Part 1: Don’t Ask What You Can Do with AI – Ask What Problem You Want to Solve

AI Week Frankfurt 2025 has shown that AI has long since become a key challenge for strategic decision-makers in companies. But while the hype continues unabated, an uncomfortable truth remains: Of the billions of dollars invested in AI projects worldwide, only 5% actually make it into production and generate measurable added value. 95% fail or get stuck in the pilot phase.

Why is that? And more importantly, what needs to change for AI initiatives in companies to finally move from hype to reality?

The answer lies in a fundamental question.

The Core Problem: The Wrong Question at the Beginning

The core problem with many failed AI projects lies in the fundamental approach to them. Companies often start by asking themselves: “What can we use generative AI for?” But this is the wrong starting point – and therefore the beginning of the end for many AI projects.

This is because value creation is not created simply by using AI. Added value is created by using AI intelligently to do what you do today more productively: with fewer staff, higher throughput or to open up completely new opportunities, products and business models.

AI is a tool. Nothing more and nothing less.
And – this is crucial – it is not the tool for every problem.

The Right Logic: Problem ➔ Algorithm ➔ Data

This is why managers in companies should ask themselves the following three questions – in precisely this order:

1. Where do we have specific challenges or problems?
   More precisely: Which manual processes cost us the most time? Where do the most frequent errors occur? Where are we currently wasting resources? Or: Where do we lack resources?
2. For which of these tasks can automation bring real added value?
   And: Which technology fits this problem? Is it AI or would a simpler solution suffice?
3. Is our data suitable in terms of scope, variety and quality? Can we really solve the problem with this?

It sounds trivial, but it is fundamental. Because this reversal means that I first understand my problem. Then I choose the right algorithm – and this is not automatically generative AI. Then I check whether my data is suitable.

Those who take this path will not fall into the trap.

Why the Hype Is So Dangerous

One of the reasons why the hype surrounding generative AI is so dangerous? It is driven by big tech capital. The message is: “AI can do anything, and it’s so easy.”. Tools are springing up like mushrooms and actually make it supposedly easy to click together your AI agents as an application.

But simplicity does not necessarily bring success. I would even go so far as to say that this form of simplicity rarely brings companies the desired success in the long term.

The reality of the big LLMs is different. They are freemium business models. The financial stakes are huge. And there is a big bet behind this: after the relocation of data to the cloud, our intelligence is now being outsourced to other hands. In combination with throughput-oriented pricing models, this is a dangerous development for companies.

The Economic Reality: Costs That Are Exploding

Let’s take a look at the cost structure: with traditional IT, you pay for infrastructure and software. With generative AI, you pay per transaction – and the transactions scale with usage. This is not simply a different pricing model – it is a structural problem.

What’s more, generative AI lacks reproducibility. The same input does not deliver the same output. This leads to endless iterations in testing and operation. That’s not innovation, it’s a waste of resources.

The result: this is becoming an inflationary danger for companies. And for Big Tech? The business models will not work out mathematically. That is foreseeable – a self-fulfilling prophecy. The deal between Oracle and OpenAI is de facto a zero-sum game. But the stock market still believes it.

The further evidence is obvious: OpenAI, Perplexity and others are now launching their own devices and browsers on the market. Why? Because the real value is not in the AI – that’s just the tool. The value is in the data. In your data. .

Recommendation for Action: The Right Strategy

For decision-makers, this means: Don’t start with AI. Start with your problem. Really understand it. Then choose the right algorithm – this can be analytical AI, classic machine learning or generative AI. And then check: Do I have the data for it?

If you take this route, you won’t fall into the 95% failure rate trap.

In the next part of this series, we will show you how to make AI systems truly reliable – and why generative AI alone cannot be the answer.