Blog series 2 Automation & AI

Part 2 of the blog series “Automation in the age of AI”

2025 was a year in which many companies realized that the introduction of AI does not automatically lead to operational benefits. Models work, pilot projects deliver results, but there is no operational impact. The reason for this rarely lies in the quality of the AI itself, but in the execution layer: AI can generate recommendations and decisions, but it cannot control, secure or monitor processes. This puts automation at the heart of any serious AI transformation. Without scalable, transparent and controllable execution, AI remains an isolated analytics function – technically impressive, but business limited. 2025 has made it clear that companies are no longer just deciding on use cases, but on operating models. 2026 will be the year in which it becomes clear whether automation is seen as a tool – or as the foundation of an AI-enabled company.

AI creates a load that existing automation cannot bear

The introduction of AI increases cycle times, data volumes and decision-making frequency. However, this acceleration often affects processes that have grown incrementally over the years: manual handovers, time-controlled scripts, individual solutions between systems. A study by the Capgemini Research Institute from 2025 shows that only 28% of AI initiatives in large companies are scaled productively because the necessary process and automation basis is missing. Many organizations scale AI on an operational basis that is itself not scalable – and then experience bottlenecks, error cascades and shadow processes rather than greater efficiency.

Speed increases, control decreases: lack of observability and governance

As AI use increases, the relationship between execution and monitoring is shifting. Processes run faster, data streams become denser, human intervention decreases – but transparency, auditability and governance remain the same as before. According to ISACA Europe, 83% of companies were already using generative AI in 2025, but only 31% had a binding AI policy or control mechanisms in place. This means that operational speed is increasing, but controllability is decreasing. Without end-to-end observability, automation will not become more efficient, but more opaque – and therefore more risky, especially in regulated industries.

Individual tools instead of architecture: isolated solutions prevent scaling

Many companies have automation components, but no automation architecture. A bot here, a scheduler there, a script as a bridge between two systems – but no platform that consistently links events, data, decisions and execution. McKinsey points out in the “State of AI 2025” that the biggest barriers to AI scaling are not the technology, but the lack of integration and process control. Tools automate tasks, architectures orchestrate companies. As long as automation is treated as a local efficiency tool, technology will emerge without strategic impact.

Why this is not an AI problem, but an automation maturity problem

The visible braking factors are not in model training, but in the execution chain. The European Commission’s Digital Decade reports show that even in data- and cloud-rich companies, the maturity level of process automation lags behind AI adoption. If you want to make AI productive, you don’t just need more automation, you need better automation: controllable, observable, connectable. The question is no longer what is automated, but how automation is designed so that AI, data and applications no longer exist in separate operating logics.

What companies really need to decide in 2026

2026 will be characterized less by the question of which new AI use cases are possible, but rather which operational and architectural decisions can support their implementation. Companies are faced with a decision: will automation continue to be viewed as a technical project – or as a strategic control level for processes, systems and decisions? In future, the focus will be on event-driven orchestration, reliable observability, clear governance, platform consolidation and role models that connect IT, architecture and specialist departments. Those who understand automation as an infrastructure can scale AI. Those who see it as a tool will reach their limits.

Conclusion: What 2025 has shown – and what 2026 requires

2025 has made the operational gap visible: the biggest challenge is not the ability to develop AI, but to use it in a safe, scalable and controllable way. Automation will not disappear in the shadow of AI – it will become the foundation of any viable AI transformation. Companies that talk about AI maturity in 2026 need to talk about automation maturity. Only when execution, governance and transparency work together will technology create real business value.

If AI is making decisions faster and faster, but processes are not growing with it – how must automation be designed so that it not only processes tasks, but also supports corporate logic?

Sources

• Capgemini Research Institute – “Harnessing the value of AI: Unlocking scalable advantage”
• McKinsey & Company – “The State of AI 2025: How organizations are rewiring to capture value”
• European Commission – “Digital Decade 2025: Digitalization of Business in the EU Member States”
• ISACA Europe – “AI Use is Outpacing Policy and Governance, ISACA Finds”
• Bitkom (via Silicon Saxony) – “Bitkom: AI use is booming – but the fear of dependence on foreign countries is great”
• Gartner (via Beta Systems) – “Navigating the Future of I&O Automation: Key Insights from Gartner’s 2025 Hype Cycle”
• Eurostat “Digitalization in Europe – 2025 edition”

Continuation of the series

This article is part 2 of our four-part blog series on the transformation of automation in the age of AI. While part 1 looked at the strategic shift in perspective, this article shows why scaling AI will fail in practice if the automation architecture does not grow with it. In the next parts, we will shed light on the maturity levels that companies will need to reach in 2026 – and how roles, platforms and operating models will change when automation is no longer a tool but the basis for AI-enabled processes.