Céline Solenthaler is a digital transformation expert and the founder of solenthaler.ai. She helps Swiss SMEs and government agencies recognize and harness the potential of AI. Previously, she served as Chief Digital Officer for the Zurich City Police.
Céline, you are self-employed and work with companies to implement AI. How did you come to be so deeply involved with AI and now support companies in this area?
I have been working in digital transformation for over 15 years, in a wide variety of roles: As a technical writer, as a UX designer, and most recently as Chief Digital Officer for the Zurich City Police. This has given me an understanding of the entire development process, from the initial idea to the solution that has to work in everyday life. When AI made the leap from experimentation to everyday work, I wanted to be right in the middle of it and help shape the future myself: closer to implementation and working with many organizations at the same time. Since October 2025, I have been working as a freelancer to help companies use AI strategically. The common thread has remained the same: I bridge the gap between the technology and the people who work with it.
What is your take on this: Are companies currently buying too much AI and not thinking enough about what they actually need it for?
That happens, of course. Some companies pay for Copilot licenses that hardly anyone uses. More often, though, I see the opposite: They are not investing enough. I regularly hear the question, «Can’t we just share access?» And many companies have not even set up an official solution yet. Employees still realize how useful AI is. But they do not want to pay for their own account, so they use the free version. That is exactly how shadow AI arises. Company data ends up in tools that no one has control over. The supposed cost savings turn into a risk. Anyone who wants to prevent this must offer a good official solution and clearly define what is allowed. A free template for this is available in my sample AI policy for SMEs.
What are the biggest challenges you currently see companies facing when it comes to using AI?
The first hurdle is imagination. Many people do not even know what is possible with AI. AI can do much more than just write nicer emails. That is why concrete examples are needed to inspire them. Once the first use cases are identified, the second hurdle arises: D ata. Either the data resides in systems that cannot be integrated – such as industry-specific ERPs – or the digital infrastructure is not yet in place, and the data is scattered across folders, drives, and desks, both digitally and in physical form. You can still get started, though. Some use cases require very little company data – social media posts, for example. But the real impact comes only when AI can work with the company’s own data. The third hurdle is people. When AI takes over tasks, responsibilities shift: Who reviews the results, who bears responsibility, and what does the person who previously handled this work do now? Questions like these create uncertainty, and uncertainty slows progress. If you do not address them early on, you will end up with a functioning tool that no one uses.
What exactly is the difference between an AI tool and an AI agent?
The difference lies less in the product than in how it operates. In a chat, the AI responds to a request, and I decide what happens with the result. When it operates as an agent, it is given a goal and works its way through several steps independently: It reads emails, searches for information in the CRM, writes a draft, and saves it. ChatGPT, Claude, and Copilot can now do both. So the question is not which tool to buy, but how much work you want to let the AI handle on its own, and what permissions you are willing to grant it to do so.
The selection of AI tools and AI agents seems to be growing every day. How is a company even supposed to decide which tool truly fits its daily work?
Start with your day-to-day workflow, the tool comes last. First ask: What costs time, money, stress? And does it even require AI? Only then should you turn to the tool, and four things matter here. First, data protection: A business version, no use of your own data for training, and clarity about where the data is stored. Second, the system landscape: If you work in Microsoft 365, Copilot is an obvious starting point. If you work in Google Workspace, Gemini is. Third, the people: The best tool is the one that gets used. That is why you should test it on a small scale with the people who will actually be working with it and evaluate the results honestly after a few weeks. Fourth, the company’s underlying philosophy. Some companies deliberately choose not to become dependent on American tech conglomerates and rely on European providers – often extending beyond AI as well. For others, the provider’s origin is less important. They make pragmatic decisions, for example, choosing Microsoft – because their customers are already using M365, which simplifies collaboration. There is no right or wrong here, but the decision should be a conscious one. And because the tool landscape changes monthly, it is worth staying flexible: Store knowledge and context in SharePoint, Google Drive, or Notion and connect them to the AI tool, rather than storing everything within the tool itself.
What surprises you time and again in your consulting work? Are there instances where companies come to you with a very specific expectation, and you end up saying, «No, you actually need something completely different»?
It hardly surprises me anymore, but it surprises the clients all the more. Often, companies come to me with a problem they want to solve using AI. Then it turns out: They do not need AI at all. Sometimes simple automation or a well-designed form is enough. And sometimes the problem resolves itself when people collaborate differently. Here is an example: Of course, you can transcribe a meeting and have the minutes generated automatically. But maybe the meeting is not even necessary, and a regular exchange via Teams is more effective. That is why my rule is: Process before tool. Because AI does not solve chaos. It just makes chaos happen faster.
If you could give companies just one piece of advice right now, before they purchase the next AI solution, what would it be?
Define the problem before you look for the solution. With AI, we ar e currently experiencing the opposite: We have a solution and are looking for problems that fit it. Those who outline the problem first know what they actually want to achieve and can set metrics. Those who run around with a solution and look for problems to fit it usually don’t have these metrics. And in the end, they can’t say whether the investment was worth it.