The City of Raisio in Finland began a Microsoft 365 Copilot rollout in autumn 2025 with Sogeti, part of Capgemini, training nearly 100 municipal employees through a staged program designed to support the city’s 2026 strategy for better data use. The notable part is not that another public-sector employer bought AI licenses. It is that Raisio treated Copilot less like a software deployment and more like an organizational literacy test. In a year when AI is being pushed into office work by default, Raisio’s case argues that the safest path to productivity may be the slowest-looking one.
Most enterprise AI stories are written in the future tense. Vendors promise that knowledge workers will spend less time summarizing, searching, drafting, triaging, and repeating themselves. Buyers repeat the language of acceleration because nobody wants to admit that the expensive new assistant may sit unused, mistrusted, or dangerously misunderstood.
Raisio’s rollout is interesting because it begins somewhere more mundane: with municipal employees who needed permission to learn. The city had already bought Microsoft 365 Copilot licenses, but the project was not framed as a race to turn on a feature. It was framed as a learning journey, with the human organization treated as the system that had to be upgraded first.
That matters in local government, where the phrase “knowledge work” hides a very wide range of jobs. A city is not a sales floor, a software team, or a consulting practice. It is education, infrastructure, HR, communications, administration, resident services, procurement, and policy execution, all sharing overlapping records and obligations.
The mayor’s comparison to the Industrial Revolution may sound grandiose, but the anxiety is familiar. AI has the aura of inevitability, and inevitably invites resistance. Raisio’s bet was that resistance is not best overcome by executive enthusiasm or vendor demos, but by giving employees a way to encounter the tool without embarrassment, coercion, or magical thinking.
Raisio’s stated goals were telling. The city wanted a safe and supportive learning environment, a cultural shift toward shared AI practice, and a better understanding of employees’ future development needs. None of those is a pure IT objective. They are change-management objectives, which is precisely why they are often underfunded until a rollout disappoints.
The city’s CIO, Eero Rostiala, described the employer’s task as enabling people to succeed. That is not the usual language of software procurement, and it should not be dismissed as HR varnish. If Copilot is positioned as a way to remove drudgery and improve well-being, then the deployment has to prove that employees are being supported rather than measured against a new machine-enabled pace.
The participant quote supplied by Capgemini is almost too tidy, but it captures the adoption problem better than most executive sound bites. An employee who “started from zero” and later described Copilot as an everyday assistant is the constituency every public-sector AI program must reach. The early adopters will experiment anyway; the institutional payoff comes when cautious workers learn enough to use the tool without surrendering judgment to it.
That is why Raisio’s emphasis on privacy, security, and usage guidelines is not just boilerplate. Microsoft’s enterprise pitch for Copilot rests heavily on the claim that existing Microsoft 365 permissions, labels, retention settings, and administrative controls continue to matter. But those controls are only as good as the data hygiene around them and the users’ understanding of the boundary between assistance and authority.
This is the uncomfortable truth inside every Copilot deployment: AI does not only automate work; it reveals how work was already organized. If SharePoint permissions are sloppy, Copilot can make that sloppiness more visible. If teams have no shared conventions for storing, naming, classifying, or retaining information, the assistant inherits that mess.
Raisio’s people-first language can sound soft until viewed through that lens. Training employees before pushing them into daily use is not a sentimental choice. It is risk management. The most dangerous Copilot user is not the skeptic who refuses to touch it; it is the confident user who has never been taught when to distrust it.
That ordering matters because many AI programs invert it. They begin with transformation talk, then scatter generic training, then wonder why departments fail to adopt the technology evenly. Raisio’s model instead accepts that a city’s different functions need different examples, different boundaries, and different definitions of success.
The introductory training created a shared baseline. That is especially important for generative AI, where employees may arrive with wildly different assumptions shaped by consumer chatbots, headlines, personal experimentation, or fear. A common starting point reduces the chance that one department treats Copilot as a search engine, another as a ghostwriter, and a third as a compliance risk to be avoided entirely.
The role-specific workshops were the more important layer. AI adoption becomes real only when workers can map it onto tasks they already recognize. “Copilot can summarize information” is an abstraction; “Copilot can help prepare this kind of internal briefing from these kinds of source materials, under these rules” is a work practice.
