In many organizations, AI is already in use—yet the value it creates remains limited. According to the MIT study “The GenAI Divide: State of AI in Business 2025,” 95 percent of generative AI pilots fail to deliver measurable impact on revenue or profitability. The reason is rarely the technology itself, but rather how companies organize AI, embed it into processes, and scale it across the business.
With Microsoft Copilot Wave 3, this challenge is moving further into focus. While Wave 2 primarily addressed individual productivity—positioning Copilot as a personal assistant in Teams, Outlook, or Word—Microsoft is now fundamentally expanding its approach: Copilot is becoming a platform for integrated agents and cross-process automation. At the same time, another development is becoming increasingly evident: the use of powerful AI capabilities and autonomous agents can no longer be viewed solely as a licensing topic. With usage-based pricing models for certain AI services, cost control, value tracking, and prioritization are moving to center stage. Organizations therefore need to decide not only where AI can be deployed from a technical perspective, but also where it delivers measurable value to enterprise-wide performance.
However, achieving this requires more than technology. The more AI is embedded in cross-functional processes, the more critical governance, data availability, integrated workflows, and clear ownership become. In many cases, the technology is evolving faster than the organizational structures required to scale it effectively.
From Pilot to Impact
Use cases emerge where there is a clear, tangible need—within individual business functions and close to their processes. Teams develop solutions that work well within their specific context. A common pattern: marketing teams automate content creation with Copilot, sales builds its own agent for proposal processes, and HR develops solutions for onboarding workflows. Each solution works effectively within its own domain and delivers visible value there.
However, as soon as end-to-end processes come into play, these isolated solutions begin to reach their limits. Systems are not aligned with one another, permission models do not integrate seamlessly, and processes are not designed holistically.
As a result, these approaches remain confined to individual functions. Learnings are not systematically transferred, and broader evolution—or integration into interconnected processes—rarely takes place.
As integration across chained processes increases, this tension turns into a structural challenge—one we see in many client engagements: a single solution works, but as soon as three functions share the same process, the chain breaks.
Understanding the Root Causes—and Their Implications
These root causes rarely appear in isolation; instead, they tend to reinforce each other.
1. Lack of Clear Governance
Without clear guardrails, parallel solutions begin to emerge. Teams develop their own use cases and agents without alignment or prioritization. At the same time, binding approval processes for production deployment are often missing. As a result, solutions remain stuck in pilot phases or are not scaled further due to uncertainty. Responsibilities remain unclear, and standards evolve inconsistently. In practice, this leads to similar solutions being developed multiple times—with varying quality and little to no reuse.
Microsoft addresses this challenge with Wave 3. Agent 365 introduces a centralized control layer and provides transparency into which agents are in use, what they are used for, and under which rules they operate.
The use of Cowork also requires clear guardrails, as it is capable of executing actions autonomously and across applications. The potential is significant. At the same time, however, the demands on governance increase. Organizations must not only ensure that agents operate in compliance with policies and that sensitive information is protected—they must also be able to assess which use cases actually deliver value. This question becomes even more important as usage-based pricing models continue to expand. Governance therefore extends beyond compliance and risk management to include the economic steering of AI: Which agents are being used? Where are costs incurred? And what business value do individual use cases create?
Flat, one-size-fits-all allocation of budgets or credits falls short in this context. Successful organizations deliberately prioritize where automation delivers measurable impact—such as reducing bottlenecks, lowering manual effort, or creating additional capacity. This is an important foundation. However, it does not replace the need for a clear organizational decision on who owns AI governance and how prioritization is actually managed.
How to successfully scale AI? Governance is the key.
In our AI Governance Check-Up, we analyze your current setup, identify risks, and outline the most impactful levers for your next steps.
Request a Workshop: AI Governance Check-Up
2. Data and Integration as Bottlenecks
Much of the relevant data is either unavailable or not usable. It is fragmented across different systems, incomplete, or accessible only to a limited extent. While AI agents can be implemented from a technical standpoint, they often operate with too limited a context. As a result, outputs remain superficial or require manual supplementation.
The bottleneck, therefore, is not the AI itself, but the underlying data foundation and its accessibility. This is where many organizations are currently rethinking their approach: integrating relevant data is increasingly seen as a critical prerequisite for enabling true end-to-end automation through agents. At the same time, many organizations are still in the early stages of this integration. Data is distributed across systems, permissions are inconsistently managed, and connections between datasets are difficult to establish.
With Wave 3, this challenge becomes more critical in two ways. First, context-based assistance can only deliver value if the underlying data is complete and accessible. Second, as the use of agents increases, questions arise around which data an agent is allowed to access and use. Without clear data classification and access controls, risks emerge that go beyond quality issues alone.
