F-08 · What AI Saves
How much time does AI save knowledge workers?
Workers who use generative AI say it saves them about 3 to 5.4 percent of their work hours, roughly 1 to 2 hours of a 40-hour week. The one large trial that measured time found users spent 1.4 to 2 fewer hours a week on email, with no measurable change in time in meetings or writing documents.
3–5.4% of work hours, by the estimate of people who use AI (about 1 to 2 hours a week)
Across all 6 studies: about 1 to 9 hours a week, counting vendor surveys.
- Medium confidence
- 6 primary sources
- 2024 to 2026
- Last checked
Every study on one scale
Shaded: 3–5.4% of hours Dashed: with vendor surveys One figure A range in one study
Chart values as text
- Scale: hours a week saved (self-reported). Shaded range: 3–5.4% of hours; dashed, with vendor surveys: 1.2 h to 8.75 h.
- Bick, Blandin and Deming (2025): 2.2 h (5.4%)
- Humlum and Vestergaard, Denmark (2026): 1.2 h (3%)
- UK civil service (2025): 26 min a day (2.2 h)
- Microsoft (2024): 14 min a day (1.2 h)
- Atlassian (2024): 53–105 min a day (4.4–8.75 h)
- Workday (2026): 1–7 h (85% of users)
Why medium confidence: Three independent creators with different surveys (a US national panel, a Danish survey linked to government records, and the UK civil service) agree on the range. All of them are self-reports, and the only large measured trial covers email time only.
Sources
| Study | Creator | Year | Sample | Method | Figure | Link |
|---|---|---|---|---|---|---|
| The Rapid Adoption of Generative AI (NBER Working Paper 32966) | Alexander Bick (St. Louis Fed), Adam Blandin (Vanderbilt) and David Deming (Harvard) | 2025 | More than 10,000 US adults aged 18 to 64 in the Real-Time Population Survey; time question asked in November 2024 | Nationally representative online survey. Asks how many extra hours last week's work would have taken without AI. | 5.4% of work hours among people who used generative AI for work, about 2.2 hours of a 40-hour week; 1.4% across all workers “We estimate a mean time savings of 5.4 percent among all genAI work users,” | nber.org ↗ |
| Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777) | Anders Humlum (Chicago Booth) and Emilie Vestergaard (University of Copenhagen), with Statistics Denmark | 2026 | About 25,000 workers at about 7,000 workplaces in 11 occupations exposed to AI chatbots | Surveys linked to Danish payroll records. Time saved is self-reported in bands; earnings and paid hours come from the records. First circulated as "Large Language Models, Small Labor Market Effects". | About 3% of work hours for people who use AI chatbots (0.7% for teachers up to 6.1% for marketing staff); no change in earnings or paid hours in the records “On average, adopters in our sample report savings of about 3% of their work hours.” | nber.org ↗ |
| Microsoft 365 Copilot Experiment: Cross-Government Findings Report | UK Government Digital Service | 2025 | 7,115 survey responses from a trial with 20,000 civil servants | End-of-trial survey, October to December 2024, with answers in time bands averaged at their midpoints. No control group for the time question. | 26 minutes a day saved on average; 17% noticed no clear time saving “Trial participants saved an average of 26 minutes a day when using M365 Copilot.” | assets.publishing.service.gov.uk ↗ |
| AI Data Drop: The 11 by 11 Tipping Point | Microsoft WorkLab | 2024 | 1,300 Microsoft 365 Copilot users, including Microsoft employees, after at least three weeks of use | Self-reported daily time saved, in answer bands. Microsoft sells Copilot. | 14 minutes a day saved on average; 30 minutes for the most efficient users “the average person saved 14 minutes a day, or nearly five hours each month” | microsoft.com ↗ |
| AI Collaboration Report | Atlassian Teamwork Lab | 2024 | About 5,000 knowledge workers in Australia, the US, India, Germany and France | Online survey, self-estimated time saved, grouped by how advanced people's AI use is. Atlassian sells AI-enabled software. | 53 minutes a day for simple AI users, 105 minutes for the most advanced “We asked people to estimate the amount of time that AI saves them” | atlassian.com ↗ |
| Beyond Productivity: Measuring the Real Value of AI | Workday, survey by Hanover Research | 2026 | 3,200 full-time employees who use AI, at organizations with $100 million or more in revenue | Online survey, November 2025, self-reported. Read in Workday's press release; the full report is gated. Workday sells AI products. | 85% say they save 1 to 7 hours a week, and that nearly 40% of the savings is lost to rework “85% of employees report saving one to seven hours per week using AI” | newsroom.workday.com ↗ |
How much of all work hours does AI save across the economy?