The final automation component points to where Microsoft and its partners want this market to go. Copilot begins as an assistant inside documents, email, chat, and meetings, but the strategic prize is workflow redesign. Raisio appears to have treated that as a future horizon rather than a day-one demand, which is probably why the project reads as credible rather than breathless.
The missing middle is the layer between Microsoft’s platform claims and the employee’s desk. Someone has to translate tenant readiness, governance, prompt practice, departmental workflows, training materials, and leadership messaging into a coherent rollout. That is where firms like Sogeti can make themselves valuable, especially for organizations that do not have a large internal AI adoption office waiting in the wings.
Raisio’s CIO said the city chose Sogeti because the partner began with needs and goals before proposing the way of working. That is exactly what buyers should demand, because generative AI has been plagued by solution-first selling. If the opening move is a demo rather than a diagnosis, the customer is already being nudged into the vendor’s preferred story.
This does not mean every municipality needs a major consultancy to introduce Copilot. Smaller organizations can and should build internal champions, reuse official training resources, and start with narrowly defined scenarios. But the Raisio case shows why “enablement” is not a decorative add-on. It is the mechanism by which a license becomes a changed habit.
That integration is also why Copilot is not a neutral chatbot bolted onto the side. It operates inside an organization’s existing information architecture. The better that architecture, the more useful the assistant can be; the worse it is, the more likely Copilot is to surface stale, overexposed, badly labeled, or context-poor material.
Raisio’s strategy elevated data as a daily asset, not merely an executive reporting input. That is the important strategic move. If employees experience data only as something demanded by management, AI becomes another extraction tool. If employees experience data as something that helps them find, summarize, communicate, and decide, the assistant has a much better chance of becoming useful.
The risk is that organizations hear “Copilot respects your permissions” and mistake that for “Copilot makes your information estate safe.” It does not. It can respect bad permissions just as faithfully as good ones. It can summarize what a user can access, even if the user should not have had that access in the first place.
This is where public-sector organizations should be especially careful. The adoption conversation cannot be separated from data classification, lifecycle management, records obligations, and internal access reviews. A friendly training environment is necessary, but it should sit beside a hard-nosed governance program.
That modesty makes the case more persuasive. The first wave of practical Copilot value is not likely to be autonomous bureaucracy. It is more likely to be a reduction in the friction that makes administrative work feel heavier than it should: finding the relevant file, turning notes into a coherent message, extracting the gist from a long thread, drafting a first version of a routine document.
Those small savings are not trivial in local government. Municipal employees often face rising service expectations without a matching expansion of staff. If AI can reduce the cognitive tax of routine composition and retrieval, the benefit may appear less as a dramatic headcount story and more as an incremental improvement in attention.
But the word “if” deserves to remain in the sentence. Productivity gains from AI are uneven because work is uneven. A communications specialist, HR administrator, teacher, engineer, and city executive will not all get the same value from the same assistant in the same week. Raisio’s role-specific approach acknowledges that reality instead of flattening it into a single adoption metric.
Hesitation is often misread as resistance to change. Sometimes it is. But with generative AI, hesitation can also be a rational response to uncertainty. Employees may worry about privacy, accuracy, plagiarism, accountability, job security, or simply looking foolish in front of colleagues who seem more technically fluent.
A good rollout lowers the social cost of learning. Raisio’s program appears to have done that by allowing employees to move at their own pace, revisit materials, and learn from peers. Peer learning is particularly important because AI practices spread through examples more than policy documents.
This is the part of adoption Microsoft cannot automate. Copilot can help draft the policy, summarize the training notes, and propose use cases. It cannot create trust among employees who suspect the tool is being imposed on them. That trust has to be built locally, by managers and colleagues who understand the work.
It is better understood as a different form of ambition. The city did not merely want a pilot group of enthusiasts to produce impressive anecdotes. It wanted to begin moving AI into the shared operating culture of the organization. That is slower, harder, and more consequential.
The project’s scale — nearly 100 employees directly participating — is meaningful but not massive. The wider impact came from open materials and the exchange of experiences across the organization. That is how adoption escapes the pilot trap: the first cohort becomes a source of local examples rather than a sealed demonstration unit.
