What happened
The Department for Work and Pensions trialed the licensed version of Microsoft 365 Copilot, a generative AI tool, from October 2024 to March 2025. It allocated 3,549 licenses to central office staff. DWP's own analysts published their evaluation on 29 January 2026. They surveyed all license holders in February 2025, with 1,716 responses, and a stratified random sample of staff without licenses in March 2025, with 2,535 responses. A Seemingly Unrelated Regression compared the two groups on task efficiency, job satisfaction and work quality. With controls for grade, gender, age, job category, health condition and interest in AI, license holders saved an estimated 19 minutes a day across eight routine tasks. The 95% confidence interval was 17 to 22 minutes. The largest savings were 26 minutes on searching for information, 25 minutes on writing emails and 24 minutes on summarizing. Job satisfaction rose by 0.56 points and work quality by 0.49 points on a 7-point scale. The evaluators note that licenses were not allocated at random, that time was self-reported and that there was no pre-trial baseline.
Sources
-
Evaluation report, by the organization that used it.
Arzilli, F., Lynch, C. S. and Page, L., Department for Work and Pensions. An Evaluation of DWP's Microsoft 365 Copilot Trial. Research and analysis, GOV.UK, 29 January 2026.
The analysis estimated that users in the treatment group saved an average of 19 minutes per day across eight routine tasks when controlling for all demographic, occupational, and
Details
AI system
- Workflow
-
DWP allocated 3,549 Copilot licenses to staff in central office roles through volunteers and peer nominations, in 12 license groups by business area. Most users learned the tool by trial and error, online resources and peer support, and nearly two-thirds used it daily. The evaluation compared license holders with a stratified random sample of non-users and controlled for grade, gender, age, job category, health condition and interest in AI.
Follow-up
- Conditions for replication
-
Users must still check and edit Copilot's outputs, which they described as a starting point. Staff need training and role-specific guidance to use the tool well. The estimate rests on self-reported time and a self-selected treatment group, so it may not hold for the wider workforce.
Entry details
- Sector
-
Public administration.
- Purpose
-
Microsoft built Microsoft 365 Copilot as a generative AI assistant inside its Microsoft 365 applications, to draft, summarize and search for its users.
30 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this benefit
Peter H. Vartanian. (2026). Microsoft 365 Copilot saves DWP staff 19 minutes a day in a trial (AIBD-1195). Artificial Intelligence Benefits Database, edition of 2 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1195.html (archived at https://doi.org/10.7910/DVN/AADSII, version 9.0)