AI in financial services: From experimentation to trust

AI in financial services: from experimentation to trust

It is no longer a matter of whether AI works in financial services. What matters is which applications create demonstrable value, how they can be integrated into the core business and under what conditions we are willing to trust them. We can only assess whether such systems deserve trust if we understand and explain the results, identify where the relevant risks materialise, determine how they can be mitigated, and establish who ultimately assumes responsibility.

The issue of trust becomes even more pressing with the emergence of AI agents. Such systems do not merely provide answers. They can plan tasks, execute individual steps, access other systems and act independently within a defined or even undefined framework. This opens new opportunities but also raises fundamental questions for financial institutions around responsibility, control, client profiles and access to services. While agents can operate within predefined permissions, defining the appropriate scope of those permissions and maintaining effective control over their actions are key trust challenges.

Under the title “Building Trustworthy AI Frameworks in Financial Institutions”, SFTI | Swiss Fintech Innovations will bring together representatives from banks, insurers, technology companies and consulting firms at the Swiss AI Summit. The roundtable will explore what trustworthy AI means for financial institutions in concrete terms – not as an abstract guiding principle, but as a practical design challenge.

When prototypes become business reality

Many financial institutions now have a considerable number of AI initiatives. The greater challenge is often no longer to identify a possible use case. Today’s real challenge is selecting the relevant use cases, measuring their benefits properly, and taking promising solutions beyond the pilot stage.

Even when all relevant aspects of the SFTI Framework have been considered, from governance and compliance through organisational readiness and ongoing monitoring, a technically convincing prototype is not yet a viable application. When moving into production, additional questions arise:

• Is the output reliable even under changing conditions?
• Who bears business responsibility?
• How is the application monitored?
• And does the expected benefit justify the effort and risk involved?

Trustworthiness and scalability are sometimes treated as opposites. Governance may then appear as an additional hurdle that slows down promising innovation. A well-designed framework, however, can have the opposite effect. By establishing clear approval criteria, defined responsibilities, and continuous monitoring, governance can provide the confidence needed to scale AI solutions responsibly and efficiently.

It is therefore particularly interesting to discuss how institutions recognise scalable use cases, which criteria they use to prioritise initiatives and how they decide whether to continue, adapt or stop a pilot.

Trustworthy AI becomes a practical question

The question of trust becomes particularly visible where AI no longer merely supports analysis, but moves closer to decisions, processes and client interfaces. AI agents could in future collect information, compare offers or guide clients through financial decisions.

This raises new questions: Who controls this interface? According to which criteria does an agent act? How transparent must recommendations be? And when is human intervention required?

Trustworthy AI thus becomes highly practical. It is about reliable systems, high-quality data, understandable results, clear responsibilities, ongoing monitoring and a culture that empowers employees to use AI responsibly and challenge outcomes when necessary. In the financial sector in particular, trust must not only be asserted – it must be embedded in the organisation. Governance is therefore neither a purely technical nor a purely legal task. It emerges from the interaction of business, IT, risk, compliance, legal and leadership.

What financial institutions need to discuss now

The SFTI roundtable at the Swiss AI Summit will address how financial institutions can enable innovation without losing sight of trust, control and responsibility.

It is not intended to present a supposedly definitive blueprint. Business models, risk profiles and use cases differ too widely between individual institutions. Instead, the focus will be on practical experience, different positions and the questions that have not yet been conclusively answered.

Because trust in AI is not created by a single control instrument. It is created through comprehensible decisions, clear responsibilities and the ability to learn from experience. Which elements financial institutions should address first will be one of the questions to be discussed at the roundtable.