There is also a democratic quality to the approach that suits the public sector. If AI is going to reshape everyday work, access to learning should not be limited to the already confident. Raisio’s emphasis on a supportive environment suggests an understanding that digital transformation can otherwise widen internal divides between power users and everyone else.
Microsoft’s enterprise assurances are central to why Copilot is even in the conversation for governments and regulated organizations. Prompts and responses in commercial Microsoft 365 contexts are covered by enterprise data protection commitments, and Microsoft says customer data accessed through Microsoft Graph is not used to train foundation models. Those commitments lower a major barrier, but they do not eliminate the need for local governance.
Recent reporting on Copilot-related bugs and exploit chains has also made clear that “enterprise-grade” is not a synonym for invulnerable. AI systems introduce new surfaces for prompt injection, oversharing, tool misuse, and unexpected data retrieval. Even when vendors patch specific vulnerabilities, the broader lesson remains: assistants that can search, summarize, and act across organizational data deserve serious administrative attention.
That is why Raisio’s co-created AI usage guidelines are more than a training artifact. They are a signal that the city understood adoption as a behavioral and governance problem. Users need practical rules for what data to enter, how to verify outputs, when to disclose AI assistance, and when not to use the tool at all.
The next maturity step for organizations like Raisio will be measurement. Not surveillance for its own sake, but evidence of which scenarios work, where errors occur, what support employees need, and whether governance rules are being understood. Without that loop, early enthusiasm can decay into scattered habits.
This is especially important when AI is used to summarize large amounts of information. Summaries feel efficient because they compress attention, but compression is also where omissions become policy risks. A summary that misses a caveat, deadline, exception, or minority concern can be more dangerous than no summary at all, because it gives the user confidence.
The same applies to drafting. Better first drafts are valuable, but a fluent draft is not the same as an accurate or appropriate one. Municipal communication has tone, legal, accessibility, and trust obligations that cannot be delegated to probabilistic text generation.
Raisio’s reported focus on limitations is therefore critical. AI literacy is not prompt wizardry. It is knowing when the machine’s confidence is cheap, when verification is mandatory, and when the task should remain human from the start.
Raisio offers a more sustainable template. Start from strategy, not novelty. Tie AI to employee development, not only executive efficiency. Train broadly enough to create shared norms, then specifically enough to make the tool relevant. Keep privacy and security in the same conversation as productivity.
This template also helps Microsoft avoid the worst version of its own platform advantage. Because Copilot is embedded in tools employees already use, it can feel inevitable. But inevitability breeds resentment if workers experience AI as something done to them rather than with them.
The city’s language of shared learning may seem quaint beside the scale of Microsoft’s AI ambitions. Yet it is precisely the kind of language that makes enterprise AI survivable. Organizations do not become AI-enabled because a button appears in Word or Outlook. They become AI-enabled when people develop a common sense of how, when, and why to use it.
Raisio Makes the Unfashionable AI Bet: Training Before Transformation
Most enterprise AI stories are written in the future tense. Vendors promise that knowledge workers will spend less time summarizing, searching, drafting, triaging, and repeating themselves. Buyers repeat the language of acceleration because nobody wants to admit that the expensive new assistant may sit unused, mistrusted, or dangerously misunderstood.Raisio’s rollout is interesting because it begins somewhere more mundane: with municipal employees who needed permission to learn. The city had already bought Microsoft 365 Copilot licenses, but the project was not framed as a race to turn on a feature. It was framed as a learning journey, with the human organization treated as the system that had to be upgraded first.
That matters in local government, where the phrase “knowledge work” hides a very wide range of jobs. A city is not a sales floor, a software team, or a consulting practice. It is education, infrastructure, HR, communications, administration, resident services, procurement, and policy execution, all sharing overlapping records and obligations.
The mayor’s comparison to the Industrial Revolution may sound grandiose, but the anxiety is familiar. AI has the aura of inevitability, and inevitably invites resistance. Raisio’s bet was that resistance is not best overcome by executive enthusiasm or vendor demos, but by giving employees a way to encounter the tool without embarrassment, coercion, or magical thinking.
