“We are all bloody users.” It was one of those sentences that was not intended to become the headline of our conversation, but in hindsight probably captures it better than anything we had prepared.
It came from Friedemann Wecker, CEO and shareholder of Bauck, during our latest Consumer & Retail Executive Call. Bauck is a German Mittelstand company producing organic food, with a history going back more than 90 years. It is certainly not the first company you would associate with the current AI hype. Which is exactly why I found the conversation so interesting.
Friedemann does not position himself as an AI expert. Quite the opposite. His point was that nobody in his organisation should pretend to have figured this out – including the CEO. Technology is developing too quickly for that. What management can do, however, is create an organisation that learns how to use it.
Looking back at Bauck's journey over the past few years, there is one aspect I found particularly relevant from a leadership perspective. The company did not start with an AI strategy, a governance framework or a list of efficiency targets. It started by creating excitement. And perhaps we underestimate how important that is.
We tend to use the word “hype” negatively, particularly when talking about AI. There is obviously good reason for some scepticism. Hardly a day goes by without another tool, agent or allegedly revolutionary use case being presented as the next big thing.
From a change perspective, however, hype also has a useful function: it creates energy.
At Bauck, the first phase was deliberately broad. People were encouraged to experiment with ChatGPT and other applications. Friedemann brought his leadership team together for several offsites and mini-hackathons. Teams presented what they had tried, compared results and challenged each other. There was an element of gamification to it, and that was intentional.
The company later introduced what it calls the “AI Americano”: a short, regular internal format in which people share new developments and examples from different parts of the business. The interesting thing about the format is that it is not primarily management communicating about transformation. Employees see colleagues experimenting with AI in purchasing, packaging, HR or other functions. The change becomes tangible.
Friedemann describes this as horizontal integration. Before going deeply into processes, Bauck wanted to get a broad part of the organisation using AI, talking about it and understanding where it might be useful.
I think there is a broader lesson in this. We often design transformation from the end backwards. We define the target operating model, introduce governance and establish rules before enough people have developed a genuine interest in the subject. Bauck took a different route: create the energy first and structure it afterwards.
That does not mean allowing experimentation to continue indefinitely. In fact, the point at which Bauck started introducing more structure is probably the more interesting part of the story.
After the third hackathon, Bauck had identified 117 potential AI use cases. For a company of its size, that is an impressive indication of engagement. It is also far too many initiatives to pursue seriously. This is where the nature of the discussion changed. The relevant question was no longer what people could do with AI, but what the company should actually do with it.
Bauck established an AI Circle and started evaluating use cases according to their expected impact and complexity. Relatively simple applications can be developed internally. More complex projects receive external support. Others are deliberately not pursued. In other words, the hype had done its job. It had generated curiosity, ideas and involvement. Now it had to be translated into governance and prioritisation.
I find that sequence important because I see both extremes in organisations. Some are so concerned with governance, data protection and potential risks that they struggle to create meaningful adoption in the first place. Others have hundreds of people experimenting with AI but no mechanism for turning that activity into organisational capability. Neither is particularly useful.
The management challenge is to recognise when the organisation needs to move from one phase to the next. Initially, leadership needs to give people permission to explore. Later, it has to become much more selective.
This selectivity also applies to the way companies assess the value of AI. The obvious business case is productivity: fewer manual tasks, faster processes, lower costs. Bauck has applications that address exactly that, but some of the examples Friedemann shared point towards something more interesting.
In sourcing, for example, AI can combine information on agricultural developments, weather patterns and their potential impact on crops. If this enables purchasing to understand market developments four or six weeks earlier, the value is not primarily a reduction in working hours. It is a potentially better commercial decision.
Another example is meeting management. Bauck is working on connecting transcription with Asana so that decisions and tasks move directly from meetings into the company's workflow. Again, there is an efficiency argument. But Friedemann's broader point was about the quality of meetings, preparation and follow-through.
The same applies to an AI leadership agent Bauck is developing based on its own leadership principles and frameworks. It can support managers in preparing employee conversations, remembering commitments or thinking through a conflict. This is particularly interesting in a mid-sized organisation that will never have the HR infrastructure of a multinational. The common denominator is not simply doing existing work faster. It is improving the quality and consistency with which an organisation works.
