Everyone is talking about AI. Few know what to do with it.
Imagine you are sitting in a strategy meeting. Someone says: We have to do something with AI. Everyone nods. No one objects. And then—nothing happens. Or worse: everything happens all at once, without a plan, without prioritization, without measurable goals.
We have been seeing this scenario in almost every company for three years. Not because those responsible are incompetent. But because the AI landscape is changing so fast that even tech teams can barely keep up.
Over the past ten years, we have recorded 297 mentions of AI in our trend corpus. In 2017, there were exactly 6. In 2025, there were 86—the highest single value in the AI cluster of our dataset. This 14-fold increase is impressive. But a number alone says nothing about business value.
The explosion curve: 2022 to 2026 in numbers
Look at the development of the last five years. Not as an abstract trend line, but as what it is: a signal that has become too loud to ignore.
Each of these numbers stands for companies that either traded—or discussed PowerPoints.
The jump from 2023 to 2024 – from 26 to 64 mentions – is not gradual growth. That is a regime shift. And 2026 is already at 59 mentions from a single report. The signal is clear: AI is not on the horizon. AI is here.
A consulting firm that introduced an AI copilot for proposals in early 2024 has halved its bid preparation time. The win rate has increased by 4 percentage points. Not because the consultants have gotten better, but because they are faster. Every quarter you wait, the lead grows.
The vocabulary shift that most people overlook
What is hidden between the numbers is almost more important than the numbers themselves. Because the way people talk about AI has changed fundamentally since 2023.
Before 2023
Neural Network
Deep Learning
As of 2023
LLM
GPT
Coding Agent
Agentic AI
Before 2023, AI was a topic for data science teams. The keywords – Neural Network, Deep Learning – were technical, specialized, far removed from the day-to-day business of most employees.
Since 2023 dominate LLM, GPT, Coding Agent, and Agentic AI the discourse. The difference? These new forms of artificial intelligence do not just affect the IT department. They are changing how sales works, how support functions, how code is written. They are no longer laboratory experiments. They are tools that are on your employees' desks today—whether you planned for it or not.
Three highest-leverage use cases – now
Enough analysis. Let's get down to business. Our Innovation Radar identifies three AI use cases that can deliver the greatest business value in the next 24 months. No theory. No buzzwords. Just use cases with defined problems, measurable KPIs, and built-in controls.
Level of Evidence A
Proposal teams spend too much time searching for past decks, case studies, and delivery artifacts.
Human Approval
Prompt-Injection Testing
Evidence B
L1 support queues are overloaded with repetitive incidents and inconsistent ticket routing.
Escalation Fallback
Audit Logs
Level of evidence C
Manual remediation of legacy Python and PowerShell assets is slow and inconsistent.
Mandatory Test Gates
Secret Scanning
Take note of the evidence classes: A, B, and C. This is no coincidence. The Proposal Copilot has the strongest evidence base, while the Code Modernization Agent has the weakest. This does not mean that the third use case is unimportant, but rather that it requires further validation before you scale it up.
Just as important: What you should avoid
Strategy isn't just about deciding what to do. Strategy is also about having the discipline to say no. Our radar explicitly identifies what you should stop or put on hold.
Stop
Don't build your own foundation model training infrastructure. The costs are disproportionate to the benefits. Use existing models and invest in fine-tuning and orchestration.
Stop
Isolated chatbot pilots without explicit workflow integration and a sponsor with KPI responsibility. A chatbot without process integration is a demo, not a product.
Defer
Multimodal avatar assistants for external workshops. Technically fascinating, but the maturity level does not yet justify the investment.
The "stop" list is inconvenient. But it protects your budget from the most costly mistakes: projects that neither solve a measurable problem nor have a clear owner.
Regulation is coming – and that is good news
Many see the EU AI Act as a hindrance. We see it as a catalyst. That’s because regulation provides something the AI market has lacked until now: legal clarity. And legal clarity reduces the risk associated with investment decisions.
Here is the schedule you need to know:
In power
In power
In 4 months
In 5 months
In 17 months
The message is clear: the regulatory clock is ticking. But it is ticking predictably. Those who establish governance structures now will have a competitive advantage—not a burden.
Governance is not a roadblock. Governance is an accelerator.
That is not an opinion. That is a hypothesis that our data support.
If governance controls are embedded from the start—evaluation harness, human-approval gates, audit logging—then the pilot-to-production conversion rate more than 40 % within twelve months lie. First pilot samples point in this direction, even though reliable long-term data are still pending. Companies without embedded controls? They remain stuck in the pilot stage, quarter after quarter.
Teams with model routing, human approval gates, and audit logging scale safely into production—while others remain stuck at the pilot stage. This is not a hypothetical scenario. This is the reality we will observe starting in 2026.
The reason is simple: without governance, a point eventually comes when Legal, Compliance, or the executive board stops a project. Then you have invested months and are back to square one. With built-in governance, that does not happen because the controls run from day one.
The accompanying voices: Cloud and Semiconductor
AI does not exist in a vacuum. Two technologies form the foundation upon which every AI initiative stands—and both show a clear signal on our radar.
Cloud Computing records around 20 mentions over six years and holds the status ADOPT. But not as an independent innovation topic. Cloud is table stakes—a prerequisite, not a differentiator. If your company is still debating the cloud question in 2026, you have a different problem.
Semiconductor shows around 11 mentions, distributed intermittently over the years, with the status TRIAL. The chip crisis has shown how fragile AI infrastructure is. Anyone planning AI workloads must keep an eye on the semiconductor supply chain – even if it is not a day-to-day operational issue.
Together with AI, this results in a cluster of 328 combined mentions. The three technologies form a system – and should be evaluated as a system.
What does that mean for your company – concretely?
If you take away one thing from reading this article, let it be this: AI is no longer a technology topic. AI is a business topic. The question is not whether you use AI. The question is whether you do it systematically or in chaos.
ADOPT does not mean: everything at once. ADOPT means: Start with a clearly defined use case. Build governance in from the beginning. Measure success after 90 days. The three use cases above—Proposal Copilot, Service Desk Triage, Legacy Code Modernization—are proven entry points, depending on whether your biggest pain point lies in sales, operations, or engineering.
And stop launching isolated chatbot pilots without clear workflow integration. Without a measurable goal and a sponsor with KPI responsibility, AI pilots are just burning budget with a clear conscience.
The data is clear. The regulation is predictable. The use cases are defined. What is missing is your decision.