Bridging the AI PoC – Production Gap – Keys to Deployment Success in Swiss Financial Industry

Background 

The adoption of Artificial Intelligence (AI) in the Swiss financial industry has accelerated in recent years, yet a significant share of initiatives remains trapped in pilot stages, unable to deliver sustained business impact. This follow-up study, titled “Bridging the AI PoC–Production Gap – Keys to Deployment Success in Swiss Financial Industry”, directly addresses this challenge. Led by OST (Eastern Switzerland University of Applied Sciences), Institute for Finance & Law, in cooperation with SFTI (Swiss FinTech Innovations) and supported by ELCA Advisory, the study provides a rigorous and practice-oriented analysis of how Swiss banks and insurers can transition AI projects from proof-of-concept (PoC) into fully productive and scalable implementations. 

Objectives and Approach 

The primary objective of this study is to equip Swiss financial institutions with concrete strategies, frameworks, and tools to overcome the “pilot purgatory” phenomenon. The focus lies on identifying actionable pathways to unlock AI’s business value while remaining aligned with regulatory, governance, and operational realities when moving from pilot to deployment (?). Specific objectives include: 

  • Validating AI business cases and expectation management to ensure realistic ROI and stakeholder confidence. 
  • Defining robust model selection, testing, and monitoring practices to build trust and facilitate regulatory approval. 
  • Addressing infrastructure and system dependencies while ensuring data sovereignty and confidentiality. 
  • Establishing governance and compliance-by-design principles to embed trust and accelerate approval processes. 
  • Developing a pragmatic success checklist and best-practice framework for scaling AI initiatives. 

The study employed a mixed-methods design: a comprehensive online survey among 24 Swiss financial institutions (including two-thirds of SFTI members) provided quantitative breadth, complemented by in-depth expert interviews offering qualitative insights. This approach ensured both strategic relevance and practical grounding. 

Key Findings 

  1. AI Use Case Life Cycle: The portfolio of AI initiatives shifted upstream, with more institutions engaged in ideation and small pilots (40%), while fewer reached deployment (17%) or scaling (11%). Stricter governance and ROI discipline slowed deployments but increased success rates: 65% of production cases met or exceeded expectations. 
  2. Business Case Realization: Efficiency remains the dominant driver (100% of respondents), followed by internal user experience (71%). Cost reduction has declined as a decisive factor (42%), reflecting a shift toward productivity, quality gains, and employee enablement. Projects that reached production generally delivered on business case expectations. 
  3. Decision Criteria for AI Applications: Hosting, cybersecurity (17% each), and data sovereignty (15%) dominate vendor selection, confirming that compliance and risk management remain critical gates for deployment. Transparency and open-source flexibility are secondary but rising considerations. 
  4. Roadblocks to Scaling AI Beyond PoC: The most significant barriers are data privacy, security, and quality issues, as well as regulatory uncertainty and weak business-case proof. Technical limitations and legacy system dependencies further impede scaling, while cultural resistance and management support play a lesser role. 
  5. Success Factors: Winning deployments are driven by data access, executive sponsorship, and robust governance frameworks, reinforced by strong IT–business cooperation and MLOps capabilities. Testing, monitoring, and compliance-by-design approaches emerged as critical enablers. 
  6. AI Model Card Expectations: Respondents ranked bias, risks, and limitations, as well as intended use and evaluation, as the most important aspects for model documentation. Environmental impact and other non-core criteria received lower emphasis. 

Scalability and Recommendations 

The study demonstrates that scaling AI in Swiss finance requires a production-first mindset, anchored in disciplined data foundations, robust MLOps, and compliance-by-design. Institutions should introduce stage gates tied to deployability, cap concurrent PoCs, and require clear path-to-production artifacts at project inception. Business case discipline – tying outcomes to measurable KPIs and assigning accountable owners – is essential to sustain credibility. 

Key recommendations include: 

Mandate production-readiness from the start: Require deployable architectures, governance structures, and observability at PoC sign-off. 

  • Invest in data and MLOps foundations: Shared feature stores, automated monitoring, and reproducible pipelines are decisive separators between pilots and scale. 
  • Embed compliance early: Standardized model cards and policy-by-design shorten approval cycles. 
  • Focus business cases on user value and productivity: Prioritize employee-facing AI that improves throughput, quality, and adoption. 
  • Balance security with innovation: Ensure sovereignty and compliance through hosting and licensing safeguards, while leveraging scalable vendor solutions. 

Conclusion 

This follow-up study confirms that while many Swiss financial institutions remain in early AI adoption stages, those that succeed in moving beyond PoCs achieve tangible business impact. By bridging the gap between pilots and production, institutions can unlock AI’s transformative potential in efficiency, risk management, and user experience. The study provides a pragmatic roadmap – combining empirical survey evidence, case-based insights, and actionable tools – that positions financial industry players at the forefront of AI industrialization in Swiss financial services. 

Study: Final Report_SFTI_OST_AI FROM POC TO PROD_20251029_FINAL

Management Summary: Management Summary_SFTI_OST_AI FROM POC TO PROD_20251029_FINAL

Study Presentation at the Digital Finance Day 2025: SFTI OST ELCA – AI Production Gap_20251029_FINAL_NO NOTES

Contact for Additional Details 

The project co-leads are happy to provide additional details. 

Eastern Switzerland University of Applied Sciences (OST), Institute for Finance & Law 

SFTI (Swiss Fintech Innovations)