NBER Working Paper 34836: The Executive Perception Gap on AI

Source credibility: HIGH. Independent academic working paper from NBER/Stanford (Nicholas Bloom and 12 co-authors). No commercial interest in outcome. Largest multi-country executive AI survey in corpus at time of publication. Senior executive respondents across four countries, n~6,000. Working paper status (not yet peer-reviewed at ingestion); however, NBER papers from this author group have a strong publication track record. The finding directionally corroborates McKinsey, BCG, and MIT CISR independently.

Temporal tier: TIER 1. Published February 2026, revised March 2026. Data collection spans the prior three years with prospective projections. Current at ingestion date.

Executive Summary

Sixty-nine percent of firms now use AI. Nine in ten executives say it hasn’t moved their numbers yet.

That is the central finding in NBER Working Paper 34836, a survey of nearly 6,000 senior business executives across the US, UK, Germany, and Australia published in February 2026. The same executives who report zero past impact predict meaningful future gains: 1.4% productivity improvement, 0.8% output growth, and 0.7% employment reduction over the next three years.

The paper frames a precise diagnosis: AI is widely deployed, lightly used by the leaders who are supposed to champion it, and not yet visible in the numbers. The perception gap between reported past impact and anticipated future impact is the most consequential data point for any executive deciding how hard to push in 2026.

Key Data Points

Adoption and Usage

The Prediction Gap: Past vs. Future

Employer vs. Employee Expectations (Next 3 Years)

The direction is reversed. Executives and employees are not debating the size of AI’s impact on jobs — they disagree on the sign.

What This Means for Your Organization

The 90% are not wrong — yet

Nine in ten executives reporting no measurable past impact is not evidence that AI fails. It is evidence of where most organizations are in the adoption arc. The timeline matters: AI reached meaningful enterprise penetration only in 2023–2024. Organizational change lags tool deployment. Three-year retrospectives measuring an adoption wave that is 18 months old will show thin results.

What the data identifies is a window. Executives believe impact is coming. The work now is ensuring the organization is positioned to capture it when it arrives.

What separates the 10% from the 90%

The NBER finding triangulates cleanly with three other independent data sources:

McKinsey State of AI (November 2025, n=1,993, 105 countries): Only 6% of organizations attribute more than 5% EBIT impact to AI. Among those 6%, the distinguishing behavior is not tool selection — it is workflow redesign. Fifty-five percent of high performers fundamentally redesigned workflows when deploying AI, versus 18% of other firms.

BCG AI Radar 2026 (n=640 CEOs, January 2026): The ~15% of CEOs BCG calls “Trailblazers” commit 73% of transformation budget to AI and upskill 69% of the workforce. The other 85% put in roughly one-third as much and show one-third the results.

Stanford AI Index 2026 (April 2026): 88% organizational adoption, 4–8% reporting substantial financial impact. The gap between the two numbers is the organizational work that has not been done — governance, workflow architecture, role redesign.

The pattern across all four data sources is the same: tools are deployed, workflows are not redesigned, and so results do not appear in the numbers. The 10% who do report impact are not smarter or better resourced — they changed how work is done, not just what tools are available.

The 1.5 hours per week problem

The NBER finding that executives personally average 1.5 hours per week of AI use deserves direct attention. Over two-thirds of executives use AI regularly, but 1.5 hours per week is roughly 4% of a 40-hour work week. That is not enough exposure to develop the informed judgment needed to direct AI strategy, evaluate vendor claims, or redesign workflows credibly.

The organizations that move from the 90% to the 10% typically have one or two people who use AI intensively enough to understand its actual capabilities and failure modes — and who then redesign specific workflows around what they learn. That is not an AI budget decision. It is a time allocation decision.

The employer-employee expectation gap

Executives predict AI will reduce employment 0.7% over the next three years. Employees predict it will increase employment 0.5%. Neither group is reading the same future.

The practical implication is a communications and change management problem that will surface regardless of which projection proves correct. If executives believe headcount will fall and employees believe it will rise, that misalignment will manifest in resistance, attrition, and governance friction before any outcomes are known. The time to address it is before the predictions become policy.

Where to start

The NBER data suggests a diagnosis before a prescription. Before committing new AI spend, three questions locate the organization on the arc:

1. What percentage of knowledge workers use AI tools for more than 10% of their daily work? (If below 20%, tool access is not the constraint — workflow design is.)
2. Has the organization redesigned any complete workflow around AI, end-to-end, with measured outcomes? (A single documented case creates the organizational template for the next one.)
3. Do the executives sponsoring AI investment use it personally for more than one hour per day? (Below that threshold, the strategy is not grounded in direct knowledge of what the tools actually do.)

If any of those questions produce an uncomfortable answer, that is the right starting point. Reach out to brandon@brandonsneider.com to work through the diagnosis.

Sources

See also (wiki): executive-ai-optimism, ai-maturity-models, roi-evidence

Brandon Sneider | brandon@brandonsneider.com April 2026