When I saw the film Her at the cinema in 2014 — a call-center agent falls in love with an operating system — it was science fiction. Today forty percent of young AI users would rather chat with AI than with their friends.
I watch with fascination as utopias become reality. What worries me is something else: the way many people in leadership positions are dealing with this transformation.
Recently I gave a talk to 150 HR managers, and was struck by the helplessness in the room. “Do we in HR even have the backing of our management to shape the AI transformation?” someone asked. Or: “We lack the competence to be an important factor.”
Yet I am convinced: the AI transformation is less a technical challenge than a human one. At its core, it is about the future collaboration between people and AI, less about tools and servers. A field study by Professor Danielle Li at MIT shows that the productivity of new employees in customer service in particular can be raised by up to forty percent with AI assistants.
And in the executive suites? There one often sees hesitation, worry, delegation to others — while companies in China, India and the US extend their lead at high speed.
Understanding together instead of guessing alone
Because no one can see the whole of this development alone, I founded the Zukunftsallianz Mensch & KI (Alliance for People and AI) with entrepreneurs, decision-makers and experts.
At one of our think days we dared to do a pre-mortem: what would you have to do to fail spectacularly? The answers came quickly, and recalled patterns that have already kept Germany from being among the front-runners in digitalization. Blind activity. Leaving AI to the technicians. Or waiting and eating chocolate because it will not be that bad.
At another think day we had an AI assistant with us as a sparring partner. When we asked it at the end what the elephant in the room was — the one thing we had not addressed although it was highly relevant — the answer was surprising: power structures.
Anyone who wants to shape the AI transformation successfully must be prepared to make their own power transparent as well: How do we make decisions? On the basis of which data? Which options did we assess, and how?
AI therefore demands an open, modest understanding of leadership that is oriented to the matter at hand and to purpose, and in which personal interests carry no weight. I have to be aware of this, and so should every person in a leadership role.
Two scenarios
Neither is destiny. As you read, check honestly which direction your company is currently heading in.
Mindset before technology
Many business leaders believe they must first formulate a strategy that describes the future in detail. That is often the most elegant excuse for not starting.
What is needed is a narrative that answers three questions.
Where do we stand right now? A radically honest stocktaking. It asks not only which use cases we want to solve but names the weak points: fragile processes, disorganized data, missing competence. Even if it hurts. Whoever talks only about opportunities builds the target picture on sand.
What is our target picture? What possibilities does AI open up for becoming a more future-proof company? These possibilities need to be translated into concrete use cases.
What are the next steps? Which results, derived from the target picture, do we want to achieve in the coming months?
Anyone who wants to use AI successfully over the long term has to tell the same story again and again, and it has to do with customers and their needs. It is like a good restaurant: the menu changes, but the level of service, the experience and the atmosphere remain.
You do not learn AI in a seminar but as a heavy user. Employees need room to experiment, and management has to create it.
Three insights from the Zukunftsallianz: no forbidden thoughts. Data and relationships are the new gold — what matters is data farming, that is, training the AI on your own definitions. And: stay independent. Listen to independent experts, not to those who want to sell you AI applications. You do not ask the frogs whether the pond should be drained.
Scale now costs money
Something has shifted. Vendors are changing their pricing models, away from flat fees toward billing by token.
How serious this is is shown, of all places, by a company that sells AI itself. According to a report in the German business daily Handelsblatt, SAP plans to budget the token consumption of its own employees — tiered from one hundred euros a month for the broad base up to five thousand euros for a few intensive users. Anyone who needs more on an ongoing basis gets approval from their manager. CFO Dominik Asam says the company is moving from piloting to industrializing.
So scale is no longer free. Costs grow not with the number of users but with the complexity of what they do: an agent that runs through several steps on its own consumes many times as much as a single query.
This is in tension with the autonomy heavy users need, and both are true. Unlimited access without a goal burns budget; a ban prevents learning. SAP solves it through transparency and budgets rather than bans. The answer lies in prioritization: whoever knows the most valuable use cases can be generous there and frugal elsewhere.
Success has three letters
Bring all key people into one room and derive the most important strategic use cases from the needs of the market and your customers. The most important thing is to set out together.
A client of mine rebuilt his company, a procurement platform for mechanical engineering, as AI-native. Now he has both: the existing company for buyers who value personal contact, and the AI-native version for customers who want to have everything done automatically.
What this text does not do
The two scenarios are exaggerated. In practice you find elements of both in the same company, often in neighboring units. The value lies not in assigning yourself to one scenario but in arguing in the leadership team about which line currently applies.
And the forty percent from the MIT study apply to new employees in customer service — a narrowly defined case with clear rules and a lot of repetition. They cannot be transferred to knowledge work in general. Anyone who uses them in a business case should include that caveat.