The AI ​​Dilemma: Why Swiss companies know what to do – and yet hesitate

The Swiss IT Study 2026 by Computerworld puts it unequivocally: Artificial intelligence (AI) has definitively arrived in Swiss boardrooms and IT departments as an omnipresent topic, the first tangible application, and a strategic urgency. Today, generative AI makes it possible to generate simple applications and code syntax in a very short time using clever prompting. However, a huge gap exists between this technological democratization and consistent, value-creating implementation. This gap is less technical and more organizational, structural, and cultural in nature. This article examines what this changed situation means for companies that use custom software as a business-critical core tool.

Between experiment and everyday life (and the temptation of shadow IT)

Almost no Swiss decision-maker doubts today that AI will have a fundamental and disruptive impact on their business model. The latest IT study clearly demonstrates this. And yet, a look behind the scenes reveals a completely different picture: The vast majority of organizations remain in pure experimentation mode. Initial, isolated pilot projects are underway, simple chatbots have been set up, and initial successes in everyday office life have been celebrated. However, consistent, strategically embedded, and scalable use of AI in core operational systems remains the absolute exception.

This is no coincidence, but rather the result of a perfectly rational and justified reluctance in the Swiss market. What's often forgotten when quickly and seemingly easily creating a single, self-directed app are the massive risks beyond simply generating the code. Building simple apps independently can work well as long as the complexity remains manageable, no sensitive personal data is processed, no complex third-party interfaces are required, and no professional IT infrastructure is needed.

However, as soon as business-critical data is involved, open questions surrounding the revised Swiss Data Protection Act (nDSG), inadequate data governance, and the cybersecurity risks of uncontrolled applications stifle the initiative of management. Anyone operating in Switzerland with sensitive customer, financial, or business data has excellent reason to meticulously examine, from an architectural perspective, where, by whom, and under what conditions AI processes this data, and whether digital sovereignty is maintained.

Efficiency beats innovation – and the danger of “prompt-and-pray”

The Swiss IT study clearly shows where Swiss companies are already using AI productively and what benefits they are deriving from it. The direction is remarkably pragmatic: increasing efficiency and reducing costs are unchallenged at the top of the objectives. Reducing repetitive tasks, automating processes, and optimizing the use of internal resources are the primary drivers. In contrast, genuine, radical innovation through AI—that is, the creation of entirely new digital products, disruptive business models, or exclusive differentiating features—still plays a significant role.

This stance is understandable and makes sound business sense, as efficiency gains are quickly measurable and calculable, while the risk remains manageable. True innovation, however, requires a far higher risk tolerance, deeper technological data maturity, and strict governance.

For companies that work with custom-built software, this presents a crucial opportunity: AI doesn't realize its true ROI as an isolated, external tool or as a self-generated, amateurish construct (shadow IT). It delivers the greatest added value when seamlessly and securely integrated into existing business processes and core applications. Those who operate their own specialized applications have the unique opportunity to integrate AI functions directly, via secure connections, where the operational impact is greatest, without risky detours through generic platforms.

The scaling gap and the infrastructure problem

The central problem described in the IT study is therefore not the lack of AI technology itself. Many companies have their isolated flagship projects: individual departments where AI tools provide targeted support. But very few have yet managed to make the leap from these volatile, isolated solutions to company-wide, strategically managed, and cyber-secure use.

What's missing in practice is rarely the algorithm itself. What's lacking are stable data architectures, unambiguous governance structures, and the organizational willingness to fundamentally rethink established ways of working. AI is not an optional feature that can simply be tacked onto a shaky legacy system or an amateurish app. It fundamentally changes processes and the demands placed on IT infrastructure.

This is precisely where the economic dividing line lies, redefining the classic CHF 50,000 threshold: While AI reduces the pure "syntax-writing costs" of code, aspects such as a professional IT infrastructure, data protection, scalability, and fault-resistant interfaces to third-party systems become all the more important. The first step to sustainable AI success is therefore never simply choosing a trendy tool, but rather a thorough and uncompromising analysis of one's own system landscape: What data is available, what is its quality, how accessible are the APIs, and which processes are architecturally viable?

What this means for your software strategy

Anyone planning new, custom software projects or modernizing existing enterprise systems today can no longer ignore the question of AI. This doesn't mean that every application needs to be immediately overloaded with complex neural networks. However, it does mean that today's architectural decisions will determine your company's AI capabilities and future viability.

Specifically, a future-proof software strategy requires:

• Cleanly structured data storage: Data must be modeled from the ground up in a way that is compliant with the nDSG (National Data Protection Act), consistent, and centrally accessible.
• Open and extensible interfaces: The consistent avoidance of rigid closed-box solutions in favor of flexible, protected APIs.
• Precise process modeling: Processes must be thought through so precisely from a human perspective that automation potential can be clearly isolated, instead of relying on “prompt-and-pray” code.

The AI ​​dilemma is ultimately a classic decision dilemma for CEOs. Those who wait until all global regulations are clarified inevitably lose their competitive edge. Conversely, those who invest hastily in AI tools without a strategy, professional infrastructure, or a solid data foundation, or who amateurishly cobble together systems, squander their budget on applications that will never scale and pose massive security risks. The only sensible approach is a highly structured, phased build-out: hardening the data foundation, consistently designing the software architecture to be "AI-ready," and piloting initial, clearly defined use cases with clear business benefits.