The AI ROI business data is finally in — and it comes from the most credible source possible: nearly 6,000 CEOs and CFOs surveyed across four countries by the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank, and Macquarie University. This isn’t a vendor white paper. It isn’t a paid survey of self-selected optimists. It’s representative, peer-reviewed, and published as NBER Working Paper No. 34836 in February 2026.
The findings are striking — not because AI adoption is low, but because the gap between adoption and measurable impact is enormous. And that gap has serious implications for every SaaS company, PE investor, and enterprise buyer making decisions based on AI-driven value propositions right now. If you’ve been following our analysis of AI valuation premiums and the AI bubble’s structural vulnerabilities, this data provides the most authoritative empirical grounding yet for the contrarian position.
The Headline: 69% Adoption, 80%+ Zero Impact
Let’s start with what the hype cycle gets right. AI adoption at the enterprise level is genuinely widespread. Across the US, UK, Germany, and Australia, 69% of firms surveyed say they currently use some form of AI technology. In the US, that figure reaches 78%. Text generation using large language models leads all categories, used by 41% of firms on average.
So the adoption narrative isn’t wrong. What’s wrong is everything that gets claimed on the back of it.
Over 80% of firms report zero measurable impact from AI on either employment or productivity over the past three years.
Specifically: 90% of business managers across all four countries estimate no impact of AI on their employment over the past three years. And 89% report no impact on labor productivity — defined as sales per employee — over the same period. The average measured productivity boost across all firms? A cumulative 0.29% over three years. Not per year. Cumulative. Over three years.
That is the AI ROI reality that vendors don’t put in their pitch decks — and it connects directly to what we documented in our earlier analysis of the margin crisis hiding inside enterprise AI adoption.
| SOURCE — NBER WORKING PAPER NO. 34836 (FEBRUARY 2026)Survey of ~6,000 CFOs, CEOs, and senior executives across the US, UK, Germany, and Australia. Conducted November 2025 – January 2026. Respondents recruited by phone to confirm identity and position. Sponsored by four central banks and Macquarie University. Lead authors: Nicholas Bloom (Stanford), Steven J. Davis (Stanford/Hoover), Jose Maria Barrero (ITAM), and teams from the Federal Reserve Bank of Atlanta, Bank of England, and Deutsche Bundesbank. |
Why This Data Is Different From Everything You’ve Been Reading
Survey methodology matters enormously when studying AI adoption, and this study’s authors are explicit about the problem. McKinsey’s 2025 State of AI report estimated that 88% of organizations use AI in at least one business function. Meanwhile, the US Census Bureau’s Business Trends and Outlook Survey (BTOS) — a nationally representative, unpaid survey — estimated AI use at roughly 9% of businesses as of early 2024, rising to approximately 18% by December 2025.
That’s nearly a 10x difference between two studies released within months of each other. The NBER authors attribute this gap to two factors: who is answering the survey, and whether they are being paid to answer it.
Paid online surveys targeting executives face a well-documented problem: imposter rates can exceed 80%, meaning most respondents claiming to be executives are not. This study solved that problem. Executives were recruited by telephone, their identity and position confirmed before being moved into the survey panel. They were not compensated. The sponsoring institutions — central banks — lent credibility that improved recruitment of genuine C-suite respondents.
The result: over 90% of respondents in the UK and Germany hold CEO, CFO, or senior management positions. These are the people who actually know what their firm is doing with AI — and what they’re reporting is far more measured than the vendor narrative suggests.
This methodology gap is something we examine directly in our evidence-first research framework — the principle that the quality of the data source matters as much as the finding itself.
The Executive Usage Problem Nobody Talks About
Here’s a finding that deserves more attention than it typically gets: even among executives who say their firm uses AI, personal usage is minimal.
Across all four countries, the average senior executive uses AI for just 1.5 hours per week. The modal response — the most common answer — was ‘up to 1 hour a week.’ A full 28% of executive respondents report using AI not at all during a typical working week.
| Executive AI Usage (Weekly) | Share of Respondents |
| Not at all | 28% |
| Up to 1 hour per week | 41% (modal response) |
| 1 to 5 hours per week | 24% |
| More than 5 hours per week | 7% |
CEOs use AI more frequently than CFOs or other senior executives — a finding consistent across both the US and UK samples. And usage has increased meaningfully through 2025: among UK executives, average weekly usage jumped from 0.9 hours per week in early 2025 to 1.4 hours by late 2025, a roughly 50% increase in under a year.
