Do AI assistants save time on everyday office work?
No good evidence
We show no impact figure until at least two studies qualify.
There is not enough evidence to rate this yet.
Business plans of AI assistants list about $20–$30 a user a month (checked 2 Nov 2026), so about $240–$360 a year before checking time.
What the studies found
No impact figure yet. Fewer than two studies qualify, so under our rules we do not show how big the effect is. The studies we found are listed below, with what each measured.
How sure are we?
- Starting level1 of 5, from 4 studies, including grade D.
- Down one: old models0 of 4 studies show they used a model released since May 2025 (the 18-month window); the rest used older models or do not say which.
- Down one: self-reportedThe headline outcome is mostly self-reported rather than measured.
- Publication biasnot testable with fewer than 10 studies.
- Result0 of 5. There is not enough evidence to rate this yet.
Behind the average
- Most vs least affected quintiles (machine-learning CATEs) (Dillon et al. (2025)): Outlook time -1.65 vs -1.18 hours a week; equal effects and zero heterogeneity cannot be rejected (section 2.2.1)
- Junior APS3-6, EL1 and ICT roles (Digital Transformation Agency (2024)): Perceived the most efficiency gains (about an hour a day on summarising and first drafts) (Executive summary)
How, where and why it works
1 of 4. Studies show where time moves (less email, faster documents), not why.
1 of 4. Few studies split results by role or task.
2 of 4. Government evaluations report training, licences and policy steps in some detail.
1 of 4. Licence costs are known; time spent checking output is rarely measured.
What to do next
Measure your pilot →If you are trying it yourself
- Pick two or three repeated tasks and time them with and without the assistant.
- Check every summary and draft before it goes out.
If you are rolling it out
- Measure time from systems (email, documents), not only surveys.
- Run a comparison group if you can.
- Use the rollout kit's measurement section.
What it takes
- Give people training and examples for their own tasks.
- Set clear rules on what information can go into the assistant.
- Expect take-up to vary a lot between people and teams.
What can go wrong
- Self-reported time savings are often larger than measured ones.
- Errors in drafts or summaries can spread if nobody checks them.
The studies
| Study | Grade | Design | People | AI model (released) | Company role |
|---|---|---|---|---|---|
| Shifting Work Patterns with Generative AI Eleanor W. Dillon, Sonia Jaffe, Nicole Immorlica, Christopher T. Stanton, 2025. NBER Working Paper 33795 (working paper) | B | field RCT, deployed work | 7,137 knowledge workers (3,684 treated) at 66 large firms; about 6,441 workers in the main Outlook regressions | Microsoft 365 Copilot (Outlook, Teams, Word, Excel, PowerPoint, M365 Chat) | vendor-run: Microsoft recruited the firms and ran the experiment through its Customer Research Program |
| Microsoft 365 Copilot Experiment: Cross-Government Findings Report Government Digital Service (UK government), 2025. GOV.UK corporate report, June 2025 (government report) | D | cross-sectional, deployed work | 20,000 licensed civil servants in 12 organisations; 7,115 end-of-trial survey responses; usage data for 14,500 users; 5 focus groups | Microsoft 365 Copilot (incl. Copilot Chat) | none |
| Evaluation of the whole-of-government trial of Microsoft 365 Copilot: full report Digital Transformation Agency (Australian Government), with Nous Group, 2024. digital.gov.au, October 2024 (government report) | D | cross-sectional, deployed work | Over 5,765 licences (executive summary; Appendix A says over 7,769) across almost 60 agencies; evaluation reached over 50 agencies and 2,000+ participants; surveys 1,556 pre-use, 1,159 pulse, 831 post-use (330 linked); 17 focus groups; 32 interviews (24 DTA, 8 Nous) | Microsoft 365 Copilot (formerly Copilot for Microsoft 365) | none |
| An Evaluation of DWP's Microsoft 365 Copilot Trial Francesco Arzilli, Callum Seán Lynch, Lucy Page (Department for Work and Pensions), 2026. GOV.UK research and analysis, published 29 January 2026 (government report) | D | cross-sectional, deployed work | 3,549 licensed central-office staff; surveys of 1,716 Copilot users and 2,535 non-users (stratified random sample); 19 interviews | Microsoft 365 Copilot (licensed version) | none |
Does the rating hold up?
- Run without company-run studies3 studies: strength 0, verdict no good evidence, headline band no clear effect.
- Run without working papers and preprints3 studies: strength 0, verdict no good evidence, headline band no clear effect.
How up to date is this evidence?
Method →Evidence searched to 2 Nov 2026. The studies do not all state model release dates. Rating computed 5 Nov 2026 under Module 15 v0.3 (pilot).
How this page was made: AI research agents searched for studies and read each paper; a second pass checked every number on this page against the paper itself, with the quote and where it appears. The rating is computed by fixed rules, not chosen. No person outside AI Adoption has reviewed it yet. Editorial owner: AI Adoption editors. Spot a mistake? Corrections.
Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.