Operator Insights

What are business and product leaders learning from rolling Microsoft Copilot out across their teams?

19 accounts: 2 failures, incidents or rollbacks, 2 run it beside another tool, 11 evaluated it, 3 adopted it and 1 reported by others

Failures, incidents and rollbacks2

Australian Centre for EvaluationEvaluation authorsAustralian Treasury
Organization-wide. Australian Treasury. 218 staff, May 20 to August 23, 2024, report published February 2025.
Australian Treasury, Australian Centre for Evaluation, published Feb 5, 2025Read the report (opens Australian Treasury, Australian Centre for Evaluation in a new tab)
Observation
Australia's Treasury found Copilot easy to deploy, and the trial failed to meet staff's expectations.
Takeaway
Budget training time before the trial starts; a smooth install does not deliver use.
What happened
Treasury's 14-week trial with 218 staff found the technical rollout smooth. Use was lower than expected: most participants used Copilot two to three times a week or less. Training and time to learn were limited, and staff had high expectations the product did not meet. Almost two-thirds found it useful for basic administrative tasks, while product limits and privacy and security restrictions held back complex work. The survey table computes response rates from 232 staff, a figure the report does not reconcile with 218.
“Implementation: the technical implementation was smooth, however, training and time to learn to use Copilot was limited, and participants had high expectations of the product which were not met.”
Australian Centre for Evaluation
- Public organization write-up (government evaluation report)
- Usage and expectations self-reported
Deployment details
Organization-wide. Australian Treasury. 218 staff, May 20 to August 23, 2024, report published February 2025.
- Public organization write-up (government evaluation report)
- Usage and expectations self-reported
Team
218 Treasury staff
Duration
14 weeks, May 20 to August 23, 2024
Workload
Administrative tasks, drafting and summaries
Product version
Microsoft 365 Copilot
Measurement
Usage frequency and staff ratings

Tony RedmondFeb 13, 2026Office 365 for IT ProsA Copilot Chat bug let confidential email past a DLP policy in Tony Redmond's own tenant until Microsoft fixed it in February 2026.
Tony RedmondLead author, Office 365 for IT ProsOffice 365 for IT Pros
Individual tenant operator. A Microsoft 365 tenant protected by a DLP policy for Copilot.
Its own announcement, published Feb 13, 2026Read the write-up (opens the announcement in a new tab)
Observation
A Copilot Chat bug let confidential email past a DLP policy in Tony Redmond's own tenant until Microsoft fixed it in February 2026.
Takeaway
Treat a DLP policy for Copilot as one layer: keep restricted content out of Copilot's reach and watch Microsoft's service advisories after you enable it.
What happened
Under Microsoft advisory CW1226324, a code issue let Copilot Chat answer from Sent Items and Drafts stamped with a confidential sensitivity label, although a DLP policy was built to block exactly that. Redmond confirmed his own tenant was affected and reported on February 13, 2026 that Microsoft's fix was rolling out. Customers first reported the incident on January 21, 2026.
“An embarrassing security glitch appeared in Microsoft 365 when the DLP policy constructed to stop Copilot Chat processing email and files stamped with specific sensitivity labels failed to suppress confidential material appearing in Copilot responses.”
Tony Redmond
Public practitioner write-up
The author reports an affected tenant
Deployment details
Individual tenant operator. A Microsoft 365 tenant protected by a DLP policy for Copilot.
Public practitioner write-up
The author reports an affected tenant
Team
The author's own Microsoft 365 tenant
Duration
January 21 to February 2026
Workload
Copilot Chat over mail labelled confidential
Product version
Microsoft 365 Copilot Chat with a Purview DLP policy

Run it beside another tool2

Ministry of JusticePublishing bodyUK Ministry of Justice
Organization-wide. UK justice department. One-year update of its AI Action Plan, September 24, 2026.
GOV.UK (Ministry of Justice), published Sep 24, 2026Read the report (opens GOV.UK (Ministry of Justice) in a new tab)
Observation
The UK Ministry of Justice runs targeted rollouts of Copilot and ChatGPT Enterprise in parallel.
Takeaway
The Ministry of Justice is rolling out both assistants to targeted staff; plan procurement and data rules for each tool your roles need.
What happened
The Ministry of Justice's one-year update on its AI Action Plan, published September 24, 2026, describes targeted, ongoing rollouts of both Copilot for Microsoft 365 and ChatGPT Enterprise to staff. It gives no licence counts or measured results for either.
“…targeted, ongoing rollouts of Copilot for Microsoft 365 and ChatGPT Enterprise…”
Ministry of Justice
- Public organization write-up (government progress update)
Deployment details
Organization-wide. UK justice department. One-year update of its AI Action Plan, September 24, 2026.
- Public organization write-up (government progress update)
Duration
Ongoing at September 2026
Product version
Copilot for Microsoft 365 and ChatGPT Enterprise

