Federal Reserve data shows AI saves workers 5.4% of work hours on average (2.2 hours/week for full-time). Gains vary wildly by profession: tech workers save 2.5%, service workers save 0.4%. Most dramatic claims (40%+ savings) come from short-term studies or vendor research.
Vendors claim 40% productivity gains. Headlines promise AI will revolutionize work. But what does independent research actually show? We dug into data from the Federal Reserve, MIT, Stanford, and industry surveys to find out.
The Federal Reserve Numbers
The St. Louis Fed ran comprehensive surveys in late 2024, asking workers how much time AI actually saves them. The results are more modest than headlines suggest.
For a 40-hour work week, that's 2.2 hours saved. Meaningful, but not the transformation some expected. And this is only among the 28% of workers who use AI at all.
Savings Vary Dramatically by Job Type
The Fed found huge differences across professions:
- Computer/Math occupations: 2.5% time savings (used AI in 12% of work hours)
- Business/Financial: 1.8% savings
- Management: 1.5% savings
- Personal services: 0.4% savings (used AI in only 1.3% of work hours)
The pattern is clear: knowledge workers in tech-adjacent roles see the biggest gains. Service workers see almost none.
The MIT/Stanford Findings
This study focused on customer support workers using AI chat tools. Notably, the biggest gains went to lower-skilled workers, while experienced workers saw smaller improvements.
"AI triples productivity on one-third of tasks, reducing a 90-minute task to 30 minutes. But these gains don't apply uniformly across all work."
— MIT/Stanford Study
The Surprising Developer Study
Here's where it gets interesting. A 2025 METR study tracked experienced developers using AI coding assistants on their own projects. The result?
Yes, slower. Developers expected AI to speed them up by 24%. Even after experiencing the slowdown, they still believed AI had helped by 20%. The gap between perception and reality is striking.
Self-reported AI productivity gains are often inflated. When possible, look for randomized controlled studies with actual time measurements.
Industry Survey Data
Vendor-adjacent research tends to show higher gains:
- Upwork: 40% productivity boost (self-reported by AI users)
- Microsoft: 11 minutes saved daily with Copilot
- Accenture: $4.40 return per $1 spent on AI
- McKinsey: 70% time savings on specific content tasks
The difference between the Fed's 5.4% and industry's 40% tells you something about who's funding the research.
What This Means for ROI Calculations
When calculating expected AI savings, here's a realistic framework:
- Conservative estimate: 5-10% time savings (Fed data)
- Moderate estimate: 15-25% on specific tasks (MIT/Stanford)
- Optimistic estimate: 40-50% on highly automatable tasks (industry benchmarks)
Most calculator tools use the optimistic numbers. That's fine for quick estimates, but budget for the conservative scenario when making purchasing decisions.
The Adoption Gap
Three-quarters of workers haven't adopted AI tools yet. For those who have, consistent daily use matters more than which tool they chose. Occasional users see minimal gains.
The Realistic Picture
AI tools can deliver real productivity gains. But the gains are task-specific, require consistent use, and take time to materialize. Expect modest overall savings (5-15%) with potential for higher gains (30-50%) on specific highly-repetitive tasks.
The headline-grabbing 40%+ claims usually come from short-term studies on specific tasks in controlled conditions. Real-world results across a full workday are more modest.
The TaskROI team researches AI productivity tools and helps businesses calculate real ROI before purchasing. Our data comes from industry studies by McKinsey, Harvard Business Review, and the Federal Reserve.