90% of Executives Say AI Hasn't Boosted Productivity — and AI Layoffs Are Making It Worse

TL;DR. A Federal Reserve survey of ~750 executives finds 90% say AI hasn't boosted productivity at their companies. A University of Pittsburgh study explains the mechanism: companies using AI to justify layoffs create employee insecurity that blocks the adoption that would actually deliver gains. The cut-costs-with-AI strategy is producing near-zero or negative returns. Executives who fund AI while laying off workers are, on average, getting neither.

For most of 2024 and 2025, the story companies told investors was clean: invest in AI, reduce headcount, unlock efficiency. That story has an empirical problem.

The Federal Reserve's executive survey

The Federal Reserve Bank of Atlanta surveyed ~748 corporate executives across diverse sectors, firm sizes, and geographies between November 2025 and January 2026 — two survey waves including CFO Survey respondents and executives from FEI/NASDAQ member firms and Duke alumni. The headline: 90% of executives say AI has not yet boosted productivity at their companies.

That's not a soft opinion poll. These are the decision-makers who approved the AI budgets and are watching the operational metrics. When they say it isn't working, they mean they're not seeing it in the numbers they're responsible for.

The same survey found nearly 60% of firms invested in AI in 2025, with that figure rising to 80% for large firms. Over 80% expect to invest in 2026. Investment is accelerating. Returns, by the executives' own account, are not.

The measured average gain — from firms' own reporting — was 1.8% productivity improvement per worker in 2025. Non-zero. But modest, concentrated in high-skill services and finance, and far below what the level of investment and expectation would suggest.

Why the productivity gains aren't materializing

A University of Pittsburgh study by Professor Mark Ma offers the mechanism. Ma's team analyzed millions of Glassdoor employee reviews, thousands of corporate financial reports, and hundreds of AI-related layoff announcements from U.S. public companies over five years — including about 10,000 earnings-call transcripts to measure management sentiment about AI.

The finding is counterintuitive only if you think productivity is a function of tools. It's actually a function of people using tools. And people don't use tools they think will get them fired.

When companies announce AI investments alongside job cuts — explicitly attributing the layoffs to automation — employee sentiment toward AI in reviews turns sharply negative. Ma's data shows that employee AI sentiment has a stronger correlation with firm productivity than management optimism about AI investments. Companies where workers feel threatened by AI are companies where workers are not adopting AI, which means the technology delivers nothing.

The stock-market data underlines the problem: companies announcing AI-attributed layoffs have seen near-zero average stock returns on announcement. Negative or near-zero for over half. The market has priced in that the "cut costs via AI" strategy isn't producing the efficiency gains it promises.

The loop looks like this:

1. Company invests in AI
2. Company cuts jobs to offset cost, citing AI as justification
3. Remaining workers fear displacement → resist or underuse AI tools
4. Productivity gains fail to materialize
5. Company reports AI hasn't boosted productivity
6. Executive pressure to cut further

Ma calls this a counterproductive strategy. The data suggests he's right.

What executives are actually finding works

The Atlanta Fed study found that where AI does deliver at the organizational level, two patterns are consistent:

• High-skill analytical work sees the largest gains — not because AI is replacing analysts, but because AI is increasing what each analyst can cover and produce
• Firms that treat AI as complementary to skilled workers outperform firms that position it as a substitute

The University of Pittsburgh data adds: firms where employee sentiment toward AI is positive are the firms showing real productivity gains. That means employee buy-in — built through trust, genuine training, and not weaponizing AI against workers — is not a soft factor. It's the primary lever.

What this means for professionals using AI today

If you're inside an organization that's cutting jobs and blaming AI, you're inside one of the organizations making this mistake. The executives leading that strategy are, per the data, getting neither the cost savings nor the productivity gains they promised their boards.

That's not the outcome you want to be caught in, but it does clarify what's valuable:

Professionals who build real AI fluency become the counter-evidence that the technology works. The organizations that can't make AI deliver at scale will eventually recognize that individual AI capability — people who can demonstrate measurable output improvements — is the ingredient they're missing. Being that person is a better position than waiting to see whether the strategy works.

The practical steps:

1. Document your AI delta. For every task you've made faster or better with AI, keep a plain log: before time, after time, quality outcome. This converts "I use AI" (which everyone says) into "AI saves me 2 hours per [task], here's the evidence" (which almost nobody can show).

2. Stay inside sanctioned tools. The insecurity driving resistance is real. Don't add fuel by routing around your organization's approved tools — you need the documented wins to be on compliant infrastructure.

3. Don't conflate organizational failure with tool failure. ChatGPT, Claude, and Gemini are not under-delivering because the models are weak. The individual-level studies (GitHub Copilot at MIT, Claude internal usage data) show real gains for engaged users. The organizational failures are governance, training, and trust problems — not model problems.

4. Build your case file now. If your organization is in the 90% and the current strategy shifts — or if you're considering a move — a documented track record of AI-assisted productivity is the asset that travels with you.

The honest picture

This isn't evidence that AI doesn't work. The 1.8% aggregate gain is real, and the individual-level data on engaged users is better still. It's evidence that the strategy of using AI investment as cover for headcount reduction is failing — for the organizations pursuing it, for the executives responsible for the returns, and for the workers caught in the middle.

The companies getting value from AI are, almost uniformly, the ones making workers feel the tools are their advantage — not their replacement. That's not a soft management principle. It's what the Federal Reserve data shows.

Sources

• Federal Reserve Bank of Atlanta Working Paper: Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives (748 executives, Nov 2025–Jan 2026)
• Federal Reserve Bank of Atlanta Macroblog: How Might AI Change the Workplace? Evidence from Corporate Executives
• Fortune: 90% of executives say AI hasn't boosted productivity. Some are still cutting jobs — covers University of Pittsburgh study by Professor Mark Ma