The City of Raisio in southwest Finland has embarked on a landmark initiative to integrate generative AI into municipal operations, launching a Microsoft 365 Copilot training program for nearly 100 employees in autumn 2025. The effort, executed in partnership with Sogeti—part of the Capgemini group—aims to equip staff with the skills to leverage AI safely, while establishing rigorous governance around permissions and data access. For a city of roughly 25,000 residents, the move signals a bold step toward digital transformation in the public sector, setting a precedent that other Finnish and Nordic municipalities are likely to watch closely.

Raisio’s decision comes as public administrations across Europe grapple with how to harness AI’s potential without compromising citizen data or trust. By choosing Microsoft 365 Copilot, an AI assistant deeply integrated into the Office suite, the city bets that productivity gains can be achieved within a familiar toolset, reducing the learning curve. Yet the real innovation lies not in the software itself, but in the structured rollout that prioritizes workforce readiness and ethical deployment.

The Raisio Initiative: From Planning to Pilot

The program did not emerge overnight. According to city officials, an extended planning phase preceded the autumn 2025 launch, during which Raisio assessed its existing IT infrastructure, data classification maturity, and staff digital literacy. Working with Sogeti’s change management specialists, the municipality designed a phased adoption roadmap that would first introduce Copilot to a core group of approximately 100 employees across various departments—including administration, social services, and urban planning.

This pilot group was selected to represent a cross-section of roles, ensuring that the training outcomes would reflect genuine workplace scenarios. Employees ranged from clerical staff who handle routine correspondence to managers overseeing complex citizen-facing processes. By starting with a manageable cohort, Raisio intends to gather feedback, refine permissions models, and build a cadre of internal champions before expanding access city-wide.

Microsoft 365 Copilot leverages large language models grounded in the organization’s own data via the Microsoft Graph. When a user asks a question in Word, Excel, or Teams, Copilot pulls context from their emails, documents, calendars, and chats—but only what the user is explicitly permitted to see. This architectural approach makes permissions management the linchpin of any deployment. Raisio recognized early that without ironclad access controls, the risk of sensitive information leaking across municipal silos could derail public confidence.

Why Microsoft 365 Copilot? Familiarity Meets Advanced AI

Unlike standalone AI chatbots, Microsoft 365 Copilot is woven into the applications that millions of information workers already use daily. For Raisio’s staff, this meant that adoption did not require learning an entirely new interface. Instead, they could summon AI assistance within the familiar ribbon of Word, ask natural-language questions in Excel to generate formulas, or draft meeting summaries in Teams with a single click.

The city cited several anticipated use cases: automatically generating summaries of lengthy policy documents, drafting responses to citizen inquiries with consistent tone and terminology, analyzing budget spreadsheets for anomalies, and creating presentations for council meetings. Copilot’s ability to synthesize information from multiple sources—provided those sources are accessible to the user—promises to slash the time spent on routine knowledge work, freeing employees for higher-value interactions with residents.

However, these benefits are contingent on proper configuration. Copilot respects existing Microsoft 365 permissions, but many organizations discover that their SharePoint sites, Teams folders, and OneDrive files have accumulated overly broad access rights over the years. Raisio conducted a thorough permissions audit as part of the preparation, tightening sharing links and removing stale access grants. This hygiene work not only secures Copilot but also improves overall data governance, a secondary payoff that the city’s IT leadership valued independently.

The Training and Change Management Approach

Sogeti’s role extended far beyond technical implementation. The consultancy provided a comprehensive change management framework that acknowledged the human dimensions of AI adoption. Employees were not simply expected to experiment with Copilot on their own; they received structured workshops that blended hands-on exercises with discussions about ethical use, limitations, and bias awareness.

The training curriculum covered:

  • Prompts and Precision: How to craft effective prompts that yield accurate, relevant outputs. Staff practiced transforming vague requests into specific instructions that reference named documents or datasets.
  • Verification Habits: Because generative AI can produce plausible but incorrect information, employees learned to treat Copilot’s output as a first draft rather than a final product. Verification against source materials became a mandatory step.
  • Data Sensitivity Awareness: Employees were taught to identify scenarios where even permitted data might be inappropriate to feed into an AI model, such as when handling highly sensitive personal data without additional anonymization.
  • Permission Boundaries: Practical exercises showed how Copilot operates strictly within the user’s own access scope, reinforcing the importance of not sharing confidential documents unnecessarily.

A dedicated “AI ambassador” program paired less tech-savvy staff with early adopters who could provide peer support. This buddy system, modeled on successful digital transformation projects in the private sector, proved effective in lowering anxiety. Initial feedback indicated that employees appreciated having a go-to colleague rather than relying solely on IT help desks.

Crucially, the city did not mandate Copilot usage across the board. Instead, it encouraged voluntary experimentation, with department heads identifying specific pain points where AI assistance could make a tangible difference. This bottom-up demand generation helped overcome skepticism, as employees saw firsthand how Copilot eliminated repetitive tasks they genuinely disliked.

Focus on Permissions and Data Governance

Public sector entities operate under strict data protection regulations, including the EU’s General Data Protection Regulation (GDPR). Municipalities handle a wide array of sensitive information: social welfare records, health-related data, tax details, and confidential council deliberations. Any AI system that ingests such data must meet the highest standards of security and compliance.

