The municipality of Raisio, Finland, has taken an unusual step in its generative AI journey: it trained employees first. In autumn 2025, the city partnered with Sogeti, part of Capgemini, to roll out a comprehensive adoption program for Microsoft 365 Copilot, ensuring its roughly 1,500 municipal workers understood the tool before they started using it. The move contrasts sharply with many organizations that have deployed Copilot with minimal preparation, only to confront confusion, security missteps, and wasted licenses.
At the core of the initiative is a recognition that generative AI in the public sector demands more than just a software license. City officials understood that Copilot’s ability to summarize documents, draft emails, and analyze data inside Word, Excel, Teams, and Outlook could transform administrative workflows, but only if employees trusted it and knew when—and when not—to rely on it. The program, designed with Sogeti’s change management and AI expertise, focused on practical, hands-on workshops, scenario-based learning, and clear governance guidelines.
Raisio’s approach is a case study in public sector AI adoption done right. While governments worldwide rush to embrace AI for efficiency gains, many have stumbled over data privacy concerns, employee skepticism, and a lack of internal expertise. Raisio flipped the typical sequence: instead of launching Copilot and then scrambling to fix misuse, the city built a foundation of literacy and responsibility. The training covered prompt engineering, recognizing AI hallucinations, protecting sensitive citizen data, and integrating Copilot into daily tasks without replacing human judgment.
The partnership with Sogeti brought change management to the forefront. Sogeti consultants worked with Raisio’s IT and HR departments to identify power users who could become internal champions, tailoring workshops to specific job functions. A finance officer learned how to automate budget report drafts, while a social worker discovered how to summarize case notes without exposing confidential client information. This role-based customization helped overcome the common one-size-fits-all training which often leaves employees disengaged.
Raisio’s timing is strategic. Microsoft has been aggressively pushing Copilot into enterprises and government bodies, with new features rolling out monthly. Yet, a 2024 Gartner survey found that fewer than 30% of organizations provided any formal training before introducing generative AI tools, leading to low adoption rates and security incidents. In Finland, where digital literacy is high but public trust in AI remains cautious, Raisio’s proactive stance sets a national precedent.
The program also addresses a critical weak spot: the human risk factor. Copilot can access vast amounts of organizational data, and without proper training, an employee might inadvertently surface confidential citizen records in a prompt or rely on an AI-generated response that contains subtle factual errors. Raisio’s training emphasized a “human-in-the-loop” principle: every AI output must be verified, and sensitive data must never be entered freely. The city even simulated real-world scenarios where Copilot gave plausible but incorrect answers, teaching staff to spot inaccuracies.
Finland’s municipal structure makes Raisio’s experiment particularly interesting. Cities in Finland enjoy considerable autonomy, with responsibility for education, healthcare, and social services. Copilot’s potential to streamline the heavy administrative burden across these sectors is enormous, but so are the risks. A misstep in handling health data or student records could violate the EU’s General Data Protection Regulation (GDPR) and Finland’s strict national privacy laws. Raisio’s legal and compliance teams were therefore integrated from day one, helping design guardrails rather than being consulted after deployment.
The financial calculus behind the training-first model is compelling. Copilot licenses cost approximately $30 per user per month, a non-trivial sum for a small city. Ensuring each employee can extract real value means the investment pays off not just in productivity but in avoided costs—fewer help desk tickets, reduced risk of breaches, and faster uptake. Early anecdotes from the program suggest that trained employees adopted Copilot at double the rate of untrained peers in a pilot phase, though the city has not yet released full data.
Sogeti’s role extends beyond training delivery. As part of Capgemini’s global AI practice, it brings a framework that combines technical readiness with organizational psychology. Their methodology includes measuring baseline AI awareness, conducting empathy mapping to understand employee fears, and co-creating a responsible AI charter. For Raisio, this meant addressing genuine concerns among staff: would Copilot replace their jobs? Would it report their keystrokes? The transparency built during training helped defuse these anxieties, reframing Copilot as a copilot, not an autopilot.
This human-centered design is what sets the project apart from typical IT rollouts. Rather than a top-down mandate, Raisio opted for a voluntary early adopter phase, then scaled based on internal demand generated by positive word-of-mouth. Departments that saw colleagues drafting council meeting minutes in half the time became curious rather than resistant. The city’s communications unit even used Copilot to help draft public updates, though always with a human editor reviewing the final text—a practice now embedded in policy.