1.4 to 2.2 percent of all US work hours, counting the many workers who do not use AI. The share rose from 1.41% in late 2024 to 2.17% in the second quarter of 2026, as the share of employed adults who used AI for work in a given week grew from 28% to 39%. Both figures are self-reported, from the St. Louis Fed's survey series.
| Study | Creator | Year | Sample | Method | Figure | Link |
|---|---|---|---|---|---|---|
| Does generative AI save time at work? (FRED Blog, series RPSGENAITSALL) | Federal Reserve Bank of St. Louis, data series by Bick, Blandin and Deming | 2026 | Employed US adults aged 18 to 64, quarterly survey waves | Same self-reported question as the NBER paper, with non-users counted as zero | 1.41% of all US work hours saved (Q4 2024) rising to 2.17% (Q2 2026) “the share of work hours saved thanks to AI assistance grew from 1.6% to 2.2%.” | fred.stlouisfed.org ↗ |
What does measured data show about time saved?
In a randomized trial of Microsoft 365 Copilot with 7,137 workers at 66 firms, people who used it spent about 2 fewer hours a week on email (1.4 hours averaged over everyone offered the tool), out of about 11.7. Time in Teams meetings and in Word did not change measurably, and they replied to as many email threads and finished as many documents. In Denmark, payroll records show no change in paid hours or earnings, which fits saved time going into other work.
| Study | Creator | Year | Sample | Method | Figure | Link |
|---|---|---|---|---|---|---|
| Shifting Work Patterns with Generative AI (NBER Working Paper 33795) | Dillon, Jaffe and Immorlica (Microsoft Research) and Stanton (Harvard Business School) | 2025 | 7,137 knowledge workers at 66 large firms | Randomized field experiment, 2023 to 2024, timed with Microsoft 365 usage data. Most authors work at Microsoft. | 1.4 fewer hours a week on email for everyone offered Copilot (12%), about 2 fewer for those who used it (17%); no significant change in meeting or document time “who used this tool spent two fewer hours on email each week” | nber.org ↗ |
Are people's own estimates of AI time savings accurate?
Often not, and they miss in both directions. In a Microsoft lab test, people guessed AI saved them 36 minutes when the measured saving was 12. In METR's 2025 trial, experienced developers felt 20% faster with AI but took 19% longer. In a 2022 GitHub Copilot test, developers guessed a 35% gain when the measured one was 55.8%.
| Study | Creator | Year | Sample | Method | Figure | Link |
|---|---|---|---|---|---|---|
| Early LLM-based Tools for Enterprise Information Workers Likely Provide Meaningful Boosts to Productivity (MSR-TR-2023-43) | Microsoft Research | 2023 | Several lab studies and an early-access survey | Lab tasks timed and compared with participants' own estimates. Tasks were chosen because AI was likely to help. | Participants guessed 36 minutes saved; the measured saving was 12 minutes “They guessed 36 minutes on average, when actual time savings were 12 minutes” | microsoft.com ↗ |
| Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | Joel Becker, Nate Rush, Beth Barnes and David Rein (METR) | 2025 | 16 experienced open-source developers, 246 tasks | Randomized trial at the task level, compared with developers' own estimates afterwards | Tasks took 19% longer with AI; developers believed they were 20% faster “we find that allowing AI actually increases completion time by 19%” | arxiv.org ↗ |
| The Impact of AI on Developer Productivity: Evidence from GitHub Copilot | Sida Peng (Microsoft Research), Eirini Kalliamvakou and Peter Cihon (GitHub), Mert Demirer (MIT Sloan) | 2023 | 95 freelance developers recruited on Upwork | Controlled experiment, 2022, with an exit survey | Developers estimated a 35% gain; the measured time saving was 55.8% “participants in both treated and control groups estimated a 35% increase in productivity” | arxiv.org ↗ |
Numbers you may have seen
These figures circulate widely. Here is what the original source says.
- Generative AI saves 5.4% of all work hours. 5.4% is the average among people who used AI for work. Across all workers it was 1.4% in November 2024 and 2.2% by mid-2026.
- Danish study: AI's measured time savings are only 2.8%. The 2.8% (about 3% in the 2026 revision) is what workers report in a survey. What the authors measure in payroll records is earnings and paid hours, where they find no change.
- Thomson Reuters: AI saves professionals 5 hours a week. This is a forecast. Respondents predicted AI would save them 5 hours a week within the next year (4 hours in the 2024 edition).
- Microsoft: Copilot saves 11 minutes a day. In Microsoft's survey, 11 minutes a day is the point at which users start to feel value. The average reported saving was 14 minutes.
- Microsoft Work Trend Index 2024: AI saves workers time, 90% say. 90% agree AI helps them save time. The report gives no amount of time saved.
- Slack: daily AI users are 64% more productive. Slack's 2025 survey found daily users 64% more likely to rate their productivity "very good". It is a gap in self-ratings, with no time or output measured.
More from What AI Saves
- F-0925–40%How much faster do people finish writing tasks with AI, and is the work better?
- F-1014–15%How much more do customer support agents get done with AI?
- F-1121–26%How much faster do developers code with AI assistants?
- F-1227–46%Does AI help less-experienced workers more than experienced ones?
- F-1314–25%When does AI make work slower or worse?
The whole chapter · Chapter poster · All of The AI at Work Fact Book