AI consulting for financial services and insurers

AI consulting for financial services and insurers differs from other industries mainly because of the regulatory guardrails: FINMA expects governance, traceability and data protection under the revDSG, regardless of which tool runs in the background. Sensible first use cases often sit in underwriting and customer service, where clearly bounded, recurring tasks tie up a lot of manual time.

Regulatory guardrails: FINMA and the revDSG

Switzerland has no dedicated AI ordinance for the financial sector. FINMA regulates in a technology-neutral way: existing requirements on governance, risk management and outsourcing apply unchanged, even when a task is newly taken over by an AI system. In 2024, FINMA published a supervisory communication that sets out its expectations for supervised institutions using AI, focused on a central inventory of the applications in use, clearly defined responsibilities, testing, ongoing monitoring and staff training. Details are on FINMA's dossier page on artificial intelligence. In practice that means before a use case goes live, it should be clear who inside the business owns it professionally, how the system's mistakes get caught and corrected, and how the outcome can be traced afterwards, including towards customers or the regulator.

Anyone using AI tools from external providers, such as cloud-based language models or outsourced document processing, also falls under Circular 2018/3 on outsourcing for banks and insurers. It requires careful selection of the provider, a contract that secures audit and control rights, and ongoing oversight, not just a one-off check at the start.

On top of that comes the revised Federal Act on Data Protection (revDSG): anyone processing personal data with AI, for example in customer communication or risk assessment, needs a clean legal basis, transparency towards the people affected and, for automated individual decisions with legal effect, particular care. What that means concretely for your website and processes is covered in the article on data protection and AI.

A general AI strategy is not enough on its own once supervised processes are involved. A financial services firm additionally needs an inventory that records, for every application in use, what purpose it serves, what data it processes, who owns it professionally and when it was last reviewed. That sounds like paperwork, but it prevents exactly the scenario regulators worry about most: a system running in production that nobody in the business can actually explain.

Use cases in underwriting and customer service

In underwriting, AI helps most where large volumes of structured and unstructured data, such as applications, claims or policy documents, need to be pulled together and pre-screened before a person decides. Concretely, that often means capturing incoming documents automatically and routing them into the right fields, flagging unusual patterns in claims that deserve closer review, and proposing straightforward, clearly rule-based cases for direct approval. That speeds up the initial review but does not replace the professional judgement needed on more complex or borderline cases.

In customer service, the biggest, fastest-visible benefit is often pre-qualifying enquiries: which enquiry needs which documents, which can be answered directly, and which has to go to a specialist. Another use case is a multilingual first response, a noticeable time saver in a country with several national languages. Summarising long customer histories before an advisory call also saves noticeable time, without the AI taking over the advice itself: the advisor walks in better prepared, instead of spending the first twenty minutes digging through old notes.

A third area that comes up less often but saves a similar amount of time is internal knowledge work: compliance rules, product terms and internal guidance change constantly, and customer-facing staff spend real time hunting for the current version. A well-maintained, searchable knowledge base built on that internal rulebook cuts that search time noticeably and produces more consistent answers for customers too.

The same principle applies across all three areas: AI takes on the groundwork, the decision with legal or financial weight stays with a person. That is not only cleaner from a regulatory point of view, it also builds more internal acceptance for the new tools, because nobody has to fear being replaced rather than supported.

How a first project typically runs

It starts with a short analysis: which processes tie up the most manual time today, which data is already structured, and where is the regulatory risk contained enough for a first test. That produces a tightly bounded pilot, often in customer communication or document review, that delivers first results within a few weeks and is tested with a small, controlled group of users before it extends across the business.

In parallel, the control environment is built: who owns the application, how is it tested, how is it monitored, and how is it documented in the internal inventory. That includes how the AI's mistakes are handled, for example through a clear escalation step back to a specialist, and how results are checked on a regular, spot-check basis. Only once the pilot has shown its value and governance is in place does it extend to further use cases. What that path could look like for your business is best clarified in the no-obligation initial conversation on AI consulting.

One point is often underestimated in planning: training the staff who will work with the system going forward. AI pre-screening is only as good as the trust the specialist behind it places in it. If the output gets accepted blindly, or distrusted on principle and double-checked anyway, the efficiency gain disappears. A short, clear explanation of how the system reaches its suggestions and where its limits lie belongs in the project from the start, not as an afterthought.

If you also want to know how your own business is currently perceived in ChatGPT, Perplexity and other AI search engines, the AI visibility check gives a first overview.