Yale UniversityOct 28, 2026Yale University (AI at Yale)Yale's AI tools catalog lists Microsoft 365 Copilot and Claude Enterprise side by side, each with its own price and data classification.
AI at YaleUniversity IT and AI servicesYale University
Organization-wide. Yale faculty and staff, as Yale's AI tools catalog lists its assistants (undated page, read Oct 28, 2026).
Yale University (AI at Yale), published Oct 28, 2026Read the write-up (opens Yale University (AI at Yale) in a new tab)
Observation
Yale's AI tools catalog lists Microsoft 365 Copilot and Claude Enterprise side by side, each with its own price and data classification.
Takeaway
Yale lists free Copilot Chat, paid Microsoft 365 Copilot on request and Claude Enterprise beside each other: write down which data class each assistant is approved for before you add a second one.
What happened
Yale's AI tools catalog lists Microsoft 365 Copilot for staff and faculty at US$18 per user per month, on request, at a High data classification, with Copilot Studio included in the licence at High with exceptions such as ePHI. Beside it the catalog lists Claude Enterprise at US$20 a month for a Standard licence or US$50 for Premium, with Claude Chat, Code and Cowork, at High with a special environment for ePHI, where Cowork is not available. Copilot Chat and Google Gemini for Education are free.
Public organization write-up (Yale's AI tools catalog, undated, read Oct 28, 2026)
Prices and data classifications stated by Yale
Deployment details
Organization-wide. Yale faculty and staff, as Yale's AI tools catalog lists its assistants (undated page, read Oct 28, 2026).
Public organization write-up (Yale's AI tools catalog, undated, read Oct 28, 2026)
Prices and data classifications stated by Yale
Team
Yale staff and faculty who request a licence
Workload
Copilot inside Microsoft 365 apps; Claude for chat, code and Cowork, with an ePHI environment
Product version
Microsoft 365 Copilot, Copilot Studio, Copilot Chat, Claude Enterprise, Gemini for Education

Evaluated it11

HMRC evaluation teamEvaluation authorsHM Revenue & Customs
Organization-wide. UK tax authority. Phase 3 trial, September to December 2024, published July 9, 2026.
GOV.UK (HM Revenue & Customs), published Jul 9, 2026Read the report (opens GOV.UK (HM Revenue & Customs) in a new tab)
Observation
HMRC randomly allocated 3,000 Copilot licences, measured about an hour a week saved, then scaled past 28,000 licences.
Takeaway
Random licence allocation tells you who uses Copilot, and it still leaves the hour a week as a self-reported estimate until you measure output against unlicensed staff.
What happened
HMRC randomly allocated 3,000 full Microsoft 365 Copilot licences in proportion to its business groups and gave 500 more directly. 83% of staff given a licence used it. A survey of 1,364 users put the saving at 2 to 3% of the working week, about 60 minutes for a full-time worker, after HMRC cut the self-reported figures by about 20% because not every licence holder used Copilot and active users answered more often. 46% of non-users named security and data privacy as their main reason. HMRC's July 27, 2026 roadmap update says it had issued over 28,000 licences by March 2026 and plans 50,000 during 2026.
“The figures therefore should be treated as a baseline estimate rather than a precise measure of impact…”
HM Revenue & Customs
- Public organization write-up (government evaluation report)
- Self-reported time savings, reduced by HMRC
Deployment details
Organization-wide. UK tax authority. Phase 3 trial, September to December 2024, published July 9, 2026.
- Public organization write-up (government evaluation report)
- Self-reported time savings, reduced by HMRC
Team
3,000 randomly allocated and 500 directly allocated licence holders; over 28,000 licences by March 2026
Duration
September to December 2024 for the trial
Workload
Everyday Microsoft 365 work: meetings, email, documents
Product version
Microsoft 365 Copilot full licences
Measurement
Usage (83%), self-reported time saved (about 60 minutes a week), and reasons for non-use