The Real Product Was Confidence, Not Copilot
Microsoft 365 Copilot is often sold as an interface to the information already living inside Microsoft 365. In practical terms, that means it can help users draft documents, summarize meetings and messages, search across organizational content, and turn natural-language prompts into office-work outputs. But that same promise depends on a prerequisite that is easy to skip: workers must understand what the assistant can and cannot responsibly do.Raisio’s stated goals were telling. The city wanted a safe and supportive learning environment, a cultural shift toward shared AI practice, and a better understanding of employees’ future development needs. None of those is a pure IT objective. They are change-management objectives, which is precisely why they are often underfunded until a rollout disappoints.
The city’s CIO, Eero Rostiala, described the employer’s task as enabling people to succeed. That is not the usual language of software procurement, and it should not be dismissed as HR varnish. If Copilot is positioned as a way to remove drudgery and improve well-being, then the deployment has to prove that employees are being supported rather than measured against a new machine-enabled pace.
The participant quote supplied by Capgemini is almost too tidy, but it captures the adoption problem better than most executive sound bites. An employee who “started from zero” and later described Copilot as an everyday assistant is the constituency every public-sector AI program must reach. The early adopters will experiment anyway; the institutional payoff comes when cautious workers learn enough to use the tool without surrendering judgment to it.
A City Is a Bad Place for AI Theater
Municipal government is a particularly unforgiving environment for AI theater because the work touches citizens who cannot simply opt out. A poorly summarized internal memo is one thing. A mistaken interpretation of service information, personal data, procurement rules, or resident correspondence is another.That is why Raisio’s emphasis on privacy, security, and usage guidelines is not just boilerplate. Microsoft’s enterprise pitch for Copilot rests heavily on the claim that existing Microsoft 365 permissions, labels, retention settings, and administrative controls continue to matter. But those controls are only as good as the data hygiene around them and the users’ understanding of the boundary between assistance and authority.
This is the uncomfortable truth inside every Copilot deployment: AI does not only automate work; it reveals how work was already organized. If SharePoint permissions are sloppy, Copilot can make that sloppiness more visible. If teams have no shared conventions for storing, naming, classifying, or retaining information, the assistant inherits that mess.
Raisio’s people-first language can sound soft until viewed through that lens. Training employees before pushing them into daily use is not a sentimental choice. It is risk management. The most dangerous Copilot user is not the skeptic who refuses to touch it; it is the confident user who has never been taught when to distrust it.
The Three-Stage Model Is Boring in Exactly the Right Way
The learning path Raisio and Sogeti built had three pieces: introductory training for the whole organization, role-specific workshops, and a look at automation opportunities. That sequence is almost aggressively sensible. It begins with common vocabulary, moves into job-specific practice, and only then gestures toward broader process redesign.That ordering matters because many AI programs invert it. They begin with transformation talk, then scatter generic training, then wonder why departments fail to adopt the technology evenly. Raisio’s model instead accepts that a city’s different functions need different examples, different boundaries, and different definitions of success.
The introductory training created a shared baseline. That is especially important for generative AI, where employees may arrive with wildly different assumptions shaped by consumer chatbots, headlines, personal experimentation, or fear. A common starting point reduces the chance that one department treats Copilot as a search engine, another as a ghostwriter, and a third as a compliance risk to be avoided entirely.
The role-specific workshops were the more important layer. AI adoption becomes real only when workers can map it onto tasks they already recognize. “Copilot can summarize information” is an abstraction; “Copilot can help prepare this kind of internal briefing from these kinds of source materials, under these rules” is a work practice.
The final automation component points to where Microsoft and its partners want this market to go. Copilot begins as an assistant inside documents, email, chat, and meetings, but the strategic prize is workflow redesign. Raisio appears to have treated that as a future horizon rather than a day-one demand, which is probably why the project reads as credible rather than breathless.
Capgemini and Sogeti Sell the Missing Middle
Capgemini’s client story is, naturally, a client story. It is designed to make the partner look good, the customer look enlightened, and the product look useful. Still, beneath the case-study polish is a recognizable market reality: the hardest part of Microsoft 365 Copilot is no longer buying the license.The missing middle is the layer between Microsoft’s platform claims and the employee’s desk. Someone has to translate tenant readiness, governance, prompt practice, departmental workflows, training materials, and leadership messaging into a coherent rollout. That is where firms like Sogeti can make themselves valuable, especially for organizations that do not have a large internal AI adoption office waiting in the wings.