Raisio’s permissions-centric approach aligned with Microsoft’s documentation for Copilot, which emphasizes that the AI does not gain omnipotent access. Instead, it inherits the user’s existing range of sight. If a clerk cannot view a particular SharePoint folder, Copilot cannot surface its contents in that clerk’s responses. This inheritance model is designed to prevent accidental leaks, but it places a heavy burden on IT teams to ensure that permissions are correctly assigned and regularly reviewed.

To manage this complexity, Raisio implemented an automated permissions review tool that flags unusually broad access rights and alerts data owners to review them. The city also enforced a “least privilege” principle more rigorously than before, recognizing that Copilot’s efficiency gains would be meaningless if they came at the cost of unauthorized data exposure.

The municipality published an internal policy document outlining acceptable Copilot use cases and prohibited scenarios. For example, employees were instructed not to use Copilot to draft responses that contain legally binding decisions without full human review, nor to process large-scale personal data analyses without engaging the data protection officer. These guardrails were designed not to stifle innovation but to provide clarity, reducing the fear that an inadvertent action could cause a compliance breach.

Building Trust in Public Sector AI: A Delicate Balancing Act

Trust is the currency of public administration. Residents need to believe that their government handles personal data with care, and employees need confidence that new tools won’t jeopardize their jobs or professional standing. Raisio’s program addressed both dimensions.

For the public, the city communicated transparently about the pilot through its website and local news outlets. Key messages emphasized that Copilot would assist, not replace, human decision-making, and that all AI-generated content would undergo review. By framing the initiative as a productivity enhancer rather than a workforce reducer, the city preempted fears of automation-driven layoffs.

Internally, Raisio’s leadership acknowledged that some employees might view Copilot with suspicion—fearing it could render their skills obsolete or expose their work to unwanted scrutiny. The training program directly addressed these concerns, showing how Copilot handles mundane tasks like formatting and summarization, allowing staff to focus on judgment-intensive work. This reframing proved crucial in gaining buy-in, especially among long-tenured employees who had witnessed previous IT projects fizzle under poor change management.

Union representatives were consulted early in the planning process, ensuring that the rollout respected existing labor agreements. No positions were earmarked for elimination; instead, the city emphasized that any time savings would be redirected toward improving service quality, reducing backlogs, and enabling employees to engage in more meaningful community-facing work.

A Model for Finnish Municipalities—and Beyond?

Finland has long been a digital frontrunner, with strong e-government infrastructure and high public trust in technology. However, concrete examples of municipal AI adoption remain relatively rare. Raisio’s pilot, therefore, carries significance beyond its city limits. The Association of Finnish Local and Regional Authorities has expressed interest in monitoring the results, and several neighboring municipalities have already reached out for informal advice.

The partnership with Sogeti/Capgemini also suggests a replicable model. Rather than building bespoke AI solutions from scratch, smaller cities can leverage the existing Microsoft 365 ecosystem with expert guidance on governance and training. If Raisio’s experience demonstrates measurable productivity gains without security incidents, it could accelerate the uptake of Copilot across Finland’s public sector, potentially influencing national guidelines on public sector AI.

At the European level, the Raisio case aligns with broader policy directions. The EU AI Act classifies certain public sector uses as high-risk, demanding robust oversight. By proactively establishing a governance framework before full deployment, Raisio positions itself to comply with emerging regulations while reaping early-mover advantages. Other European municipalities, particularly in the Nordic and Baltic regions where digital maturity is high, are watching the pilot with interest.

Challenges Ahead

The road to broad AI adoption in public administration is far from smooth. Several challenges loom on Raisio’s horizon:

  • Sustaining Engagement: After the initial excitement, maintaining momentum can be difficult. Without continuous reinforcement, employees may slip back into old workflows, and the productivity gains may plateau. The city plans to counter this with regular refresher workshops and a library of use-case examples that evolve over time.
  • Evolving Permissions Complexity: As employees change roles or departments, permission sets must be updated promptly. Even a small oversight could lead to Copilot exposing data that should no longer be accessible. Raisio is investigating automated lifecycle workflows powered by Microsoft 365’s identity governance features to address this.
  • AI Quality and Hallucinations: Generative AI’s occasional tendency to fabricate information remains a concern. While training mitigates this, the city is exploring ways to integrate additional verification layers, such as requiring AI-generated summaries to cite source paragraphs explicitly.
  • Cost-Benefit Uncertainty: Microsoft 365 Copilot requires an additional per-user license fee, and the cost of Sogeti’s consultancy services adds to the initial investment. Raisio must demonstrate clear returns—whether through quantifiable time savings, improved service metrics, or employee satisfaction—to justify ongoing expenditure to taxpayers.
  • Inequity Across Departments: If some departments benefit more obviously than others, resentment could build. The city must ensure that Copilot’s value is communicated and, where possible, extended to roles traditionally less aligned with office-based knowledge work.

Conclusion: A Pragmatic Path to Public Sector AI

The City of Raisio’s Microsoft 365 Copilot pilot exemplifies a measured, human-centric approach to AI adoption that many larger organizations would do well to emulate. By placing training, permissions, and trust at the core of its strategy, the city mitigates the risks that often derail technology rollouts in the public sector. The partnership with Sogeti provides the expertise necessary to navigate both technical and cultural hurdles, while the pilot’s phased structure allows for course-correction before broader deployment.

As 2025 progresses, the eyes of Finnish local government will be on Raisio. If the promise of Copilot translates into tangible improvements in municipal efficiency and citizen service—without compromising security or trust—this small city could become the blueprint for a nationwide AI transformation. For now, the message is clear: generative AI in the public sector is not about flashy demos but about painstaking groundwork in governance, education, and cultural change.