UK Department for Business and TradeAug 28, 2025GOV.UK (Department for Business and Trade)The Department for Business and Trade found high satisfaction with Copilot, and its evaluators found no evidence that the time saved improved productivity.
Digital Data and Technology Monitoring and Evaluation teamEvaluation authorsUK Department for Business and Trade
Organization-wide. UK government department. 1,000 licences, October to December 2024, report published August 28, 2025.
GOV.UK (Department for Business and Trade), published Aug 28, 2025Read the report (opens GOV.UK (Department for Business and Trade) in a new tab)
Observation
The Department for Business and Trade found high satisfaction with Copilot, and its evaluators found no evidence that the time saved improved productivity.
Takeaway
DBT's evaluators found no evidence that time saved on summaries improved productivity, so test the tasks your roles actually do, Excel included, before licensing widely.
What happened
DBT gave 1,000 staff Microsoft 365 Copilot licences for three months. 72% of survey respondents were satisfied. In observed sessions with 11 participants, Copilot users summarized a report in 12 minutes 37 seconds against 41 minutes 34 seconds without it, and took 25 minutes 1 second on an Excel analysis against 20 minutes 33 seconds, with less accurate answers. The staff diary recorded extra time on scheduling and images. The evaluators found no evidence that the time saved improved productivity, and the six control group staff they spoke to had not seen productivity gains in colleagues; the control group took part only in conversations with evaluators and observed tasks.
“The evaluation did not find evidence that time savings have led to improved productivity, and control group participants had not observed productivity improvements from colleagues taking part in the M365 Copilot pilot.”
Department for Business and Trade evaluation team
Public organization write-up (government evaluation report)
Observed tasks for 11 staff; other figures self-reported
Deployment details
Organization-wide. UK government department. 1,000 licences, October to December 2024, report published August 28, 2025.
Public organization write-up (government evaluation report)
Observed tasks for 11 staff; other figures self-reported
Team
1,000 licensed staff; 19 spoke to evaluators, 13 users and 6 from the control group
Duration
Three months, October to December 2024
Workload
Report summaries, email, Excel analysis, PowerPoint, scheduling, images
Product version
Microsoft 365 Copilot
Measurement
Satisfaction (72%), timed observed tasks, staff diary
Department of Health and Social Care (UK)Oct 21, 2025GOV.UKNHS England is rolling Copilot out to 505,000 staff on the strength of a pilot whose method was never published.
Dr Zubir AhmedHealth Innovation MinisterDepartment of Health and Social Care (UK)
Organization-wide. 90 NHS organizations in the pilot, more than 30,000 staff. National rollout announced June 8, 2026.
GOV.UK, published Oct 21, 2025Read the report (opens GOV.UK in a new tab)
Observation
NHS England is rolling Copilot out to 505,000 staff on the strength of a pilot whose method was never published.
Takeaway
Point paid licences at the meeting and email load you can measure, and treat the NHS's 43 minutes as a target to test: no method stands behind it.
What happened
On October 21, 2025 the Department of Health and Social Care reported a Microsoft 365 Copilot pilot across 90 NHS organizations with more than 30,000 staff, saving on average 43 minutes per person per day, and projected 400,000 hours a month at full scale, largely from meeting notes and email summaries. More than 50,000 staff used Microsoft 365 Copilot at that date, and Copilot Chat is available across the NHS at no extra cost. On June 8, 2026 NHS England announced Microsoft 365 Copilot for 505,000 clinicians and support staff, citing the same pilot. Office Watch reported on September 2, 2026 that the methodology was never published and that an FOI release puts the figure on 86 case studies, 69 in another place.
“A groundbreaking pilot of Microsoft 365 Copilot across 90 NHS organisations found that AI-powered administrative support could save NHS staff on average 43 minutes per staff member per day or more…”
Department of Health and Social Care and NHS England
Public organization write-up (government press release)
Savings reported and projected by the NHS; no method published
Deployment details
Organization-wide. 90 NHS organizations in the pilot, more than 30,000 staff. National rollout announced June 8, 2026.
Public organization write-up (government press release)
Savings reported and projected by the NHS; no method published
Team
More than 30,000 staff across 90 NHS organizations in the pilot; 505,000 staff in the 2026 rollout
Workload
Teams meeting notes, email chain summaries, and administrative tasks
Product version
Microsoft 365 Copilot, with Copilot Chat across the NHS
Measurement
Average minutes saved per person per day, plus projected monthly hours; no method published
