Artificial intelligence in banking, insurance and compliance

The state of adoption

The most reliable picture of AI use in the Swiss financial sector comes from a survey conducted by FINMA between late November 2024 and mid-January 2025 among around 400 supervised institutions, including banks, securities firms, insurers, fund managers and financial market infrastructures. About half of the institutions surveyed used AI or had initial applications in development, and a further quarter planned to introduce AI within three years. Institutions using AI had on average about five applications in operation and nine in development, with larger firms deploying more. Of the institutions using AI, 91 per cent also used generative AI such as the newer chatbots. FINMA noted that this increases dependence on a few large technology providers.¹

The same survey found that about half of the institutions had anchored AI in an explicit strategy. The main risks named by respondents were data quality, data protection, explainability, security weaknesses, inaccuracy and outsourcing. Most institutions combined internal development with external providers, although smaller institutions often relied entirely on externally developed solutions. A study commissioned by the association Swiss FinTech Innovations and published by the Swiss Bankers Association at the end of 2025 found that 40 per cent of AI initiatives in Swiss banks were still at the stage of ideas or feasibility studies, 17 per cent had reached deployment and 11 per cent were being scaled. Among the cases in production, 65 per cent had met or exceeded expectations.²

Use cases in banking

The Swiss Bankers Association has grouped the uses of generative AI in banking into four categories. The first is employee productivity, including the summarising of documents, translation, meeting transcription and internal knowledge retrieval through retrieval-augmented generation. Operational efficiency comes second, with the automation of data entry, support for compliance reporting, risk assessments and loan processing. Customer experience, including chatbots and personalised recommendations, forms the third group, and the fourth covers the development of products and software. The Association identified seven principal risks: misinformation and deepfakes, hallucinations, infringement of intellectual property, bias, cyber threats such as prompt injection, third-party risks and model drift. On banking secrecy it recommended that client identifying data should not be used by employees in prompts or for training models, and that external providers be engaged only under special contractual and technical safeguards.³

The largest Zurich bank illustrates the scale that adoption has reached. In October 2025 UBS announced that Daniele Magazzeni, previously Chief Analytics Officer at J.P. Morgan for EMEA and a former academic at King’s College London, would become its first Chief Artificial Intelligence Officer from 1 January 2026, based in London. At the time UBS reported more than 300 AI use cases in production. It was rolling out Microsoft 365 Copilot and its own AI assistant “Red” to all employees, and it had launched transformation programmes under the internal label “Big Rocks”.⁴

In compliance, the uses that matter most for a financial centre with Zurich’s reputation are client onboarding, know-your-customer reviews and transaction monitoring. AI tools can extract information from identity documents, corporate registers and source-of-wealth narratives, compare them with sanctions and adverse-media databases, and prepare a draft risk assessment for review by a compliance officer. In transaction monitoring, machine learning models can reduce the high number of false positives produced by traditional rule-based systems and can identify patterns that fixed thresholds miss. FINMA has reported that in its supervision it has examined initial AI applications in asset management, transaction monitoring and liquidity management.⁵ The legal duties themselves do not change. Under the Anti-Money Laundering Act the institution remains responsible for identifying the beneficial owner and clarifying unusual transactions, whatever tools it uses, as explained in chapter I.III.

Advisory services are a further field. Language models can prepare meeting notes, draft suitability documentation under the Financial Services Act and generate first drafts of client correspondence in several languages. They can also summarise research reports for relationship managers. In practice, institutions generally keep a human adviser between the model and the client, both for reasons of liability and because the Financial Services Act ties suitability and appropriateness duties to the financial service provider. Fully automated advice exists in the form of robo-advisers, but the combination of complex cross-border wealth and personal relationships that characterises Zurich private banking makes human-supervised AI the dominant model.

Use cases in insurance

Zurich is also the home of Zurich Insurance Group and Swiss Re, two of the world’s largest insurance and reinsurance groups, whose use of AI is as extensive as that of the banks. Zurich Insurance describes applications across claims and underwriting. In claims, systems automate the intake of notifications and, in some markets, settle simple health claims within minutes. In underwriting, AI analyses aerial imagery to score roofs and detect damage in property business, reads multinational insurance programmes across languages and answers underwriting questions through an internal chatbot used in 29 business units. The group reports that a workbench for commercial underwriting in the United Kingdom reduced processing times by 78 per cent. Its internal challenge for agentic AI in 2025 produced 218 prototypes, of which five moved into production.⁶

For insurers, the regulatory sensitivities differ from those of banks. Risk-based pricing with machine learning raises questions of fairness and discrimination, particularly in health and life insurance. The use of large volumes of personal data in underwriting engages the data protection rules on profiling and automated decisions. These questions are addressed below in the context of FINMA’s expectations and the EU AI Act, whose list of high-risk systems includes AI used for risk assessment and pricing in life and health insurance.

Notes

1. FINMA, “FINMA survey: artificial intelligence gaining traction at Swiss financial institutions”, 24 April 2025, https://www.finma.ch/en/news/2025/04/20250424-mm-umfrage-ki/. ↩
2. Swiss Bankers Association, Stefan Neumann and Patrik Schmid, “From pilots to production: Scaling AI in Swiss finance”, 19 December 2025, https://www.swissbanking.ch/en/media-politics/opinions/from-pilots-to-production-scaling-ai-in-swiss-finance. ↩
3. Swiss Bankers Association, Generative AI in Banking – A Comprehensive Overview, April 2025, https://www.swissbanking.ch/_Resources/Persistent/2/0/3/b/203b937e175f25819ea271c883a095fe1dfa1ee0/SBA_Generative-AI-in-Banking_EN.pdf. ↩
4. UBS, “UBS appoints Chief AI Officer to drive its AI strategy”, 16 October 2025, https://www.ubs.com/global/en/media/display-page-ndp/en-20251016-ai-strategy.html. ↩
5. FINMA, “Künstliche Intelligenz” (dossier Fintech), https://www.finma.ch/en/documentation/dossier/dossier-fintech/kuenstliche-intelligenz/ (accessed September 2026). ↩
6. Zurich Insurance Group, “AI at Zurich”, https://www.zurich.com/about-us/ai-at-zurich (accessed September 2026). ↩

For general information only. This article does not constitute legal or tax advice, does not replace advice on an individual case and creates no lawyer-client relationship. It reflects the legal position as of September 2025.