Raisio’s CIO said the city chose Sogeti because the partner began with needs and goals before proposing the way of working. That is exactly what buyers should demand, because generative AI has been plagued by solution-first selling. If the opening move is a demo rather than a diagnosis, the customer is already being nudged into the vendor’s preferred story.
This does not mean every municipality needs a major consultancy to introduce Copilot. Smaller organizations can and should build internal champions, reuse official training resources, and start with narrowly defined scenarios. But the Raisio case shows why “enablement” is not a decorative add-on. It is the mechanism by which a license becomes a changed habit.
Copilot’s Promise Depends on the Boring Stuff Microsoft Cannot Do for You
Microsoft has spent the past few years repositioning Copilot from a novelty into a productivity layer for work. The pitch is powerful because Microsoft 365 is already where many organizations live. Email, documents, Teams conversations, calendars, SharePoint sites, OneDrive files, and meeting artifacts form the substrate of modern office labor.That integration is also why Copilot is not a neutral chatbot bolted onto the side. It operates inside an organization’s existing information architecture. The better that architecture, the more useful the assistant can be; the worse it is, the more likely Copilot is to surface stale, overexposed, badly labeled, or context-poor material.
Raisio’s strategy elevated data as a daily asset, not merely an executive reporting input. That is the important strategic move. If employees experience data only as something demanded by management, AI becomes another extraction tool. If employees experience data as something that helps them find, summarize, communicate, and decide, the assistant has a much better chance of becoming useful.
The risk is that organizations hear “Copilot respects your permissions” and mistake that for “Copilot makes your information estate safe.” It does not. It can respect bad permissions just as faithfully as good ones. It can summarize what a user can access, even if the user should not have had that access in the first place.
This is where public-sector organizations should be especially careful. The adoption conversation cannot be separated from data classification, lifecycle management, records obligations, and internal access reviews. A friendly training environment is necessary, but it should sit beside a hard-nosed governance program.
The Productivity Story Is Plausible Because It Is Modest
The benefits Raisio reported are refreshingly ordinary: easier document production, better-quality drafts, clearer summaries of large information sets, more efficient information searches, and improved email and communication management. There is no claim that the city reinvented public administration in a quarter. There is no suggestion that AI replaced professional judgment.That modesty makes the case more persuasive. The first wave of practical Copilot value is not likely to be autonomous bureaucracy. It is more likely to be a reduction in the friction that makes administrative work feel heavier than it should: finding the relevant file, turning notes into a coherent message, extracting the gist from a long thread, drafting a first version of a routine document.
Those small savings are not trivial in local government. Municipal employees often face rising service expectations without a matching expansion of staff. If AI can reduce the cognitive tax of routine composition and retrieval, the benefit may appear less as a dramatic headcount story and more as an incremental improvement in attention.
But the word “if” deserves to remain in the sentence. Productivity gains from AI are uneven because work is uneven. A communications specialist, HR administrator, teacher, engineer, and city executive will not all get the same value from the same assistant in the same week. Raisio’s role-specific approach acknowledges that reality instead of flattening it into a single adoption metric.
The Survey Results Point to the Real Barrier: Hesitation
Capgemini says pre- and post-project surveys showed that training significantly reduced hesitation and helped employees understand both the potential and limitations of AI. That is exactly the kind of outcome AI programs should measure early. Before an organization can calculate return on investment, it has to know whether workers are willing and able to use the tool appropriately.Hesitation is often misread as resistance to change. Sometimes it is. But with generative AI, hesitation can also be a rational response to uncertainty. Employees may worry about privacy, accuracy, plagiarism, accountability, job security, or simply looking foolish in front of colleagues who seem more technically fluent.
A good rollout lowers the social cost of learning. Raisio’s program appears to have done that by allowing employees to move at their own pace, revisit materials, and learn from peers. Peer learning is particularly important because AI practices spread through examples more than policy documents.
This is the part of adoption Microsoft cannot automate. Copilot can help draft the policy, summarize the training notes, and propose use cases. It cannot create trust among employees who suspect the tool is being imposed on them. That trust has to be built locally, by managers and colleagues who understand the work.