Department for Work and Pensions (UK)Jan 29, 2026GOV.UK (Department for Work and Pensions)DWP compared licensed staff with non-users and estimated 19 minutes a day saved, led by search and email.
Francesco ArzilliEconomist and evaluation authorDepartment for Work and Pensions (UK)
Organization-wide. UK government department. 3,549 licensed staff, October 2024 to March 2025, report published January 29, 2026.
GOV.UK (Department for Work and Pensions), published Jan 29, 2026Read the report (opens GOV.UK (Department for Work and Pensions) in a new tab)
Observation
DWP compared licensed staff with non-users and estimated 19 minutes a day saved, led by search and email.
Takeaway
Compare licensed staff with unlicensed staff on the same tasks, and expect the estimate to move with what you control for.
What happened
DWP trialled Microsoft 365 Copilot with 3,549 staff. Its regression compared 1,716 survey responses from users with 2,535 from non-users and estimated 19 minutes saved per day once it controlled for attitudes to AI, and 10 to 12 minutes without those controls. Searching for information (26 minutes) and writing email (25 minutes) saved the most. 90% of users said Copilot helped them save time.
“The Seemingly Unrelated Regression (SUR) analysis estimated an average saving of 19 minutes per day.”
Department for Work and Pensions
Public organization write-up (government evaluation report)
Self-reported times compared with non-users by regression
Deployment details
Organization-wide. UK government department. 3,549 licensed staff, October 2024 to March 2025, report published January 29, 2026.
Public organization write-up (government evaluation report)
Self-reported times compared with non-users by regression
Team
3,549 licensed staff; 1,716 user and 2,535 non-user survey responses
Duration
October 2024 to March 2025
Workload
Information search, email, summaries and other routine office tasks
Product version
Microsoft 365 Copilot
Measurement
Minutes saved per day by task, estimated by regression
University of North Carolina at CharlotteJul 15, 2026arXivAt the North Carolina Department of Transportation, staff rated Copilot less useful after an eight-week pilot than before it.
Omidreza ShoghliResearcher, UNC Charlotte, studying the North Carolina Department of Transportation's Copilot pilotUniversity of North Carolina at Charlotte
Organization-wide. State transportation agency. About 175 employees in an eight-week pilot in 2025, paper published July 15, 2026.
arXiv, published Jul 15, 2026Read the report (opens arXiv in a new tab)
Observation
At the North Carolina Department of Transportation, staff rated Copilot less useful after an eight-week pilot than before it.
Takeaway
Expect usefulness ratings to fall once staff use Copilot on real work; measure after the pilot, never only at launch.
What happened
NCDOT ran an eight-week Microsoft 365 Copilot pilot with about 175 employees and surveyed the same people before training and after the pilot. Perceived usefulness fell from 3.85 to 3.62 on a five-point scale, a significant drop, while ease of use, intention to use and trust held steady.
“In 2025, NCDOT launched an eight-week pilot of Microsoft 365 Copilot involving approximately 175 employees…”
Omidreza Shoghli, Fatemeh Banani Ardecani, Amin Mohamadi Hezaveh
Public organization write-up (research paper with an NCDOT coauthor)
Matched before and after survey
Deployment details
Organization-wide. State transportation agency. About 175 employees in an eight-week pilot in 2025, paper published July 15, 2026.
Public organization write-up (research paper with an NCDOT coauthor)
Matched before and after survey
Team
About 175 employees
Duration
Eight weeks in 2025
Workload
Everyday Microsoft 365 work
Product version
Microsoft 365 Copilot
Measurement
Perceived usefulness, ease of use, intention and trust, before and after
UK Department for Science, Innovation and TechnologyJun 2, 2025GOV.UK (Government Digital Service)Across 20,000 UK civil servants, users reported saving 26 minutes a day, a figure GDS built from self-reported bands.
Government Digital ServiceExperiment organizerUK Department for Science, Innovation and Technology
Organization-wide. 12 UK government organizations. September 30 to December 31, 2024, published June 2, 2025.
GOV.UK (Government Digital Service), published Jun 2, 2025Read the report (opens GOV.UK (Government Digital Service) in a new tab)
Observation
Across 20,000 UK civil servants, users reported saving 26 minutes a day, a figure GDS built from self-reported bands.
Takeaway
Read 26 minutes as a self-reported mean estimated from time-saved bands, and budget for the change management that kept adoption near 80%.
What happened
GDS ran a cross-government experiment with 20,000 Microsoft 365 Copilot licences across 12 organizations. An end-of-trial survey drew 7,115 responses; users picked a daily time-saved band, GDS took each band's midpoint with the top band set at 60 minutes, and reported 26 minutes a day. Adoption reached 83% after October's change management and held near 80%. Telemetry covered 14,500 users.
“The time savings presented throughout are self reported through the quantitative portion of the data capture.”
Government Digital Service
