Past the stall: Why M365 Copilot rollouts plateau and the governed path through

The licenses arrived; the deployment never did

The pattern repeats across industries. An enterprise agreement gets amended, a few hundred or a few thousand Microsoft 365 Copilot seats arrive, the launch email goes out, and a webinar fills up. Ninety days later the usage report tells a different story: weekly active use sits far below paid seats, and the steering committee starts to ask what the licenses actually bought. We have sat in those reviews. The stall is common enough to be the default outcome.

The published data matches what we see in the field. A Gartner survey of 132 IT leaders, conducted in June 2024 and reported by Computerworld, found that 60 percent of organizations had started Copilot pilots. Only 6 percent had finished a pilot and were actively planning large-scale deployment. One percent had completed a rollout to all eligible office workers. Gartner analyst Max Goss summarized the gap in that report: Microsoft generated interest at remarkable speed, but the pilots were not translating into broader deployments.

The same survey explains why. Seventy-two percent of IT leaders said employees struggle to integrate Copilot into their daily routines. Fifty-seven percent reported that user engagement declines quickly after implementation. Only 3 percent said Copilot was providing significant value at the time of the survey. None of these are license problems. They are integration problems, and they surface after procurement has already closed (Exhibit 1).

The plateau has a price tag that CFOs can read without help. Forrester's cost model uses Copilot's list price of 30 dollars per user per month. Apply that to a typical stalled estate: a 3,000-seat purchase running at 25 percent weekly active use carries roughly 2,250 idle seats. That is 67,500 dollars a month, around 810,000 dollars a year, paid for software nobody opens. The arithmetic is ours, but every input is public, and we have presented versions of this table to more than one steering committee.

Demand is not the constraint. Microsoft's 2025 Work Trend Index reports that 80 percent of the global workforce lacks the time or energy to do their job, and 53 percent of leaders say productivity must increase. The same report finds that workers at the most AI-mature organizations are far more likely to describe their company as thriving, 71 percent against 37 percent globally. The appetite for capacity is real on both sides of the org chart. What is missing is the connective work between a licensed model and a specific team's Tuesday morning.

Copilot stalls are not demand problems. They are integration debt, and the debt comes due at the 90-day usage review.

Copilot pays off where the work is structured, and stalls where it is not

The strongest public evidence on Copilot value comes from the UK government, which ran one of the largest evaluated deployments published anywhere: 20,000 employees across government departments and agencies in the final quarter of 2024. The cross-government findings report, published on GOV.UK, measured an average of 26 minutes saved per user per day, roughly 13 working days per year, and 82 percent of participants said they did not want to return to working without the tool.

Read past the headline and the distribution is the real finding. Savings concentrated where the work is structured. Drafting documents saved 24 minutes. Building presentations saved 19. Scheduling and routine communication saved 9. Adoption followed the same gradient: use in Teams peaked at 71 percent of participants, while peak adoption inside Excel and PowerPoint reached only 23 to 24 percent. The report also found a correlation between a user's prior familiarity with AI and the time they saved. Skill, not seat count, predicted value.

The Department for Work and Pensions published a deeper six-month evaluation of its 3,549-user trial. An econometric analysis estimated 19 minutes saved per day across eight routine task types, with information retrieval at 26 minutes and email drafting at 25. Ninety percent of users said Copilot helped them save time, and users reported reinvesting the recovered time in work they considered more valuable, such as project delivery and strategic planning.

The UK trials also answer Gartner's engagement-decay finding. Where Gartner's respondents saw usage fall off quickly after launch, the supported and instrumented government cohorts held. The cross-government experiment reached 83 percent peak adoption and stayed near 80 percent for the duration. In the DWP trial, 65 percent of participants used Copilot daily and another 30 percent weekly, and job satisfaction rose 0.56 points on a seven-point scale. The difference between decay and retention was not the product. It was the scaffolding around it: named cohorts, defined use cases, and measurement that participants could see.

Our delivery experience says the gradient is the lesson. Teams whose work product is documents and recurring meetings gain fast. Teams with judgment-heavy, weakly structured work gain slowly or not at all, and both UK reports flagged the same limitation around nuance and contested source material. A rollout plan that treats all 5,000 seats identically is therefore wrong by design. The deployment unit is the team and its workflow, not the tenant.

Oversharing, not model quality, is what stops rollouts at the security review

Copilot answers from everything a user can technically reach. That single design fact turns years of permissive SharePoint sharing into a day-one risk: the salary file shared to 'Everyone except external users' in 2019 becomes retrievable in a chat window. Security teams understand this immediately, which is why governance, not enthusiasm, sets the rollout calendar.

The Gartner survey reported by Computerworld quantifies the delay. Sixty-four percent of IT leaders said information governance and security risks consumed significant time and resources. Forty percent delayed their rollout by three months or more specifically over data oversharing. Fifty-seven percent restricted Copilot to low-risk or trusted user groups, which caps adoption by policy before culture even gets a vote.

Forrester's Total Economic Impact study of Microsoft 365 Copilot, published in March 2025, corroborates this from the success side: 70 percent of surveyed organizations ran data security projects ahead of deployment. Microsoft now says the quiet part in its own documentation. Its deployment blueprint for a secure and governed Copilot foundation is organized into three pillars, in this order: remediate oversharing, set up guardrails, meet regulations. The recommended tooling is Microsoft Purview plus SharePoint Advanced Management, and the latter is already included with the Copilot license. The vendor's own guidance puts governance ahead of scale.

