AI in banking: what the Swiss financial centre already does, and what an SME can learn from it
Updated: 4 days ago
Banking is one of the sectors where AI arrived earliest, and one of those where it is most tightly regulated. For a financial centre like Geneva, this double movement is instructive: it shows both what AI brings when data is rich, and what it costs to deploy it without governance. We have worked for years with wealth management players and their platforms; here is what we observe, and what a financial SME in French-speaking Switzerland, independent asset manager, fiduciary or fintech, can take from it right now.
Where AI is already at work in Swiss banks
Contrary to the image of an AI "advising" clients, most of the use cases in production are discreet and operational.
Anti-money laundering and transaction monitoring. Models reduce the volume of false alerts that compliance teams have to process by hand. This is the oldest use case, and the one with the most measurable return: fewer hours of analysis for the same level of control.
Document processing. Extracting account-opening data, reading supporting documents, classifying incoming mail. In a private bank, this represents thousands of documents per month.
Client meeting preparation. A summary of a client's history, recent movements and past exchanges, produced before the meeting. The adviser keeps ownership of the relationship, but arrives prepared.
Internal assistants. Searching procedures, help drafting answers to regulatory questions, summaries of long documents. These are the uses that have spread most widely since 2023.
The common thread: AI has not replaced human judgement on the decisions that commit the bank. It has reduced the time spent before that judgement.
What FINMA expects, and why it also concerns smaller players
On 18 December 2024, FINMA published its Guidance 08/2024 on governance and risk management when using AI in supervised institutions. The text creates no new obligation, but it states clearly what the regulator expects: an inventory of AI applications with a risk classification, identified responsibility at management level, particular attention to data quality, testing before production and continuous monitoring afterwards, the ability to explain results, and an independent review of the most critical applications.
This framework applies first to banks and insurers. But an independent asset manager, a fiduciary or a fintech working for those institutions will be asked the same questions by its banking clients: which AI tools do you use on our data, how did you test them, who is accountable on your side? Better to have a one-page inventory ready than to build it in a hurry during an audit.
Add to this the European AI Act. Credit scoring of natural persons is listed among high-risk uses, which affects any Swiss SME providing that type of service to EU clients.
The case of a Geneva fiduciary firm: three months, one use case, one framework
A 30-person Geneva fiduciary firm, which handles accounting and tax for asset management companies, approached us in early 2025 with a simple request: "our bank clients are asking for our AI policy, we don't have one". We handled it in two strands, over three months.
The first strand was the framework: an inventory of real usage (four tools, two of them undeclared), a three-level classification, a one-page charter, a named accountable person within management. The second strand was a productive use case, so that the framework would not remain theoretical: preparing year-end closing files, where the assistant gathers missing documents, proposes a list of questions for the client and pre-fills the request letter. Measured gain on the first closing season: around 40 minutes per file, across 180 files, close to 120 hours over the peak period.
The firm was able to answer its banking clients with a three-page document. More importantly, it has a use case that pays for the effort.
What a financial SME can apply right now
Three lessons seem transferable without waiting to be a large bank.
Start with what is measurable. Document processing, file preparation, searching procedures. These are low-risk uses with a gain in hours visible from the first month.
Keep an inventory, however short. One page is enough: tool, use, data concerned, risk level, accountable person. It is the document everyone will ask you for, regulator, client or insurer.
Keep a human on the decision. AI prepares, summarises, proposes. The signature, the advice, the compliance decision stay with an identified person. It is the principle that structures FINMA's expectations and the only one that holds up in front of an unhappy client.
To see where your organisation stands against these practices, our AI Barometer offers a first diagnosis in a few minutes.
What we take away
Swiss finance did not wait for AI to be perfect before using it; it put it to work on precise tasks, under an explicit framework. That is exactly the posture we recommend to SMEs, in finance and beyond: a use case that pays, a framework that reassures, a person who is accountable. If you would like to build both at the same time, let's talk.
Comments