Session: Track 3 - Services
Date: June 4, 2025
Moderator: Pascal Eichenberger

At the 2025 Panoramai Swiss AI Summit, a high-stakes panel discussion revealed fundamental tensions in how Swiss public and private sectors approach AI implementation. While private companies race to deploy generative AI solutions, public sector leaders advocate for a more measured approach that prioritizes organizational readiness over technological novelty. The discussion, moderated by Pascal Eichenberger, brought together key decision-makers to examine real-world AI deployment challenges across healthcare, municipal services, energy, and insurance sectors.

The session highlighted a critical paradox: as Giorgio Pauletto from SEG revealed, « 61% of the people you have in your companies are using AI without telling you », yet institutional AI adoption remains deliberately cautious, particularly in public services where accountability and reliability are paramount.

The Organization-First Philosophy: Bienne's Strategic Approach

Barthélémy Rochat, Chief Digital Officer for the City of Bienne, emerged as the session's most provocative voice, challenging the technology-first mentality that dominates private sector AI discussions. His central thesis proved both controversial and compelling: meaningful AI implementation requires foundational organizational transformation before deploying advanced technologies.

« We should not deliver AI bullshit », Rochat declared, drawing sharp attention to what he sees as a dangerous trend in public administration. « If you try to do the opposite, and if you start with AI first, you just get public paid AI bullshit ». His criticism extends beyond mere inefficiency to a fundamental question of public responsibility—using taxpayer funds to pursue AI initiatives without clear value propositions.

Rochat's five-year tenure in Bienne has focused on organizational restructuring, process optimization, and cultural change. He argues that these foundational improvements have delivered greater efficiency gains than any AI deployment could achieve in the short term. When challenged by Eichenberger about whether AI could serve as a cost-saving mechanism for municipal governments, Rochat remained firm: « Not in the short term ».

However, Rochat emphasizes that his position is strategic, not Luddite. « I'm really a youth of fan of technology. This is part of my job », he clarifies. « But the reality on the field is that you can be more efficient when you work with people and culture and mindset ». His approach involves using AI pilots as catalysts for organizational change rather than standalone solutions, acknowledging that « we need to do both ways ».

Healthcare AI: The Three-Year Implementation Reality

Bayrem Kaabachi, a data scientist at CHUV (Lausanne University Hospital) and PhD candidate, provided sobering insights into the healthcare AI implementation timeline that challenges Silicon Valley narratives of rapid deployment. His experience developing clinical decision support systems reveals the complex journey from research concept to clinical practice.

"It takes time to go from those jupyter notebooks to an implementation inside the hospital," Kaabachi explained, describing a three-year process to deploy what could be developed algorithmically within weeks. The sepsis detection system exemplifies these challenges—while the machine learning models could predict sepsis risk effectively, the surrounding infrastructure, validation protocols, and clinical integration required extensive development.

The hospital's approach reflects broader healthcare AI challenges: balancing innovation with patient safety, integrating with existing clinical workflows, and ensuring physician adoption. Kaabachi noted that CHUV physicians may display "both extremes"—some are deeply resistant, fearing the potential impact of the technology on patient health and current healthcare practices, while others can be "over eager," driven by the promise of research opportunities and publication potential, while overlooking common pitfalls associated with the use of AI. Thus, part of the challenge as a data scientist in a hospital is communicating effectively with physicians by highlighting the true value of algorithms and prioritizing explainability over complex modeling.

The institution has invested heavily in foundational infrastructure, beginning with a 2017 Swiss Personalized Health Network initiative focused on data interoperability. By 2021, CHUV established a Biomedical Data Science Center with over 50 data scientists and biostatisticians, achieving notable gender parity (50% female) and international diversity (18 nationalities).

Kaabachi's current research focuses on responsible AI implementation, particularly differential privacy techniques to protect patient data when training large language models. His work addresses a critical healthcare AI challenge: ensuring that AI systems powered by sensitive medical data don't inadvertently expose individual patient information through model outputs.

Switzerland's AI Research Infrastructure: A Billion-Franc Foundation

Jan Kerschgens, leading AI initiatives at Innovaud (Canton de Vaud's innovation agency), provided comprehensive insight into Switzerland's substantial AI research investments. His presentation revealed the scope of Swiss federal commitment to AI excellence, anchored by world-class institutions and strategic partnerships.

EPFL (École Polytechnique Fédérale de Lausanne) serves as a cornerstone of this ecosystem, receiving « 1 billion every year which comes here to support training, research and technology transfer ». The institution's AI Center coordinates approximately 100 faculty members working on AI-related research, equivalent to over 1,000 researchers, while offering 175+ AI-related courses including executive education modules developed with IMD.

The Swiss Data Science Center (SDSC), launched in 2017 as a collaboration between federal universities, achieved National Research Infrastructure status in 2025, signaling its strategic importance. The center's partnership with Canton de Vaud involves 7.5 million Swiss francs over five years, providing companies with access to 150,000 francs in direct funding plus 250,000 francs in-kind support for AI innovation projects.

Kerschgens highlighted the University of Applied Sciences and Arts of Western Switzerland (HES-SO) led Swiss AI Center of SME’s  role in coordinating applied AI research across five universities in western Switzerland, offering structured support for companies, SMEs and start-ups seeking AI solutions. He also noted the Swiss Institute of Bioinformatics (SIB) as a « silent giant » whose protein databases were instrumental in training AlphaFold, demonstrating Switzerland's often-unrecognized contributions to global AI breakthroughs.