That distinction matters. If the only question we ask about AI is how many hours it saves, we may overlook applications that improve decisions, provide access to expertise or allow smaller organisations to build capabilities they previously could not afford.
For me, the most strategic part of the conversation came when Friedemann moved away from Bauck's current applications and talked about the cost structure of established companies.
A traditional consumer business has accumulated organisational infrastructure over decades. Finance, controlling, HR, administration and management are necessary parts of running the company, and their costs ultimately need to be carried by the products it sells.
An AI-native company may start from a very different position. Friedemann referred to examples of businesses operating with a handful of employees and dozens of agents. Not everything these companies claim will stand up to scrutiny, and translating such models into food production is obviously not straightforward. But the underlying question is worth taking seriously.
What happens when AI-native companies start competing in traditional consumer markets? The discussion about AI then moves beyond individual productivity gains. It becomes a question of operating models and structural competitiveness. Can established Mittelstand companies retain the knowledge, quality, customer relationships and experience that make them successful while reducing some of the organisational complexity built up over time?
Friedemann calls this AI readiness. He currently rates Bauck somewhere between six and seven out of ten. What I liked about that assessment was not the number itself, but the absence of any claim that the journey is finished. His expectation is that the technology will continue to change, which means the organisation has to become better at adapting to it.
There is another reason why governance becomes more important once experimentation scales: some technically possible applications simply do not fit the company.
Bauck has, for example, decided against introducing an AI customer service agent for now. Friedemann's reasoning is very pragmatic. The company has built considerable trust in its brand. Anyone who has spent ten frustrating minutes trying to escape an automated customer service loop will understand why he is reluctant to put that trust at risk for a relatively obvious efficiency gain.
The same applies internally. AI could be used to analyse employee chats and conversations and provide management with information about sentiment or emerging problems. Bauck does not do this. It does not fit Friedemann's understanding of leadership.
I find these examples more useful than abstract discussions about responsible AI. Governance becomes real when it forces management to decide not only where AI creates value, but also where efficiency is less important than trust, culture or the quality of a human interaction.
Towards the end of our conversation, we discussed what all of this means for leadership itself. Friedemann's answer was relatively traditional: experience, empathy and judgement will remain important. I suspect they may actually become more important.
If AI can analyse information, generate alternatives and prepare a management discussion within seconds, producing another option becomes less valuable. Assessing those options in context becomes more valuable. Leaders will still need to understand the shades of grey that do not sit neatly in the available data, and somebody will still have to take responsibility for the decision.
Friedemann used an expression from another AI workshop that I liked: AI can be instructed to “be the pain in the ass”. It can challenge an argument three or four times, look for weaknesses and force a leader to reconsider an assumption. That is an excellent use of AI. It is not the same as delegating the decision.
He also connected this to the Kaizen idea that most decisions should be made close to the shop floor, rather than travelling through several layers of management. Better access to information and analysis could reinforce that principle. If employees have stronger decision support, organisations may be able to distribute responsibility further.
This is where the discussion about agents becomes interesting beyond the technology itself. Hybrid teams of people and agents may change how work is organised. They may also change what we expect from managers.
Which brings me back to Friedemann's sentence. The first requirement for broad AI adoption at Bauck was not technological. It was psychological safety. People needed to believe that experimenting was expected, that getting something wrong was acceptable and that management was learning as well.
“We are all bloody users” is a surprisingly good leadership statement in that context. It does not remove accountability. It removes the pretence that the people at the top already have all the answers. Bauck's journey is still work in progress, and Friedemann was very open about that. There was no master plan at the beginning. Some experiments worked, others did not. The company created enthusiasm, learned from it and gradually introduced more structure. Perhaps that sequence is more relevant than any individual AI tool we discussed.
Create enough excitement for people to start. Give them room to experiment. Watch where genuine value begins to emerge. Then introduce governance, make choices and move from individual usage towards organisational capability.
Friedemann summed it up towards the end of our conversation in a sentence I would probably put above any AI strategy: AI is not the transformation. Organisational learning is.