But here’s the strategic implication: if the executives driving AI adoption strategy are themselves using these tools for 90 minutes a week, what does that tell you about the depth of organizational integration at the firm level? The answer, borne out in the productivity numbers, is that integration remains shallow for the vast majority of businesses. This is the same pattern we documented in our analysis of why AI coding agents are failing to deliver enterprise developer productivity gains — surface adoption masking deep integration failure.
If the C-suite averages 90 minutes of AI use per week, the productivity revolution is still mostly a future event — not a present one.
The Expectation Gap: Where the Real Risk Lives
The most strategically significant finding in this research is not the current impact data. It’s the divergence between current reality and near-term expectations — and the divergence within expectations between executives and employees.
What Executives Expect (Next 3 Years)
Looking forward, firms are significantly more optimistic. Executives predict AI will boost productivity by an average of 1.4% over the next three years and reduce employment by 0.7% over the same period. The implied net output gain is 0.8%.
US executives are the most bullish: they expect a 2.25% productivity gain and a 1.19% reduction in headcount. UK executives are close behind, expecting 1.86% productivity improvement and a 1.36% reduction in employment — the largest expected job reduction of any country in the study.
What Employees Expect — A Striking Contrast
The same questions were posed to approximately 3,000 US employees via the Survey of Working Arrangements and Attitudes (SWAA), a monthly survey of US residents aged 20 to 64. The contrast is stark. Employees expect AI to increase employment at their firms by 0.5% over the next three years — while their executives expect a 1.19% decrease. Employees also expect smaller productivity gains (0.9% vs. 2.25% from US executives).
This is not a small discrepancy. It is a structural misalignment between the people setting AI strategy and the people living inside the organizations those strategies will affect. For M&A practitioners, this gap carries specific risks: labor disruption, change management failures, and integration challenges that can erode deal value significantly.
| Metric | Executive Prediction (US) | Employee Prediction (US) |
| Productivity gain (3 yr) | +2.25% | +0.90% |
| Employment change (3 yr) | -1.19% | +0.45% |
| Implied output change | +1.06% | +1.37% |
| THE EXPECTATION DIVERGENCE — STRATEGIC IMPLICATIONThe gap between executive and employee AI expectations creates a predictable pattern in M&A integration: leaders underestimate change management costs, overestimate productivity timelines, and encounter workforce friction that wasn’t modeled in the deal thesis. Our AI-accelerated win/loss analysis framework is specifically designed to surface these misalignments before they become deal liabilities. |
Sector-Level Variance: Not All Enterprise SaaS Is Equal
The aggregate numbers mask significant sectoral variation that matters enormously for enterprise SaaS valuation and M&A strategy.
On the productivity side, Information and Communications firms lead all sectors, expecting AI to increase productivity by 2.8% over the next three years. Administrative and Support services follow at 2.5%. These are precisely the sectors where AI-native workflow tools and automation platforms are being marketed most aggressively — and where the investment thesis has the most structural support.
On the employment impact side, Wholesale and Retail (-2.0%) and Accommodation and Food (-1.8%) expect the largest AI-driven workforce reductions. For SaaS companies selling into these verticals, this creates both opportunity and risk: procurement teams are under real pressure to demonstrate AI ROI, creating urgency in the buying cycle, but integration complexity is high and workforce disruption can stall implementations.
| UK Sector | Expected AI Productivity Gain (3 yr) |
| Information & Communications | +2.8% |
| Admin & Support | +2.5% |
| Professional & Scientific | +2.5% |
| Finance & Insurance | +2.1% |
| Wholesale & Retail | +1.9% |
| Accommodation & Food | +1.2% |
| Construction | +0.7% |
| Other Production | +0.4% |
The sectors with the weakest expected productivity impact — Construction, Recreational Services, and Accommodation and Food — are also the slowest current adopters. If you’re evaluating a vertical SaaS business serving these industries, be skeptical of AI-driven growth assumptions without hard usage evidence. This directly informs how we approach product-market fit analysis for clients in these verticals.
What This Means for M&A Due Diligence in 2026
This research doesn’t suggest AI is a fiction. It suggests that the productivity benefits are real but lagged — and that the current market is pricing in expectations that firms themselves aren’t yet able to validate with data. We’ve been making this argument for some time, most recently in our piece on AI washing and the gap between marketing claims and operational reality.