Public organization write-up (government evaluation report)
Self-reported survey bands
Deployment details
Organization-wide. 12 UK government organizations. September 30 to December 31, 2024, published June 2, 2025.
Public organization write-up (government evaluation report)
Self-reported survey bands
Team
20,000 licensed staff in 12 organizations; 7,115 survey responses
Duration
September 30 to December 31, 2024
Workload
Meetings, email, documents and search
Product version
Microsoft 365 Copilot
Measurement
Self-reported minutes saved per day, adoption rate, telemetry
University of WashingtonNov 15, 2024University of Washington UW-ITThe University of Washington piloted Copilot for five months, and the pilot recommended against funding it centrally.
UW-IT and UW Medicine pilot teamPilot ownersUniversity of Washington
Organization-wide. UW Academy and UW Medicine. 200 pilot users over five months, report dated November 15, 2024.
University of Washington UW-IT, published Nov 15, 2024Read the report (opens University of Washington UW-IT in a new tab)
Observation
The University of Washington piloted Copilot for five months, and the pilot recommended against funding it centrally.
Takeaway
Fund licences from departments that show use, and fix oversharing first; UW's pilot recommended against central funding on both counts.
What happened
UW-IT and UW Medicine ran a five-month pilot with 200 users and analyzed 425 prompt tests. At the close, half of participants used Copilot only occasionally or not at all, and of the 100 who answered the closing survey, 54% of those who saved time saved an hour or more a week. UW recommended against central funding, citing value and the risk that existing data oversharing would be unacceptable.
“Of those who reported saving time, over half (54%) reported saving an hour or more…”
UW-IT and UW Medicine
Public organization write-up (university IT report)
Prompt tests and a closing survey
Deployment details
Organization-wide. UW Academy and UW Medicine. 200 pilot users over five months, report dated November 15, 2024.
Public organization write-up (university IT report)
Prompt tests and a closing survey
Team
200 pilot users across UW Academy and UW Medicine
Duration
Five months, ending October 31, 2024
Workload
425 tested prompts across office tasks
Product version
Microsoft 365 Copilot
Measurement
Prompt pass rate, self-reported time saved, ongoing use
Ohio UniversityAug 28, 2025Ohio University NewsOhio University's yearlong Copilot pilot found steady savings for only part of its 130 users.
Ohio University Office of Information TechnologyUniversity central ITOhio University
Organization-wide. About 130 staff across planning units. Yearlong pilot ended May 30, 2025, reported August 28, 2025.
Ohio University News, published Aug 28, 2025Read the write-up (opens Ohio University News in a new tab)
Observation
Ohio University's yearlong Copilot pilot found steady savings for only part of its 130 users.
Takeaway
License the communication-heavy and administrative roles that gained, and leave the rest on Copilot Chat.
What happened
Ohio University's IT office ran the Copilot add-on licence for a year with about 130 participants to judge its return. Copilot helped with quick drafting in Outlook, Word and PowerPoint, and administrative and research support roles gained more than instructional or technical roles. Only a portion of users reported consistent, measurable savings, typically 1 to 3 hours a week.
“Approximately 130 individuals across OHIO participated in the pilot program and represented a wide range of users across various planning units. Participants agreed to be active in a Microsoft Teams AI channel, provide regular feedback and evaluations, adhere to University data security policies, document use cases and experiences, and collaborate with OHIO IT to contribute to the insights and understanding of Microsoft Copilot. Monthly meetings were held with participants to share updates and information and solicit feedback from Copilot users. The pilot concluded on May 30, 2025.”
Ohio University Office of Information Technology
Public organization write-up (university news report)
Time savings self-reported
Deployment details
Organization-wide. About 130 staff across planning units. Yearlong pilot ended May 30, 2025, reported August 28, 2025.
Public organization write-up (university news report)
Time savings self-reported
Team
About 130 staff
Duration
One year, ended May 30, 2025
Workload
Drafting in Outlook, Word and PowerPoint
Product version
Microsoft 365 Copilot add-on licence
Measurement
Self-reported hours saved per week (1 to 3 for some users)
FraunhoferFeb 20, 2026arXivA German research organization licensed 550 staff and found administrators valued Copilot first, scientists later.
Carsten F. SchmidtEvaluation authorFraunhofer
Organization-wide. Non-university research organization. 550 licences from May 2024, paper published February 20, 2026.
arXiv, published Feb 20, 2026Read the report (opens arXiv in a new tab)
Observation
A German research organization licensed 550 staff and found administrators valued Copilot first, scientists later.
Takeaway
