Do AI productivity gains feel larger than they actually measure?

A survey of nearly 750 corporate executives, reported in NBER working paper 34984 (Baslandze, Graham, Meyer, Waddell et al., March 2026), documents "substantial heterogeneity in AI adoption across firms, with more than half having already invested, though many smaller firms are only beginning to do so." Where adoption has taken hold, "labor productivity gains are positive, vary across sectors, and are expected to strengthen in 2026, with the largest effects concentrated in high-skill services and finance." The paper's headline finding is what it calls "a productivity paradox, in which perceived productivity gains are larger than measured productivity gains, likely reflecting a delay in revenue realizations."

The mechanism the authors give is that the gains they do measure "are not primarily driven by firms' capital deepening but instead reflect increases in revenue-based total factor productivity, closely associated with innovation- and demand-oriented channels." Because the measured metric is tied to revenue, and revenue lags the operational changes executives can already see internally, perceived gains run ahead of what shows up in the numbers. On the workforce side, the survey finds "little evidence of near-term aggregate employment declines due to AI, though larger companies anticipate AI-driven workforce reductions, while smaller firms expect modest gains," alongside "compositional reallocation of labor both within and across firms, with routine clerical roles declining and a relative demand for skilled technical roles increasing." The authors built an index ranking which job functions are most negatively affected.

This sits as a size-stratified complement to Do firms substitute labor for AI at different rates?: both find that how AI reshapes labor depends on firm characteristics rather than applying uniformly, here split by company size (larger firms anticipating cuts, smaller firms expecting gains) rather than by exposure-driven substitution cost. It also qualifies Is generative AI displacing workers at economy-wide scale?: that finding rests on measured ADP payroll outcomes, while this survey's "little evidence of near-term aggregate employment declines" is executives' self-reported near-term outlook — a perception, not a measurement, and one this same paper's own productivity paradox suggests should be read cautiously. The reported shift of routine clerical roles down and skilled technical roles up also matches the channel Does automation raise or lower the skills that remaining work demands? predicts: removing lower-expertise clerical tasks while raising demand for higher-expertise technical ones.

The excerpt does not establish that the anticipated workforce reductions at larger firms will actually occur, nor does it validate the negatively-affected-job-function index against any outcome other than executives' own rankings — both are self-reported expectations gathered at one survey point, not behavioral or payroll data. It also does not show that perceived productivity gains will in fact convert to measured revenue once the delay the authors posit has passed. The implication the paper supports is narrower than the headline: AI is producing real, sector-concentrated productivity gains and a real compositional shift in demand toward technical over clerical roles, but claims about near-term aggregate employment and firm-size-specific workforce cuts rest on executive perception and should be weighed against measured data, such as payroll records, before being treated as settled.