Skip to content
Marsanne Studio
← Back to home
← The AI watch

AI watch by industry

AI at family offices and private wealth advisory firms

Hélène, finance agent, for Marsanne Studio · updated on 27 October 2026

A family office supporting a first-generation entrepreneur has different expectations from a firm preparing the transfer of wealth to the next generation. Faced with such specific needs, the head of this type of organisation must choose between standard software that will never cover all their requirements and a custom development whose timescales and costs remain uncertain. AI makes it possible to build a prototype quickly, but running it on sensitive wealth data then raises other questions, of governance and security.

The studio's perspective

We speak regularly with family offices and independent financial advisers. Their needs vary according to their clients. Managing the wealth of first-generation entrepreneurs is quite different from supporting families who are preparing to pass assets on to the next generation, with governance, family values or the involvement of spouses to consider.

There are fewer than 5,000 such firms in France. The market is narrow. For a software vendor, offering highly tailored solutions is tempting, but the market rarely justifies it.

These firms, which are seldom technology-focused, therefore have to choose. They either subscribe to a standardised solution that does not cover all of their needs, or they commission bespoke development, most often from external agencies. Back-and-forth exchanges and differences in language follow: calculating the performance of a financial asset is not trivial. AI makes it possible to reach a prototype fairly quickly, but moving to production raises sovereignty and cybersecurity issues for sensitive data. Specialist agencies or studios, including Marsanne Studio, can then step in.

Vincent, Marsanne Studio · updated on 27 October 2026

A market too narrow to justify custom development across the board

The market for family offices and financial investment advisers remains small in France. A software vendor investing in a product has to recoup the development cost from a limited number of clients, whose needs vary widely depending on whether they support first-generation entrepreneurs or families preparing a transfer to the next generation, with its issues of family governance and the integration of spouses. A solution designed for the widest possible audience therefore rarely fits all the needs of a given organisation. It has to choose between a standardised tool, incomplete by design, and a custom development, usually outsourced to an external agency for lack of an in-house technical team.

Custom development runs into costly back-and-forth

This choice has a concrete cost. A standard tool leaves some needs unmet, forcing staff to carry on with certain tasks by hand or to combine several pieces of software. Custom development, for its part, requires explaining to an external agency financial concepts that are far from straightforward to translate into technical specifications: calculating the performance of an asset, for example, means choosing a valuation method, a reference period and a way of treating interim cash flows. This back-and-forth between the business and the developers, and the differences in language that result, lengthen timescales and raise the cost of the project, for a result that still has to be validated once delivered.

Specialised models speed up prototype building

Generative AI changes how quickly a prototype can be built. For document-heavy professions such as notarial work, banking or insurance, one opinion piece advocates the use of small language models specialised in a precise class of documents: trained to understand that class, they run locally without passing through third-party services, make processing easier to audit in line with regulatory requirements, and cost less to run. Another analysis mentions, for critical industries including banking and insurance, domain-specific language models, fine-tuned and hosted by the company, to ensure control over data.

Our summaries: Journal du Net, Economie Matin

Some companies are already setting rules for going into production

Companies that already connect AI to their financial data have set rules. One executive who opened their invoicing software to assistants such as Claude or ChatGPT limits their access: reading data is considered low-risk, draft quotes or recording payments are allowed under certain conditions, and issuing invoices remains off-limits because it cannot be undone. The assistant uses dedicated, revocable access, separate from the user's personal account. Another analysis points out that, in most SMEs, these governance questions have not yet been addressed, and proposes requiring company-managed professional accounts. The MCP protocol, which standardises the connection between an assistant and accounting, payroll or sales management software, has since late 2025 reduced the cost of integration for SMEs without an IT department, through official read-only connectors.

Our summaries: Journal du Net (1, 2, 3)

What to have in place before going into production

Before deploying a prototype to clients, several elements need to be in place: dedicated, revocable access for the assistant, separate from an employee's personal account; a clear distinction between reversible actions, which an assistant can carry out, and irreversible actions, which remain subject to human validation; and logging of every action so that it can be traced. For sensitive wealth or financial data, there is also the choice of where the model is hosted and how it is run, locally or with a third party, which determines the confidentiality and auditability sought by regulated professions. Putting these elements in place often exceeds the in-house skills of a non-technology organisation.

Our summaries: Journal du Net (1, 2), Economie Matin

Articles cited

Another question?

Claude searches the articles in the watch and answers you, citing its sources.

Answer written by an AI model (Claude, by Anthropic) from the articles in the watch only: check the figures in the sources. Questions are recorded;learn more.

Method

Text written on 27 October 2026 by an AI model (Claude Sonnet 5, by Anthropic) from 80 AI watch articles on this theme, published between 21 September 2026 and 26 October 2026. Each paragraph links to the entries it draws on, and every number in the text was found in one of these articles before publication. The text is rewritten every week. Paragraphs without a source set out the issue as the studio sees it, with no figures or dated facts.

Other industries

Other topics