Broader implications for the Finnish public sector loom large. If Raisio succeeds, other municipalities may follow suit, and the central government could weave Copilot training into its national digital skills framework. Finland’s Ministry of Finance has already urged public agencies to explore AI, but without specific guidance on adoption methodology. Raisio’s blueprint could fill that gap, providing a tested model for training, governance, and rollout sequencing.
From a Windows workplace perspective, the integration of Copilot into the Microsoft 365 suite means that workers across devices—Windows 11 PCs, tablets, and smartphones—are affected. Raisio’s IT team had to ensure that endpoint security policies on Windows were updated to reflect new Copilot traffic patterns and data flows. That practical layer of the training included clear instructions on using Copilot only through approved browsers and applications, avoiding shadow IT scenarios.
The timing also coincides with a broader Microsoft push around “AI Trust, Privacy and Security”. At Ignite 2024, the company announced new controls for Copilot, including sensitivity labels and audit logging, but these features only work if users comprehend their purpose. Raisio’s training cleverly linked technical controls to behavioral norms, making security features understandable rather than obscure settings buried in admin panels.
Critics might argue that a training-first approach is too slow for a rapidly evolving technology. By the time employees are trained, the tool may have changed. Raisio acknowledged this by making the program iterative. Instead of a one-and-done course, the city set up quarterly refreshers and a dedicated Copilot helpdesk. A dedicated Teams channel allowed employees to share tips and flag unexpected behaviors, creating a living knowledge base that evolves with the product.
This agile learning model reflects a truth about generative AI: its capabilities are morphing so quickly that static training is obsolete the moment it’s delivered. Copilot’s integration with plugins, its expanding context windows, and its ability to connect to CRM and ERP systems mean that what employees learned in autumn 2025 will need updating by spring 2026. Raisio’s partnership with Sogeti includes ongoing advisory hours to keep pace.
Another distinctive element is the ethical dimension. Public sector workers serve citizens, and there is a moral hazard in allowing AI to make decisions that affect benefits eligibility, social services, or school placements. Raisio’s charter explicitly forbids using Copilot for decision-making about individual citizens, limiting it to drafting, summarization, and analysis support. The training drove this point home with case studies from other countries where AI had been misused in government contexts.
Employee feedback gathered during the program indicates that the most valued sessions were those dealing with prompt crafting. Workers were amazed at how different the output could be with a well-structured prompt. They learned to instruct Copilot to “adopt a neutral tone”, “extract key dates only”, or “redact names before summarizing”. These prompt engineering skills turned Copilot from a toy into a tool, and participants reported a sense of empowerment rather than threat.
For Microsoft, Raisio represents an ideal customer story. The company has been emphasizing customer success over raw license sales, knowing that high churn could damage its AI ambitions. Public sector deal cycles are long, and a reference case from a pragmatic European city can sway skeptical buyers. Microsoft’s Finnish subsidiary likely supported the effort behind the scenes, perhaps providing technical resources or fast-track access to preview features.
Yet, the ultimate test will be measured in real-world outcomes: faster permit processing, shorter response times to citizen inquiries, more accurate meeting documentation, and higher employee satisfaction. Quantitative metrics are being collected by Raisio’s innovation office, with a public progress report expected in early 2026. If the numbers align with the enthusiasm, the city may have discovered a replicable formula for AI adoption that many larger organizations have missed.
The Raisio model could also influence Microsoft’s own deployment guidance. Currently, the company offers FastTrack onboarding, but many partners complain it is too technical and light on change management. Sogeti’s structured approach might become a template for future partnerships, blending technology rollout with deep organizational immersion.
In conclusion, the City of Raisio’s deliberate, employee-first strategy for Copilot is a benchmark for public sector AI. By prioritizing literacy, safety, and trust over speed, it sidesteps the chaos that often accompanies generative AI pilots. The coming months will reveal whether training truly correlates with sustained adoption, but the early signals suggest that slowing down to go fast is, in this case, exactly the right pace. Other municipalities and even private companies would do well to watch this Finnish experiment unfold.