AI-driven advice with MCP: a new era in client interaction  

From experiment to reality: how agent-based AI is changing advice 

Until now, Swiss banks have mainly employed AI in client interaction in the form of voicebots and chatbots. These applications enhance client contact in certain respects but fall a long way short of harnessing the full potential of today’s AI technology. We must not let caution stand in the way of progress. This warning from Bill Gates applies equally to the Swiss banking centre: “People often overestimate what will happen in the next two years and underestimate what will happen in ten.” 

A survey conducted at the SBA’s Digital Finance Day on 29 October 2025 revealed that many banking experts expect to see AI-driven, partially autonomous advice within as little as three to five years. Tax, succession, investment and pension advice as well as wealth management seem especially promising. Agent-based AI goes far beyond the scope of text-based assistants. It will in future be able to plan, prioritise and execute all by itself – individual steps at first, then later complex multi-stage working processes. This should significantly improve both efficiency and the client experience while also creating new earning opportunities.

Requirements: data access, trust and regulation 

A number of fundamental challenges must be overcome in order to make the most of agent-based AI. Trust in AI agents is only built up slowly through repeated use and systems that make transparent, verifiable decisions. Authorisation models and liability issues also have to be clarified unequivocally. Legal and data protection requirements additionally play a key role. Another prerequisite is security mechanisms such as strong authentication, identity management and explicit user consent. 

Access to data and functions via programming interfaces is especially vital. Switzerland is well placed in this respect, with the first harmonised interfaces already in place and more in the planning and implementation stages.

Technological bridge: the Model Context Protocol (MCP) 

Another key technology is the Model Context Protocol (MCP), which was published at the end of 2024 as an open-source standard. MCP defines how digital services can be structured so that AI systems can access them in a controlled and secure manner and interact with them. It provides functions, data sources and reusable templates that serve to link banks’ models and systems with each other. MCP thus creates a basis for using agent-based AI in a genuinely productive way in day-to-day business. 

Swiss Fintech Innovations (SFTI) is currently leading the development of a reference architecture that describes the technical building blocks needed for access by agent-based AI.

What banks have to do today 

In parallel with this, however, banks themselves need to make a few first steps: 

• Define the importance, role and embedding of agent-based advice 
• Build up in-house expertise 
• Modernise their data and interface landscape 

Agent-based AI is not about to make a sudden breakthrough, it is set to trigger a gradual but far-reaching transformation. Institutions that play a part in shaping this transition will be able to offer advisory experiences in a few years’ time that are more personal, efficient, reliable and accessible to a broader group of clients than ever before.