Insights
How many hours does AI actually save a small business owner?
No study located for this article measures hours saved by Australian small business owners. The closest measured evidence comes from larger studies. A Danish survey of 25,000 workers found average self-reported savings of 2.8% of work hours, about 1.1 hours in a 40-hour week. Studies of specific tasks report gains of 14% to 40%.
What does the measured evidence show?
Four studies give the best available numbers. Two measured results directly. Two rely on what participants reported. The table lists the publisher, year, sample, method and limit of each.
| Study | Publisher and year | Sample | Method | Result | Limit |
|---|---|---|---|---|---|
| Humlum and Vestergaard, Large Language Models, Small Labor Market Effects | Becker Friedman Institute, University of Chicago, working paper 2025-56, May 2025 | 25,000 workers and 7,000 workplaces in 11 occupations, Denmark | Two surveys (late 2023 and 2024) linked to administrative employer-employee data | Average time savings of 2.8% of work hours, self-reported. The authors rule out average effects larger than 1% on earnings and hours | Self-reported. Danish workers in occupations chosen for AI exposure |
| Brynjolfsson, Li and Raymond, Generative AI at Work | National Bureau of Economic Research working paper 31161 (2023). Quarterly Journal of Economics 140(2), 2025 | 5,179 customer support agents | Staggered rollout of an AI assistant | 14% more issues resolved per hour on average. 34% for novice and low-skilled workers. Minimal impact on experienced workers | Customer support work. Large differences between workers |
| Noy and Zhang, Experimental evidence on the productivity effects of generative AI | Science, volume 381, 14 July 2023 | 453 college-educated professionals | Preregistered online randomised experiment with incentivised writing tasks of 20 to 30 minutes | Time taken fell 40%. Output quality rose 18% | Short, self-contained writing tasks. The authors note fact-checking and prompt writing reduce real-world speed |
| Evaluation of the whole-of-government trial of Microsoft 365 Copilot | Digital Transformation Agency with Nous Group, October 2024 | Over 7,600 staff in 60 or more agencies, January to June 2024. Sample size of survey respondents not captured | Participant surveys and agency evaluations | Respondents estimated Copilot saved up to an hour a day. Two thirds of participants used it a few times a week or less (iTnews) | Self-reported. Public servants. One product |
Which numbers are measured and which are reported?
Measured results come from output counts and graded work. Reported results come from what people estimate. The Brynjolfsson and Noy results are measured. The Humlum and Vestergaard result and the Copilot trial are self-reported.
Figures such as "AI saves 10 to 20 hours a week" circulate in vendor and agency content without a published method. This article excludes them.
What do the numbers mean for a small business owner?
Expect small average gains and larger gains on specific tasks. The Danish figure of 2.8% equals about 1.1 hours in a 40-hour week. Task studies show bigger effects where the work is writing or answering queries. The Copilot trial found users reported consistent quality and efficiency gains in summarising content, creating first drafts and searching for information.
Three findings should change your expectations.
- Gains are uneven. The customer support study found 34% for novice workers and minimal impact on experienced ones.
- New work appears. The Danish study found AI chatbots created new job tasks for 8.4% of workers.
- Time saved does not become income by itself. Humlum and Vestergaard estimate only 3% to 7% of gains passed through to earnings in their data.
How do you measure your own hours?
Measure three tasks over two weeks. This method is Clockwise's own and is not drawn from a study.
- Pick three recurring tasks. Use the task scoring method.
- Time five runs of each task without AI. Record minutes.
- Time five runs of each task with AI. Include prompting, checking and fixing.
- Score the quality of each result from 1 to 5 against the same checklist.
- Calculate minutes saved per run, then multiply by runs per week.
| Task | Minutes without AI (average of 5) | Minutes with AI including checking (average of 5) | Quality score without | Quality score with | Runs per week | Hours saved per week |
|---|---|---|---|---|---|---|
Include checking time in the AI column. Noy and Zhang note that fact-checking and prompting reduce real-world speed.
What to do next
Use your own numbers to decide where to spend more time. If you bill hourly, read how to price services when AI cuts your hours. The Clockwise Intensive is one focused hour with Megan, one-on-one on Google Meet, on where AI fits in your business and which process to hand over first. It costs $550 AUD.
Book the Clockwise Intensive →
Frequently asked questions
How many hours a week does AI save?
A Danish survey of 25,000 workers found average self-reported time savings of 2.8% of work hours, about 1.1 hours in a 40-hour week. Studies of specific tasks report larger gains, such as 14% more customer issues resolved per hour and 40% less time on writing tasks.
Is there Australian data on time saved by small businesses?
No measured Australian small business study was located for this article. The Australian Government's Copilot trial in 2024 collected self-reported estimates from public servants of up to an hour a day. That sample was public sector staff using one product.
About the author
Megan Hogan is the founder of Clockwise and works as a fractional COO inside small to medium businesses. She spent seven years at Optus and has led customer experience transformation and AI-enabled contact centre implementation.
Editorial standards
Each study is named with its publisher, year, sample and method. Measured results and self-reported results are labelled. Vendor claims without a published method are excluded. The measurement method is Clockwise's own. This article is general information.
Sources and dates
All checked 29 October 2026.
- Humlum, A. and Vestergaard, E., Large Language Models, Small Labor Market Effects. Becker Friedman Institute Working Paper 2025-56, 9 May 2025. https://bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2025-56-3.pdf
- Brynjolfsson, E., Li, D. and Raymond, L., Generative AI at Work. NBER Working Paper 31161 (2023). Quarterly Journal of Economics 140(2), 889-942 (2025). https://www.nber.org/papers/w31161
- Noy, S. and Zhang, W., Experimental evidence on the productivity effects of generative artificial intelligence. Science 381(6654), 187-192, 14 July 2023. https://www.science.org/doi/10.1126/science.adh2586
- Digital Transformation Agency, Evaluation of the whole-of-government trial of Microsoft 365 Copilot (evaluation by Nous Group). October 2024. https://www.digital.gov.au/initiatives/copilot-trial
- iTnews, Gov Copilot trial unsettled by usage metrics and unmet expectations. October 2024. https://www.itnews.com.au/news/gov-copilot-trial-unsettled-by-usage-and-unmet-expectation-612580
- MIT Economics, Study finds ChatGPT boosts worker productivity for some writing tasks. 14 July 2023. https://economics.mit.edu/news/study-finds-chatgpt-boosts-worker-productivity-some-writing-tasks
- Press summaries of the Humlum and Vestergaard study (The Register, as reproduced in a mailing-list digest, 2025), for the 8.4% new tasks and 3% to 7% pass-through figures. Re-confirm both in the working paper before publishing.