Plan role-specific training: administrators reported higher usefulness and reliability, while scientists became more positive over time.
What happened
Fraunhofer equipped 550 employees with Microsoft 365 Copilot licences in May 2024 and evaluated the rollout with repeated surveys. Administrative staff rated usefulness and reliability higher, scientific staff grew more positive over time on productivity and workload, and Copilot was rated easy to use, with the most value on clearly structured text tasks.
“The study analyzes the introduction of Microsoft 365 Copilot in a non-university research organization using a repeated cross-sectional employee survey. We assess usefulness, ease of use, output quality and reliability, and usefulness for typical knowledge-work activities. Administrative staff report higher usefulness and reliability, whereas scientific staff develop more positive assessments over time, especially regarding productivity and workload reduction. Copilot is widely viewed as user-friendly and technically reliable, with greatest added value for clearly structured, text- based tasks. The findings highlight learning and routinization effects when embedding generative AI into work processes and stress the need for context-sensitive implementation, role-specific training and governance to foster sustainable acceptance of generative AI in knowledge-intensive organizations.”
Carsten F. Schmidt, Sophie Petzolt, Wolfgang Beinhauer, Ingo Weber, Stefan Langer
Public organization write-up (research paper by the organization's staff)
Repeated employee survey
Deployment details
Organization-wide. Non-university research organization. 550 licences from May 2024, paper published February 20, 2026.
Public organization write-up (research paper by the organization's staff)
Repeated employee survey
Team
550 licensed employees
Duration
From May 2024, two survey waves
Workload
Structured text tasks, administration and research work
Product version
Microsoft 365 Copilot
Measurement
Usefulness, ease of use, output quality and reliability ratings
Virginia TechApr 7, 2025Virginia TechVirginia Tech piloted Copilot with 177 faculty and staff for 38 workdays.
AI at Virginia TechPilot ownersVirginia Tech
Organization-wide. 177 volunteer faculty and staff, January to March 2025. Outcome report dated April 7, 2025.
Its own announcement, published Apr 7, 2025Read the report (opens the announcement in a new tab)
Observation
Virginia Tech piloted Copilot with 177 faculty and staff for 38 workdays.
Takeaway
Short pilots produce high self-reported savings; set the renewal decision on use measured after the pilot ends.
What happened
Virginia Tech gave 177 volunteer faculty and staff Microsoft 365 Copilot for 38 workdays, January to March 2025, and published its outcome report on April 7, 2025. 94% of the 85 who answered the closing survey said Copilot saved them time, an average of 38 minutes a day by their own estimate.
Public organization write-up (university pilot report)
Self-reported time saved
Deployment details
Organization-wide. 177 volunteer faculty and staff, January to March 2025. Outcome report dated April 7, 2025.
Public organization write-up (university pilot report)
Self-reported time saved
Team
177 volunteer faculty and staff; 85 survey responses
Duration
38 workdays, January to March 2025
Product version
Microsoft 365 Copilot
Measurement
Share reporting time saved (94%) and self-reported minutes per day (38)
Te Kawa Mataaho Public Service CommissionOct 9, 2026Te Kawa Mataaho Public Service CommissionNew Zealand's business ministry trialled Copilot on ministerial correspondence and reported better quality and speed.
Te Kawa Mataaho Public Service CommissionOfficial Information Act response, reporting the Ministry of Business, Innovation and Employment's Copilot trialTe Kawa Mataaho Public Service Commission
Organization-wide. New Zealand government ministry. Structured trial reported in an OIA response dated October 9, 2026.
Its own announcement, published Oct 9, 2026Read the report (opens the announcement in a new tab)
Observation
New Zealand's business ministry trialled Copilot on ministerial correspondence and reported better quality and speed.
Takeaway
Pick one written workflow with a clear owner, such as ministerial replies, and trial Copilot on that before buying for everyone.
What happened
Answering an Official Information Act request on October 9, 2026, the Public Service Commission reported that MBIE had completed a structured trial of Microsoft Copilot for drafting, and that it improved the quality and efficiency of the ministerial correspondence process. The response gives no staff count or timings.
“MBIE has completed a structured trial to evaluate the use of Microsoft Copilot for drafting…”
Te Kawa Mataaho Public Service Commission
Public organization write-up (government OIA response)
No timings published
Deployment details
Organization-wide. New Zealand government ministry. Structured trial reported in an OIA response dated October 9, 2026.
Public organization write-up (government OIA response)
No timings published
Workload
Drafting ministerial correspondence
Product version
Microsoft Copilot