Private Sector AI Adoption: Hidden Innovation and Strategic Caution

Giorgio Pauletto, Head of Foresight and Innovation at SIG (Services Industriels de Genève), revealed the gap between official AI policies and actual employee behavior. His revelation that « 61% of the people you have in your companies are using AI without telling you » sparked considerable discussion about managing unauthorized AI adoption.

SIG's approach combines recognition of this reality with structured organizational change. Rather than restricting AI use, the company has trained 400 of its 1,700 employees in prompt engineering and AI applications, creating guidelines through experimentation rather than preemptive regulation. This strategy acknowledges that « people are using it, and for now, maybe later they will be more afraid. But for now, it seems that people really want to use it ».

Pauletto emphasized AI's transformative potential for workers traditionally challenged by documentation requirements. He described AI tools helping field technicians who install pipes to generate professional reports from simple notes, democratizing capabilities previously requiring additional support or external assistance.

However, Pauletto raised critical concerns about AI's energy implications, warning that « the pricing of AI is going to converge to the price of electricity » due to massive inference computing requirements. This perspective proves particularly relevant for an energy company executive, highlighting infrastructure constraints that could fundamentally reshape AI economics.

Yannick Hauser from Centre Patronal (insurance) provided practical insights into AI project management realities. His company's journey from initial ChatGPT enthusiasm to structured AI deployment took eight months for their first chatbot, significantly longer than initially anticipated. The experience taught valuable lessons about the difference between AI projects and traditional IT implementations.

Hauser identified a critical challenge in AI democratization: while the technology is accessible across age groups—noting that both his 7-year-old daughter and 75-year-old father use ChatGPT—organizational adoption creates new forms of digital divide. « You will have more disconnect between people that use it and know how to use it and people that don't use it and don't know how to use it », he warned, suggesting that AI could both bridge and widen existing inequalities.

Regulatory and Ethical Frameworks: The Swiss Advantage

The discussion revealed Switzerland's unique position in global AI competition, leveraging regulatory thoughtfulness and ethical frameworks as competitive advantages rather than constraints. Kaabachi's research into bioethics labs at ETH examining the « opportunity cost of using AI » reflects broader Swiss engagement with responsible AI development.

The healthcare panel particularly emphasized ethical considerations around AI deployment speed. While acknowledging pressure to implement beneficial AI tools quickly, speakers consistently prioritized validation and safety protocols. As one audience member noted, some medical professionals consider it « unethical to not use certain tools which are now becoming available because they're so efficient », creating ethical tensions around adoption timelines.

Switzerland's approach contrasts sharply with the « move fast and break things » mentality dominating Silicon Valley AI development. The Swiss model emphasizes sustainable innovation, rigorous validation, and stakeholder engagement—potentially offering more durable competitive advantages as AI matures beyond initial hype cycles.

The Global Context: Speed vs. Stability

Eichenberger's provocative closing challenged panelists and audience members to consider Switzerland's competitive position in global AI adoption. Citing examples like China's « Find a Good Doctor » platform treating 433 million patients annually through AI, and Amazon Health's $99 annual digital health offering, he questioned whether Swiss caution might disadvantage citizens who lack access to premium healthcare networks.

The session concluded with a revealing split when Eichenberger asked attendees to vote on AI adoption speed—roughly 50% favored acceleration, while 50% preferred continued caution. This division reflects broader tensions between innovation urgency and institutional responsibility that characterize current European AI policy discussions.

Yannick Hauser offered a nuanced perspective on this tension: « It's not about the speed. It's about thinking different ». Giorgio Pauletto reinforced this view with a compelling metaphor: « If you have the speed, you should also have the brakes. Otherwise you cannot steer your vehicle ».

Strategic Implications for AI Decision-Makers

The Panoramai panel discussion reveals several critical insights for AI strategy development:

Organizational Readiness Trumps Technology: Barthélémy Rochat's emphasis on foundational organizational change before AI deployment challenges conventional wisdom about rapid AI adoption. Organizations may achieve greater efficiency through process improvement and cultural change than through premature AI implementation.

Implementation Timelines Require Realistic Expectations: Bayrem Kaabachi's three-year healthcare AI deployment timeline provides sobering perspective on enterprise AI implementation. Organizations should plan for extensive integration, validation, and change management processes beyond initial model development.

Hidden AI Adoption Demands Proactive Management: Giorgio Pauletto's revelation about widespread unauthorized AI use suggests that organizations need proactive AI governance strategies rather than restrictive policies. Training and guidelines may prove more effective than prohibition.

Swiss Competitive Positioning: The discussion highlighted Switzerland's unique AI advantages—world-class research infrastructure, substantial public investment, and thoughtful regulatory approaches. These factors position Swiss organizations to compete on AI quality and reliability rather than speed alone.

Energy and Economic Sustainability: The growing connection between AI costs and energy consumption requires strategic consideration, particularly as inference demands scale. Organizations should factor long-term operational costs into AI investment decisions.

The Panoramai session ultimately demonstrated that successful AI implementation requires balancing innovation ambition with institutional responsibility—a particularly Swiss approach that may prove more sustainable than alternatives prioritizing speed over stability.


More on the panelists

• Bayrem Kaabachi (CHUV)

• Barthélémy Rochat (Ville de Bienne)

• Giorgio Pauletto (SIG)

• Jan Kerschgens (Innovaud)

• Yannick Hauser (Centre Patronal)

• Pascal Eichenberger, moderator - Elitebrain.ai