We treat governance as scoping work with a definition of done, not a standing committee. The error we see most is the tenant-wide program: a six-month effort to classify everything before anyone gets value. The workable alternative is cohort-scoped. Inventory and remediate the sites and libraries the pilot teams actually use, apply guardrails there, then expand the governed perimeter sprint by sprint. We build systems to SOC 2, HIPAA, and GDPR standards, and we align every implementation with the EU AI Act, so the cohort-scoped approach still leaves an audit trail a regulator can read.

Microsoft's own deployment blueprint puts oversharing remediation before scale. A rollout plan that skips it is not faster. It is just earlier in its delay.

Training costs more than licenses, and the generic version does not transfer

Forrester's TEI model is widely quoted for its headline: 116 percent ROI and 19.7 million dollars of net present value over three years, with payback in 10 months. The model rests on a 25,000-employee composite organization built from interviews with 16 decision-makers and a survey of 367 Copilot users. The number buyers should study sits on the cost side. Forrester prices training and discovery at 6.9 million dollars over three years, against 5.8 million dollars for the licenses themselves. The model assumes 10 hours of formal training per user plus 6 hours of ongoing discovery per user per year. The modeled returns are conditional on that investment. Most stalled rollouts we encounter spent the license line and skipped the larger one.

The DWP evaluation shows what happens when training is funded but stays generic. Despite formal training on offer, 89 percent of users relied on self-directed exploration and 61 percent learned from peers. Their request to the department was specific: shorter sessions, built for their role. Gartner's survey found the same pressure from the management side, with 73 percent of organizations reporting higher-than-expected change management demands.

The practitioner translation: skills are per-team because prompts are per-workflow. A financial close team and a field service dispatcher need entirely different prompt sets, grounded in their own documents and rhythms. The ten prompts that matter to a team can be identified in days by watching the work, and they transfer only when taught inside the live workflow with the team's real content. A tenant-wide webinar teaches the product. It does not teach the job.

The license is the smallest line in an honest Copilot budget. Forrester prices training and discovery above licensing. Fund the gap or keep the plateau.

Adoption has a deployable shape: two to three sprints per workflow cluster

We are an AI implementation firm. We have shipped production systems since 2014, more than 200 of them, and the Copilot stall looks to us like every other adoption problem we get hired to fix: the technology arrived before the implementation. The fix has a shape, and the shape is small.

The evidence that workflow-level work moves usage comes from our own deployments. For Rockwell Automation we built an AI-driven digital workplace serving more than 28,000 employees in over 80 countries. Content findability improved 60 percent and support tickets fell 40 percent. Licenses did not produce those numbers. Mapping how specific teams search for and reuse content did, followed by measurement against agreed baselines.

For Copilot, the shape is two to three two-week sprints per workflow cluster (Exhibit 2). Each sprint starts with executable acceptance criteria signed on day one, runs inside the client environment, and costs 10,000 dollars paid only after every criterion passes. Larger programs repeat the same unit on the same terms.

Acceptance criteria do the heavy lifting. They are binary, they are testable, and they are agreed before any build starts. For a Copilot engagement they look like this:

• Every SharePoint site used by the pilot cohort passes an oversharing scan, with exceptions documented and access remediated.
• Each pilot team has a prompt library covering its highest-volume tasks, validated in writing by the team lead.
• Weekly active use exceeds the agreed threshold of pilot seats at day 30, read directly from the Microsoft 365 Copilot usage report.
• A named owner inside IT can rerun the oversharing scan and the usage report without us.

The gating matters more than the speed. The UK government could publish credible numbers because it instrumented the trial before it started. Most enterprises cannot say what their Copilot seats returned last quarter because nobody defined what a returned dollar would look like. Acceptance criteria force that definition to exist on day one, and payment tied to criteria keeps the incentive honest. That is the difference between a rollout that plateaus quietly and one that earns its next tranche of seats.

What to do with this on Monday morning

1. 1. Pull the Microsoft 365 Copilot usage report for the last 90 days, divide weekly active users by paid seats, and put that one number at the top of your next leadership agenda.
2. 2. Pick a pilot cohort of two or three teams whose work is document-heavy and repeatable, and suspend tenant-wide ambitions until this cohort shows measured adoption.
3. 3. Run an oversharing scan on the cohort's SharePoint sites this month using SharePoint Advanced Management, and remediate before you expand access, not after an incident.
4. 4. Rebudget to Forrester's ratio: match every licensing dollar with at least a dollar for per-team training and workflow integration.
5. 5. Replace the launch webinar with per-team prompt libraries taught inside live workflows, and measure usage per team, not per tenant.
6. 6. Write executable acceptance criteria for the next phase, including a weekly-active-use threshold, and gate the next tranche of seats on passing them.

1. Forrester, The Total Economic Impact of Microsoft 365 Copilot (2025)
2. Microsoft, 2025 Work Trend Index Annual Report: The Frontier Firm is born (2025)
3. Computerworld, Microsoft 365 Copilot rollouts slowed by data security, ROI concerns (2024)
4. GOV.UK, Microsoft 365 Copilot Experiment: Cross-Government Findings Report (2025)
5. GOV.UK, An Evaluation of DWP's Microsoft 365 Copilot Trial (2025)
6. Microsoft Learn, Secure and Governed Data Foundation for Microsoft 365 Copilot: Foundational Deployment Guidance (2026)