Adopted it3

Black & VeatchSep 8, 2026Microsoft Customer StoriesBlack & Veatch brought nearly all of its Copilot users on board with volunteer groups and tiered training.
Mike AdamsChief Digital Technology OfficerBlack & Veatch
Organization-wide. Engineering, construction, and consulting firm. About 13,000 professionals, more than 6,000 using Copilot.
Microsoft Customer Stories, published Sep 8, 2026Read the write-up (opens Microsoft Customer Stories in a new tab)
Observation
Black & Veatch brought nearly all of its Copilot users on board with volunteer groups and tiered training.
Takeaway
Recruit volunteers into fixed-length groups, move graduates on to agent building, and answer the data-safety question in the first session.
What happened
Black & Veatch rolled Copilot out in three tiers. Adopter groups of 500 to 850 volunteers take a six-week course, live and asynchronous. About 3,500 graduates move to a navigator group with monthly training on Copilot Studio and the Researcher and Analyst agents. A smaller builder group trains on Copilot Studio, GitHub Copilot, and Microsoft Foundry. A Viva Engage community of more than 8,000 people posts weekly tips, and an AI Governance Committee sets the guardrails. The first question from every new group is whether its data is safe. The company reports 98% adoption and up to 9 hours a week saved per user.
Vendor customer story
Adoption and time savings reported by Black & Veatch
Deployment details
Organization-wide. Engineering, construction, and consulting firm. About 13,000 professionals, more than 6,000 using Copilot.
Vendor customer story
Adoption and time savings reported by Black & Veatch
Team
Nearly half the firm's professionals; adopter groups of 500 to 850
Workload
Daily Microsoft 365 work, Copilot Studio agents, and BV Ask, a Foundry agent over thousands of process documents
Product version
Microsoft 365 Copilot, Copilot Cowork, Copilot Studio, and the Researcher and Analyst agents
Measurement
Adoption rate (98%) and time saved per user (up to 9 hours a week)
CapitaJun 24, 2025Microsoft Customer StoriesAt Capita, 3,000 Copilot licenses saved 9,000 hours in one month, and employees built 169 agents of their own.
Tiina StephensDirector of DigitalCapita
Organization-wide. UK outsourcing and professional services company. 3,000 licenses after a trial across business areas.
Microsoft Customer Stories, published Jun 24, 2025Read the write-up (opens Microsoft Customer Stories in a new tab)
Observation
At Capita, 3,000 Copilot licenses saved 9,000 hours in one month, and employees built 169 agents of their own.
Takeaway
Name a champion in every department before opening agent building, and start with a search agent over the policies people already hunt for.
What happened
Capita started with a small trial across business areas and grew it to 3,000 licenses as the business value showed. It picked Copilot for security, since most of its data and processes already sit in Microsoft 365, and checked the rollout against the EU AI Act with Microsoft's legal teams. Employees saved 9,000 hours in one month; drafting and Teams transcription are the examples Capita gives. Champions in each department taught colleagues to build no-code agents, and 169 employee-built agents now run. The most used, AskMeAnything, answers policy and procedure questions from Capita's SharePoint estate.
Vendor customer story
Hours saved and agent count reported by Capita
Deployment details
Organization-wide. UK outsourcing and professional services company. 3,000 licenses after a trial across business areas.
Vendor customer story
Hours saved and agent count reported by Capita
Team
3,000 licensed employees across the business
Workload
Drafting, Teams meeting transcription, and no-code agents such as AskMeAnything over SharePoint policies
Product version
Microsoft 365 Copilot and Copilot agents
Measurement
Hours saved in one month (9,000) and employee-built agents (169)
EYNov 18, 2025Microsoft Customer StoriesEY deployed Microsoft 365 Copilot to more than 150,000 employees before it moved on to agents.
Anthony CaterinoGlobal Managing Partner, Business EnablementEY
Organization-wide. Global professional services firm. More than 150,000 employees, reported November 18, 2025.
Microsoft Customer Stories, published Nov 18, 2025Read the write-up (opens Microsoft Customer Stories in a new tab)
Observation
EY deployed Microsoft 365 Copilot to more than 150,000 employees before it moved on to agents.
Takeaway
Run the hands-on sessions first and open agent building after people use the assistant daily.
What happened
EY deployed Microsoft 365 Copilot to more than 150,000 employees and paired the rollout with change management and hands-on learning sessions before expanding into Copilot Studio agents. Microsoft's story gives no time-saved or usage figures for the rollout.
“EY’s transformation started with Microsoft 365 Copilot, deployed to more than 150,000 employees and expanding rapidly worldwide.”
Anthony Caterino
Vendor customer story
No usage or time figures published
Deployment details
Organization-wide. Global professional services firm. More than 150,000 employees, reported November 18, 2025.
Vendor customer story
No usage or time figures published
Team
More than 150,000 employees
Workload
Everyday Microsoft 365 work, then Copilot Studio agents
Product version
Microsoft 365 Copilot and Copilot Studio

