally limited. Just 42 percent of finance departments actively track the full costs of inference, interfaces, and software licensing. Despite this visibility gap, CFOs plan to double their AI spending from an average of 0.5 percent of total operating costs today to 1.0 percent next year.
Theoretical grounding for this divergence comes from the Stanford Graduate School of Business, where researchers including Tim de Silva and Rebecca Blank examined workplace outcomes. While generative AI reliably reduces individual working hours spent on writing and routine analysis, that efficiency does not scale to company-wide performance. The researchers found no statistically significant increase in total factor productivity across organizations. Instead of lifting top-line revenue or reducing overall payroll expenses, saved hours are largely absorbed by informal tasks and qualitative adjustments that fail to register on profit-and-loss statements.
Labor markets reflect a similar pattern of reallocation rather than outright contraction. According to the NBER findings, the net employment effect across firms hovered near zero percent over the past three years. Yet beneath that aggregate figure, over 40 percent of surveyed companies experienced significant restructuring, as job cuts in certain roles balanced hiring in others. For corporate leadership, realism is beginning to temper expectations: 70 percent of German CFOs surveyed by Deloitte indicate that reliable metrics and demonstrable payback horizons will require at least another two years to materialize.