# How to AI-Transform a Company: Vision and Leadership First, Technology Last > In episode 7 of Prompt & Proper, Steffen Ochsenreither and I take on the transformation itself. The most common question is still the tooling question, yet no tool transforms an organization. The usual maturity frameworks start with technology, which puts them upside down. Plus my AI maturity matrix, the transformation framework we run in practice, and the real lever: transformative leaders who go ahead as beacons. - Date: 2026-09-02 - Source: https://romanoroth.com/en/podcast/e07-wie-man-ein-unternehmen-ki-transformiert/ - Tags: AI, Podcast, Prompt & Proper, Transformation, Leadership, AI Strategy We have spent a lot of this podcast on the technology and on the effects of AI. In episode 7 of **[Prompt & Proper](https://promptandproper.ai/)**, Steffen Ochsenreither from Swiss Post and I take on the transformation itself. Not what AI does inside a company, but how you point a company at AI. The episode is in German, and this post captures the core of the conversation.
## The Tooling Question Is the Easy One Steffen says the main question is still the tooling question. Which tool do I take, which model do I take. I see it too, most recently at two conferences in Berlin, CIO Connect and the Industry Forward Summit. And I see it at home. I pushed Claude Code internally quite hard, and now we are renegotiating our licenses. That is a real consequence, and it is not a transformation. In my view the tool question is the easiest one there is. You put the tool in front of people and you have changed nothing yet. Steffen has the comparison that lands it: > "Even the big tools, SAP and so on, that is the tool. SAP does not transform a company either. It is the usage of the tool." So the question behind it is how you secure usage in the first place. That is where the work starts. ## The Maturity Frameworks Are Upside Down Steffen has been through a range of AI maturity models. Most of them open with how well the company is set up on technology. For him that point belongs at the end. His sequence runs from vision through leadership to skills, and technology and data come after that. Without access to the data and without a clean data model it gets hard, he does not dispute that. There are simply around fifteen other steps in front of it, and those are where you start. ## My AI Maturity Matrix: Four Stages Across All Roles At Zühlke I introduced an AI maturity matrix for this. Stages on the horizontal, our roles on the vertical, and each field spells out what the stage means for that role. The funny part: tools barely appear in it. Stage 0 means you do not use AI in your daily work at all. Stage 1 I called capable, you experiment sporadically, privately and at work. Stage 2, adopting, is where it gets interesting. AI is the default tool, and when you hit a problem your first question is how to solve it that way. Stage 3 is transformative. AI is fully integrated into your work, you train others, you rebuild whole processes and you are a champion in your field. Steffen pushed back straight away: anyone using Microsoft Office or Google search is exposed to AI already, so are they automatically at stage 2? No. A Google search stays stage 1. Stage 2 means the engineer works with coding agents by default, often several at once, and the architect puts architecture questions directly to an agent. At Zühlke the target is clear. We do not want anyone at stage 0 going forward, which is why the education program is running now. The absolute minimum is stage 1. I do not expect that to hold for long, though. Somewhere next year stage 2 will be our minimum. ## The Transformation Framework: Assessment, Vision, Roadmap, Portfolio The matrix is the personal dimension. Next to it sits the transformation of the company. At Zühlke we have our own framework for running AI transformations with clients, and we are currently applying it to ourselves. The sequence: first interviews with the leadership team and selected employees, so it becomes clear where we stand. That is a classic maturity assessment, and it makes the gaps visible. Then an AI vision workshop where we define the target picture. I used to work with a five-year horizon for this, today it is two years. From the target picture you derive the roadmap, and from the roadmap the initiative or use case portfolio. One piece that is often missing: > "Every use case you have should follow the strategic goals of the company and should pay into a company KPI." Without that anchor you collect initiatives nobody can measure. ## The Three Keys: Technology, Processes, Organization Steffen wanted the three things you cannot do without. You ask a consultant for three points, you get three points. The first point is technology. The technical foundation, and at its center the AI platform, the agentic gateway, whatever you call it. It matters most when token prices rise and you want to decouple from individual vendors. On top of it sits skill management, meaning skills as Markdown files for the models, and that is where your agents live too. The second point is the processes. You have to analyze them and change them, and you do not coordinate that from an ivory tower. We appoint AI champions in the country organizations and units who run the transformation locally and rebuild the processes on the ground. The third point is the organization. This is where the security people get restless: you have to share the data. The models need access, structured and unstructured, so agents can work on it. Which means in the end that the organization has to be built along the value chain. Line and process organization align to a product or a service, and that unit has access to every piece of data it needs to deliver. Steffen's comment: each of those three is a long project on its own. And potentially its own episode, which is now on the backlog. ## The Vision Problem Steffen sees one gap ahead of all of this. The vision is usually very vague. Writing a vision that holds for five years is hard in this period, and in his view it does not have to. The real problem sits elsewhere: the vision for AI use either clings to tooling or goes ultra broad with "we use AI everywhere". Neither translates for the management layer. What does it mean for the individual employee? A lot of potential is left lying there. ## The Executive Board Goes Through the Training In many companies I see that the topic is not really understood at executive or board level. What this technology is and how you use it profitably stays unclear. That is why I decided at Zühlke that the entire executive board goes through one of our trainings for a full day. They did not like it at all, which I understand when you pull people out of the business for a day. The content was governance, the question of which AI initiative actually makes sense, and the difference between an LLM, a RAG and an agent. I showed clearly why hallucinations happen and where the limits are. Exactly that takes the "magic" out of the topic. The effect is visible since: everyone pulls in the same direction with the same understanding, and that pays straight into the vision. Steffen agrees, with one caveat that lands: > "You forget a generalist training as fast as you signed up for it." A baseline enablement is needed, people have to understand what this technology does. After that the training has to get specific about what actually helps them in their daily work. Otherwise usage does not stay high. ## Limit Awareness The point Steffen considers most important for team leads, management and boards is limit awareness. He does not mean which task a model cannot do yet. He means the limits that come out of your own organization, and the limits of your own technology platform. Anyone who does not know those limits lands in one of two errors. Either they assume everything works without friction. Or they never start, out of fear the thing could grow too big. ## Transformative Leaders Are the Key That closes the circle back to the matrix. The managers who interest me most are the ones using AI completely. Who run sessions, who are further along than their engineers and show how they work with it. Those people are not at stage 2, they are already transformative. And you need a set of them in the right places. People who act as influencers and front runners and make the direction visible. In my view that is the key to an AI transformation. ## The Take-Away Steffen's summary at the end of the episode: you need a vision, meaning a concrete plan for where you are going. You need a set of leaders who carry that vision actively, rather than going along because the CAIO said so. Those leaders live it and build the structures for it. Anyone who will not free up their people for a baseline training does not get far. An assessment makes the current state transparent, and that is a good place to start. If you enjoy the episode, subscribe to **Prompt & Proper** on [Spotify](https://open.spotify.com/episode/7vEr25ARvQzAVULdVvIYGI), [Apple Podcasts](https://podcasts.apple.com/us/podcast/prompt-proper/id6792786850), [Amazon Music](https://music.amazon.com/podcasts/b8a56367-a595-4865-8e73-97c706fa4743) or [YouTube](https://youtu.be/COwYN2fdg3Q) and tell us which AI topic you are wrestling with right now. All episodes: [promptandproper.ai](https://promptandproper.ai/).