Reported by others1

Antony SpenceJul 6, 2026iTnewsAustralia's Attorney-General's Department moved staff to Copilot Chat in its own protected tenancy.
Antony SpenceAssistant Secretary for IT support, engagement and innovationAttorney-General's Department (Australia)
Organization-wide. Australian federal department. Copilot Chat and NotebookLM in controlled tenancies, reported July 6, 2026.
iTnews, published Jul 6, 2026Read the report (opens iTnews in a new tab)
Observation
Australia's Attorney-General's Department moved staff to Copilot Chat in its own protected tenancy.
Takeaway
Certify the tenancy for the data staff actually handle; an assistant limited to public data goes unused.
What happened
Spence told Google Cloud Summit Sydney that the department stood up Microsoft Copilot Chat and Google NotebookLM over the past year in its own controlled tenancies, certified at the protected level, after an earlier phase limited to public data fell out of favour with staff. Staff can now use AI with their own data, and people still own every final version.
“These two products are certified at the protected level for us. They're in our own controlled [cloud] tenancies, and what it now allows staff to do is actually use AI with their data.”
Antony Spence
Second-hand news report (iTnews, from a conference talk)
Reported by iTnews
Deployment details
Organization-wide. Australian federal department. Copilot Chat and NotebookLM in controlled tenancies, reported July 6, 2026.
Second-hand news report (iTnews, from a conference talk)
Reported by iTnews
Team
Department staff
Duration
Over the year to July 2026
Workload
Work on departmental data up to the protected level
Product version
Microsoft Copilot Chat beside Google NotebookLM

Failures, incidents and rollbacks2

Australia's Treasury found Copilot easy to deploy, and the trial failed to meet staff's expectations.

A Copilot Chat bug let confidential email past a DLP policy in Tony Redmond's own tenant until Microsoft fixed it in February 2026.

Run it beside another tool2

The UK Ministry of Justice runs targeted rollouts of Copilot and ChatGPT Enterprise in parallel.

Yale's AI tools catalog lists Microsoft 365 Copilot and Claude Enterprise side by side, each with its own price and data classification.

Evaluated it11

HMRC randomly allocated 3,000 Copilot licences, measured about an hour a week saved, then scaled past 28,000 licences.

The Department for Business and Trade found high satisfaction with Copilot, and its evaluators found no evidence that the time saved improved productivity.

NHS England is rolling Copilot out to 505,000 staff on the strength of a pilot whose method was never published.

DWP compared licensed staff with non-users and estimated 19 minutes a day saved, led by search and email.

At the North Carolina Department of Transportation, staff rated Copilot less useful after an eight-week pilot than before it.

Across 20,000 UK civil servants, users reported saving 26 minutes a day, a figure GDS built from self-reported bands.

The University of Washington piloted Copilot for five months, and the pilot recommended against funding it centrally.

Ohio University's yearlong Copilot pilot found steady savings for only part of its 130 users.

A German research organization licensed 550 staff and found administrators valued Copilot first, scientists later.

Virginia Tech piloted Copilot with 177 faculty and staff for 38 workdays.

New Zealand's business ministry trialled Copilot on ministerial correspondence and reported better quality and speed.

Adopted it3

Black & Veatch brought nearly all of its Copilot users on board with volunteer groups and tiered training.

At Capita, 3,000 Copilot licenses saved 9,000 hours in one month, and employees built 169 agents of their own.

EY deployed Microsoft 365 Copilot to more than 150,000 employees before it moved on to agents.

Reported by others1

Australia's Attorney-General's Department moved staff to Copilot Chat in its own protected tenancy.