3. Architecture Without a Clear Target State
The effort required to integrate systems, data flows, and permissions afterward is significant—and increases with every use case built without a target state in mind. With Wave 3, cross-process workflows and integrated agents are moving further into focus. This requires consistent alignment across all layers. An architecture designed for isolated, standalone solutions will not be sufficient in the long term.
Building without a clear target state today creates integration overhead tomorrow—ultimately offsetting the efficiency gains that AI is meant to deliver.
4. AI as a Project Rather Than a Program
In many organizations, AI is still managed as a time-bound project. Individual use cases are identified, implemented, and then considered complete. While this often results in working solutions, they typically remain limited to their specific context.
At the same time, key initiatives run in parallel: HR trains employees, IT defines policies, and business units develop their own use cases. What’s missing is the connection between these efforts. There is no shared target state, no consistent prioritization, and no clear ownership for further development beyond individual use cases.
This is precisely where the challenge for scaling arises. AI does not create value in isolated solutions, but through the interplay of use cases, data, processes, and organization. When managed as a project, responsibility often ends with implementation. Systematic evolution, reuse, and scaling rarely take place.
A program-based approach, by contrast, pursues a different objective. AI is not treated as a one-off initiative, but as a capability continuously built within the organization. Use cases are not considered in isolation, but are developed, prioritized, and advanced in alignment with shared goals. Responsibilities are anchored over the long term, and progress is managed through clear metrics. This includes governance mechanisms such as OKRs, which link business objectives, AI initiatives, and measurable outcomes—creating transparency around the actual contribution of individual efforts.
What all four root causes have in common is the lack of this overarching approach in many organizations—one that brings together technology, organization, and processes. As AI becomes increasingly embedded in operational workflows, this evolves from an efficiency issue into a strategic vulnerability.
What Successful Companies Do Differently
These challenges cannot be addressed through technology alone. Successful companies therefore go beyond tools, transforming how AI is used and managed across the organization—on two levels.
At the Employee Level
With Wave 3, the way AI is used shifts fundamentally. Copilot no longer supports just individual tasks—it enables entire workflows. To leverage AI effectively, work itself must be structured differently: goals need to be clearly defined, processes articulated, and the relevant context made available.
In practice, one thing becomes clear: those who can describe their work in terms of target states, workflows, and constraints adapt much more quickly to this new way of working. Employees increasingly take on a steering role—similar to leading and coordinating a team, but in collaboration with AI agents. The quality of outcomes depends less on the technology itself and more on the ability to clearly define and structure processes. This capability does not emerge on its own. It requires guidance, targeted training formats, and leadership that actively supports new ways of working. This is where many organizations unlock an additional lever for improvement and innovation.
Enablement must evolve accordingly—from purely tool-based training toward formats where teams analyze their own workflows and experiment with AI in context. Organizations that embed this early build internal multipliers who can sustainably transfer knowledge across the business.
At the Organizational Level
At the same time, the structural prerequisites for scaling need to be established. The challenges outlined above show that isolated measures are not sufficient.
Governance and responsibilities should be built in parallel with broader AI adoption and continuously evolved. Many organizations consolidate this within a central function—often referred to as an AI Office or AI Center of Excellence. This function defines who prioritizes use cases, who makes decisions on agents, and who is responsible for ongoing development, typically within a target operating model.
In addition, an increasing number of organizations are adopting so-called agent factories as a standardized approach to scale agent development, provide reusable components, and support business units in implementation.
Data and system integrations should be built selectively, focusing on where automation actually delivers value. Not every integration is necessary, but the relevant data must be available and usable to enable end-to-end processes. Data thus becomes the central fuel for AI-driven workflows. Without sufficient context, outputs remain generic or are based on incomplete information. At the same time, data access alone is not enough. It is equally critical to connect relevant systems so that agents can not only process information but also trigger actions along defined processes. For example, an agent could automatically update CRM records or initiate follow-up tasks based on a transcribed customer conversation.
At the same time, enablement, use case development, and IT architecture must be considered together. When business units develop solutions, IT defines standards separately, and training takes place in isolation, the very silos that prevent scaling are created. What is needed, therefore, is an integrated approach with clear governance, prioritization, and continuous evolution.
Conclusion
Governance, data quality, and architecture determine whether scaling is effective. Microsoft provides the technology with Wave 3. But the more AI is embedded in operational workflows—and the more usage-based costs come into play—the more critical organizational maturity becomes. Organizations must make clearer decisions about where AI is deployed, the value it creates, and who is responsible for steering its development.
This is exactly where we at Campana & Schott support our clients: with a structured approach that brings together technology, enablement, and organization. If you want to understand what the next steps toward scaling AI should look like in your organization, get in touch with us.
The journey from early success to scalable impact begins with clarity on your current state and the right next-step priorities.
Authors
Categories
#AI Transformation
#AI Strategy
#AI Agents
#Microsoft Copilot
