Public administration and the broader public service sector are under immense pressure. In the public service sector, we are grappling with waves of retirements, an acute shortage of skilled workers, and, at the same time, steadily rising expectations from citizens. Today, everyone expects quick, straightforward answers—preferably around the clock. At the same time, in my daily work and in discussions with other cantons, I often find that there is still a great deal of ground to make up when it comes to understanding what artificial intelligence (AI) can truly do for us. The solution does not lie in simply automating processes aimlessly or placing a basic chatbot on a website. The future belongs to what is known as “Agentic AI”—and this will fundamentally change the way we work in the public sector.
Agentic AI explained simply?
Previous AI systems were mostly reactive: they provided a predefined answer to a specific question. Agentic AI takes a decisive step further. These are “digital employees” that not only carry out commands but also proactively pursue goals, independently plan complex tasks, and coordinate various steps. Instead of simply sending a citizen a link to a form, such an AI agent can check documents for completeness, specifically request missing data, and present a finished draft decision to the case worker.
The “Values Matrix” – Lessons from the Banking World
To determine where this technology truly helps us, I use an evaluation model that has already proven highly effective for me in professionalizing customer interfaces in the banking sector—including at Raiffeisen—namely: the so-called “Value-Irritant Matrix.” There, for example, we learned that the automated unlocking of a TWINT PIN offers no internal added value for the bank but is enormously helpful to customers in self-service. When it comes to complex consultations, however, people remain central. This logic can be directly applied to public administration:
- No added value for the customer or the administration: We should neither digitize nor retain these processes, but simply eliminate or radically simplify them.
- Added value for the customer, but not for the administration: Think of the hundreds of standard questions we answer every day, regardless of the department involved (e.g., regarding my office: “How many applications do I have to submit per month?”). For citizens seeking advice, these answers are essential; for us internally, they’re often just administrative time-wasters. This is the perfect use case for automation and Agentic AI.
- Added value for both sides: Here, human interaction remains irreplaceable. When people find themselves in stressful situations—such as suddenly losing a job—empathy is needed to alleviate shame and overcome initial hesitation. No machine can replace a human in this regard.
The major advantages of Agentic AI in public service
If we use the technology correctly, there are enormous benefits for government agencies and citizens:
- Real relief for staff: Instead of answering the umpteenth call about the same standard question, our employees can use their expertise to handle complex, consultation-intensive cases. The use of such systems can drastically reduce call volume and protect staff from burnout.
- Accessibility and Multilingualism: A huge advantage is the integration of dialect recognition. For people who feel uncertain about written official language or have visual impairments, intuitive, voice-controlled operation breaks down enormous barriers.
- A 24/7 service with no wait times:Citizens can resolve their issues regardless of office hours, without getting stuck in unpopular on-hold queues.
Successful real-world examples—Why the dialect matters
Various projects in the Swiss public service sector demonstrate that this is not just a pipe dream—projects I’ve been able to follow up close, in part through my involvement on the Advisory Board of Spitch AG. For example, the motor vehicle departments of the cantons of Aargau, Schaffhausen, and Zurich, as well as the civil registry office of the city of Kreuzlingen and the Baselland Public Unemployment Fund, are already successfully utilizing intelligent, AI-based voice assistants. In Kreuzlingen, a voicebot provides round-the-clock assistance with questions regarding civil status or ID orders, thereby significantly reducing the workload at the service counter. The same applies to progressive energy providers such as the Ticino-based Società Elettrica Sopracenerina (SES). A key factor in the success of all these projects is dialect recognition, as well as the ability to understand people for whom German is a second or third language. In Switzerland, we speak dialect in everyday life. When citizens can naturally express their concerns in Swiss German—and the system understands not only the words but also the intent behind them—it lowers the barrier to use enormously. It makes the technology accessible and user-friendly, rather than forcing citizens into a rigid “Standard German straitjacket.”
Data Protection, Transparency, and Legal Certainty with a Sense of Proportion
A key issue in every digitalization initiative is the extremely high standard required for data protection. There is no question about it: public administration processes highly sensitive personal data. Compliance with the Data Protection Act and the preservation of digital sovereignty are absolutely essential prerequisites. The revised Data Protection Act actually provides clear and sensible guidelines in this regard: For example, in the case of automated individual decisions, it mandates that data subjects be transparently informed and have the right to request a review of the automated decision by a human being at any time. To make this transparency tangible not only in individual cases but for society as a whole, I consider the initiative toward public AI registries to be very sensible. Such a registry would make it clear to everyone where and for what purpose the government uses algorithms. This breaks down the often-feared “black box,” makes potential risks of discrimination visible, and strengthens democratic oversight as well as the public’s essential trust in the digital state. However, based on my daily experience, I must issue a warning: Data protection and the call for oversight must not be allowed to become a brake on innovation. If government agencies get bogged down in endless theoretical concerns or if unclear lines of responsibility stall the process, we are standing in our own way. We need clear legal guidelines that provide us with legal certainty. However, we must be careful that excessive concerns and federal silo mentality are not used as excuses to block urgently needed developments.
The Risks—Where We Must Remain Vigilant
Despite all our enthusiasm for the technology, we must not turn a blind eye to the risks. In particular, I see four key challenges for public administration:
- The “human-in-the-loop” dogma: AI must never make sovereign decisions on its own in public administration. AI may only perform preparatory work in the background as a transparent “glass box”—the final decision is always made by and is the responsibility of a human.
- Blind trust in technology: We must not forget that AI is ultimately “only” a statistical model. It is a major risk if users accept these suggestions unquestioningly as established facts. Critical thinking is essential.
- Loss of humanity in crises: If we blindly automate everything, we eliminate the human factor precisely where it matters most. The initial contact in situations of personal distress is crucial for building trust. This must not be replaced by machines.
- Lack of Courage Among Decision-Makers: In my view, the greatest risk for the administration often lies in adapting too slowly to a digitized society. In some cases, decision-makers simply lack the courage and expertise.
Outlook—No AI Without a Solid Foundation
My long-term vision is a “single point of contact”—a single, central point of contact for citizens. However, I am by no means naive and am well aware of the enormous difficulties involved in implementation: Such a project is fraught with massive federal and organizational hurdles. My pragmatic approach is therefore this: We focus on what is feasible within our own agency. This is exactly the path we are currently taking in the Canton of Bern. Before rolling out complex AI tools, we’ve done our homework: We have established a centralized, professional customer service center with a single phone number for the Regional Employment Offices (RAV) and the public unemployment insurance fund—an approach that is unique in Switzerland. We’ve consolidated locations and reimagined our counseling approaches. Only through such a structural realignment can we create the organizational foundation that makes technological change and the future use of intelligent voice assistants, such as Spitch, a practical reality. At the same time, the profile of our employees will change. In the future, critical thinking, creativity, emotional intelligence, basic digital literacy, and a high degree of adaptability will be key to our success. Let’s not view AI as a threat, but rather as a tremendous opportunity to make public service more human again—by letting machines do what they do best, so that we have more time for people.
