2026 is the year in which AI must be transferred from the experimental stage to the routine. Many companies are stuck between two poles: on the one hand they want to "use AI", on the other they don't know where to start. The danger is that they either invest too broadly (trying everything everywhere) or too narrowly (only one pilot in one department).
A three-horizon model
We have established a framework that helps managers to set priorities:
Horizon 1: Efficiency (3-6 months) Use cases that make existing work faster or more accurate. Examples: Support wizards, document summaries, categorization of incoming mail. ROI is measurable and occurs within a few months. Risk area: low. Investment: manageable.
Horizon 2: Growth (6-18 months) Use cases that open up new business value. Examples: AI-supported product recommendations, personalized self-service platforms, automated quotation generation. The ROI is indirect (more sales, higher conversion), but requires a learning phase. Risk area: medium. Investment: considerable.
Horizon 3: Transformation (18+ months) Use cases that change the business architecture. Examples: Fully automated case handling, autonomous supply chain decisions, radically new customer experiences. ROI is difficult to strategize and time. Risk area: high. Investment: significant.
The mistakes we observe
**Mistake 1: Horizon 3 ambitions without a Horizon 1 foundation ** Anyone who wants autonomous decisions to be made without clean data pipelines is building on sand. Horizon 1 is not just efficiency - it builds the foundation for the later horizons.
**Mistake 2: Too many Horizon 2 projects in parallel ** Each growth project ties up resources for 6+ months. Three parallel initiatives mean 18 months of tied-up capacity. Concentration beats breadth.
**Mistake 3: Confusing horizon and technology ** "We are doing AI now" is not a roadmap. The question is: What exactly do we want to achieve? The technology results from this, not the other way around.
What can be decided in 2026
Three decisions should be made this year:
- Platform decision: Which infrastructure will support our AI initiatives over the next three years? The choice between hyperscaler, sovereign cloud and on-premise has long-term implications. Once made, very expensive to revise. 2 Data governance: Which data assets may be used and how? Who decides on new use cases? Without clear governance structures, every single project comes to a standstill when data is released. 3 Competence strategy: Do we build AI expertise internally, or do we work with partners? The answer is rarely black and white - usually a mixture. But the weighting must be made consciously.
What can wait
Model choice: The question of whether GPT-5 or Claude 4.6 or an open source model is "better" is overrated. For 80% of operational use cases, model quality is not the bottleneck - it is the data pipelines, context preparation and orchestration.
Marketing communication: Talking about AI externally before AI works internally is risky. Credibility is tested by the first visible error. First deliver, then communicate.
The next step
An honest readiness assessment is a good starting point. Which Horizon 1 use cases can be implemented within 6 months? What data is ready? What governance gaps need to be closed before Horizon 2 can get off the ground?
We regularly accompany such assessments - usually in 4-6 weeks, with concrete prioritization recommendations at the end. More on request.