For buyers evaluating SaaS targets that lead with AI value propositions, this framework is useful:
- Ask for realized impact data, not projected impact data. If a target claims AI drives productivity gains for customers, ask for the cohort-level measurement. What percentage of their customer base can actually demonstrate measurable improvement in sales per employee?
- Distinguish adoption from integration. The NBER data shows that adoption (69% of firms) massively outpaces impact (sub-1% cumulative productivity gains). A SaaS company reporting high AI feature adoption rates may still have customers at the shallow end of this curve.
- Weight sector dynamics carefully. Information and Communications and Professional and Scientific services show the strongest near-term AI productivity expectations. Businesses serving these sectors have a more credible near-term runway than those targeting sectors with low AI integration depth.
- Factor in the executive-employee expectation gap. In any business with significant human capital, the gap between what leadership expects AI to do and what employees expect is a change management liability. Assess it explicitly in diligence.
- Be cautious of US AI premiums in late-stage deals. US executives hold the most bullish AI expectations of any country in this study. That optimism is arguably already reflected in valuation multiples — a dynamic we examined in depth in our analysis of what PitchBook’s data reveals about AI valuation premiums.
| RELATED SERVICE — STRATEGIC ACQUISITION EXIT ADVISORYIf you’re preparing a SaaS company for an M&A transaction in the current environment, our Strategic Acquisition Exit Advisory service is specifically designed to position AI-driven value propositions with the rigor that sophisticated buyers and PE investors now require. Book a call to discuss your situation. |
The Productivity Paradox Is Alive and Well
In 1987, economist Robert Solow famously observed that ‘you can see the computer age everywhere except in the productivity statistics.’ His remark became known as the productivity paradox — and it remained unresolved for nearly a decade before computerization’s macroeconomic effects became measurable.
We are likely in an analogous moment with AI. The technology is real. The adoption is real. The productivity impact, at the macro level, has not yet materialized in representative firm-level data.
The NBER authors themselves note that their forward-looking productivity estimates, if realized, ‘could imply a reversal of the long-run decline in productivity growth in many advanced economies.’ That is a significant conditional. For context, Daron Acemoglu’s 2025 analysis in Economic Policy estimated only a 0.66% ten-year TFP gain for the US from AI adoption — a figure far below the claims driving current market enthusiasm.
It’s also worth noting what Brynjolfsson, Li, and Raymond (2025) found in their study of generative AI in customer support: productivity gains of 14% in specific, well-defined tasks. Similarly, Noy and Zhang (2023) found meaningful gains in writing assignments via ChatGPT, and Dell’Acqua et al. (2023) documented gains in consulting work. Task-level productivity gains are real. Firm-level and economy-wide gains, as of early 2026, are not yet measurable at the aggregate. Understanding that distinction is critical for any investor evaluating AI-driven business models.
Task-level AI productivity is real. Firm-level and economy-wide AI productivity remains a future event — and every investment thesis that conflates the two is carrying unpriced risk.
The Bottom Line for SaaS Executives and Investors
The NBER working paper “Firm Data on AI” represents the most rigorous representative dataset on enterprise AI adoption and impact available as of early 2026. Its findings should recalibrate how SaaS founders, enterprise CTOs, and PE investors evaluate AI-driven value claims.
AI adoption is broad. Impact is narrow. Expectations are large. The gap between all three is where the real strategic risk and opportunity lives in the current market.
For companies positioning for a liquidity event in the next 12 to 24 months, this data creates a narrow window: AI hype is inflating buyer optimism and, in some cases, valuation multiples. But as representative impact data accumulates and becomes impossible to ignore, the narrative will compress toward reality. Companies that can demonstrate genuine, measured productivity impact — not just feature adoption — will command durable premiums. Those selling the story without the substance will face increasing buyer scrutiny. This is precisely what we mean when we talk about the shift from AI washing to AI accountability.
The AI ROI conversation is evolving from aspiration to accountability. Position your business accordingly.
About the Author
John is the Managing Director of DevelopmentCorporate LLC, an M&A advisory firm specializing in enterprise SaaS transactions. With 25+ years in enterprise technology and experience leading $175+ million in acquisitions, he advises founders, investors, and executive teams navigating liquidity events in the current market. Explore our Strategic Advisory Services or book a call to discuss your company’s positioning.
Primary Source
Yotzov, I., Barrero, J.M., Bloom, N., Bunn, P., Davis, S.J., et al. (2026). Firm Data on AI. NBER Working Paper No. 34836. National Bureau of Economic Research.