“Safe and Gradual” Is Not the Opposite of Ambitious
The current AI market rewards speed in public. Every vendor wants to say it is leading, every customer wants to say it is innovating, and every board wants evidence that the organization is not being left behind. Against that background, Raisio’s gradual rollout could look cautious to the point of dullness.It is better understood as a different form of ambition. The city did not merely want a pilot group of enthusiasts to produce impressive anecdotes. It wanted to begin moving AI into the shared operating culture of the organization. That is slower, harder, and more consequential.
The project’s scale — nearly 100 employees directly participating — is meaningful but not massive. The wider impact came from open materials and the exchange of experiences across the organization. That is how adoption escapes the pilot trap: the first cohort becomes a source of local examples rather than a sealed demonstration unit.
There is also a democratic quality to the approach that suits the public sector. If AI is going to reshape everyday work, access to learning should not be limited to the already confident. Raisio’s emphasis on a supportive environment suggests an understanding that digital transformation can otherwise widen internal divides between power users and everyone else.
The Security Conversation Has Moved From “Can We Use It?” to “Can We Govern It?”
In 2023 and 2024, many organizations were still asking whether generative AI could be allowed near enterprise data at all. By 2026, the question has become more operational: under what conditions, with which controls, for which users, and with what training? Raisio’s project belongs to that second phase.Microsoft’s enterprise assurances are central to why Copilot is even in the conversation for governments and regulated organizations. Prompts and responses in commercial Microsoft 365 contexts are covered by enterprise data protection commitments, and Microsoft says customer data accessed through Microsoft Graph is not used to train foundation models. Those commitments lower a major barrier, but they do not eliminate the need for local governance.
Recent reporting on Copilot-related bugs and exploit chains has also made clear that “enterprise-grade” is not a synonym for invulnerable. AI systems introduce new surfaces for prompt injection, oversharing, tool misuse, and unexpected data retrieval. Even when vendors patch specific vulnerabilities, the broader lesson remains: assistants that can search, summarize, and act across organizational data deserve serious administrative attention.
That is why Raisio’s co-created AI usage guidelines are more than a training artifact. They are a signal that the city understood adoption as a behavioral and governance problem. Users need practical rules for what data to enter, how to verify outputs, when to disclose AI assistance, and when not to use the tool at all.
The next maturity step for organizations like Raisio will be measurement. Not surveillance for its own sake, but evidence of which scenarios work, where errors occur, what support employees need, and whether governance rules are being understood. Without that loop, early enthusiasm can decay into scattered habits.
Public Administration Cannot Outsource Judgment to a Text Box
There is a subtle danger in the phrase “everyday assistant.” It is the right aspiration for Copilot, but it can invite the wrong mental model. Assistants can be helpful, but in public administration the accountable actor remains the employee and the institution, not the software.This is especially important when AI is used to summarize large amounts of information. Summaries feel efficient because they compress attention, but compression is also where omissions become policy risks. A summary that misses a caveat, deadline, exception, or minority concern can be more dangerous than no summary at all, because it gives the user confidence.
The same applies to drafting. Better first drafts are valuable, but a fluent draft is not the same as an accurate or appropriate one. Municipal communication has tone, legal, accessibility, and trust obligations that cannot be delegated to probabilistic text generation.
Raisio’s reported focus on limitations is therefore critical. AI literacy is not prompt wizardry. It is knowing when the machine’s confidence is cheap, when verification is mandatory, and when the task should remain human from the start.
The Raisio Lesson Microsoft Should Want Customers to Learn
Microsoft benefits when customers believe Copilot is easy to activate, but it may benefit more when customers understand that adoption is work. Failed rollouts are bad for the category. A city that buys licenses and then discovers low usage, user anxiety, or governance confusion becomes a cautionary tale.Raisio offers a more sustainable template. Start from strategy, not novelty. Tie AI to employee development, not only executive efficiency. Train broadly enough to create shared norms, then specifically enough to make the tool relevant. Keep privacy and security in the same conversation as productivity.
This template also helps Microsoft avoid the worst version of its own platform advantage. Because Copilot is embedded in tools employees already use, it can feel inevitable. But inevitability breeds resentment if workers experience AI as something done to them rather than with them.
The city’s language of shared learning may seem quaint beside the scale of Microsoft’s AI ambitions. Yet it is precisely the kind of language that makes enterprise AI survivable. Organizations do not become AI-enabled because a button appears in Word or Outlook. They become AI-enabled when people develop a common sense of how, when, and why to use it.
The Finnish City’s Copilot Rollout Leaves a Practical Trail
Raisio’s project should not be read as proof that every Copilot deployment will pay for itself, or that Microsoft’s assistant is the right answer for every public-sector workflow. It is better read as a concrete example of what serious adoption looks like before the dashboards and ROI slides arrive. The strongest lesson is that the rollout treated employees as learners and stewards, not simply as endpoints.- Raisio began its Copilot initiative in autumn 2025 to support a city strategy that makes data a practical asset in everyday work and decision-making from 2026 onward.
- The city had already purchased Microsoft 365 Copilot licenses, but delayed broad value claims until employees had training, guidance, and role-specific practice.
- Sogeti’s contribution was not only technical expertise, but a change-management structure that joined training, communications, materials, and AI usage guidelines.
- Nearly 100 employees participated directly, while shared materials and peer learning extended the program’s influence beyond the initial cohort.
- The most credible reported benefits were everyday productivity improvements in drafting, summarizing, searching, and managing communications, rather than sweeping automation.
- The project’s most transferable lesson is that AI governance and AI confidence have to be built together, especially in organizations handling public information and resident services.
References
- Primary source: Capgemini
Published: Tue, 30 Jun 2026 05:25:33 GMT
Loading…
www.capgemini.com - Official source: learn.microsoft.com
Enterprise data protection in Microsoft 365 Copilot and Microsoft 365 Copilot Chat | Microsoft Learn
Learn what enterprise data protection means for Microsoft 365 Copilot and Microsoft 365 Copilot Chat.learn.microsoft.com - Official source: developer.microsoft.com
Microsoft 365 Copilot | Extend and Customize Copilot
Extend, enrich, and customize Microsoft Microsoft 365 Copilot. Explore Copilot extensibility options such as agents, API plugins, and Copilot connectors to expand AI-powered productivity, skills, and creativity.developer.microsoft.com - Official source: support.microsoft.com
Loading…
support.microsoft.com - Related coverage: techtarget.com
How does enterprise data protection in Copilot work? | TechTarget
Under enterprise data protection, Microsoft promises not to train Copilot on data that users submit in their queries. Learn more.www.techtarget.com - Official source: adoption.microsoft.com
Loading…
adoption.microsoft.com
- Official source: techcommunity.microsoft.com
Loading…
techcommunity.microsoft.com - Related coverage: windowscentral.com
Microsoft 365 Copilot Chat summarized confidential emails | Windows Central
Microsoft confirms a Copilot bug that exposed and summarized confidential emails, bypassing security policies.www.windowscentral.com - Related coverage: techradar.com
Microsoft 365 Copilot can be turned into a one-click data theft tool — inbox, OneDrive, and SharePoint data all at risk, so patch now | TechRadar
Varonis found a way to chain three bugs into one exploitwww.techradar.com - Official source: learn.microsoft.com.mcas.ms
Loading…
learn.microsoft.com.mcas.ms - Official source: microsoft.com
</rdf:Alt> </dc:title> <dc:description> <rdf:Alt> <rdf:li xml:lang="x-default"/> </rdf:Alt> </dc:description> <dc:creator> <rdf:Seq> <rdf:li>Lukas V
</rdf:Alt> </dc:description> <dc:creator> <rdf:Seq> <rdf:li>Lukas Velushwww.microsoft.com
- Official source: info.microsoft.com
Loading…
info.microsoft.com - Related coverage: tomshardware.com
Admins finally get the power to uninstall Microsoft Copilot on Windows 11 Pro, Enterprise, and EDU versions — devices must meet specific conditions to allow the removal of the AI app | Tom's Hardware
One less bloatware on Windows 11.www.tomshardware.com - Related coverage: tomsguide.com
Microsoft confirms Copilot bug let its AI read sensitive and confidential emails | Tom's Guide
Microsoft confirmed a bug in Copilot was letting the AI assistant read and summarize confidential emails.www.tomsguide.com - Official source: cdn-dynmedia-1.microsoft.com