Back to articles
Blog Apana
AI in Wealth Management: 72% of Advisors Use It, but Does It Improve Advice?

AI in Wealth Management: 72% of Advisors Use It, but Does It Improve Advice?

72% of French wealth advisors (CGPs) use AI in 2026, up from 23% in 2024, according to BNP Paribas Cardif's annual survey. Adoption is a given.

Its uses remain shallow. The available data shows AI that summarizes, writes and searches. Building a recommendation comes a distant last.

On October 12, France's two financial supervisors, the AMF and the ACPR, published the findings of around sixty inspections. They do not fault the firms they inspected for poor writing. They fault them for standardized recommendations, made without knowing their clients well enough.

That is the gap. AI saves time on admin work, while the regulatory pressure is on the advice.

In short. According to BNP Paribas Cardif, 72% of French wealth advisors use AI in 2026. Documented uses, mostly from the US, focus on handling information, and recommendations remain a minority use. The AMF and ACPR found a lack of personalized advice at some of the firms they inspected. The task now is to use AI for the advice itself, with verified data, traceable calculations and advisor sign-off.

Contents

Has AI really taken hold in advisory firms?

Yes. Reported use has tripled in two years.


Year

Share of French advisors using AI

2024

23%

2025

56%

2026

72%

Source: BNP Paribas Cardif surveys conducted with Kantar, 2025 and 2026 editions.

The figure measures reach, not intensity. In 2025, the 56% combined regular and occasional use, and regular use alone stood at 27%. The 2026 edition does not give this breakdown.

Clients are changing too. 67% of advisors say AI has changed how their clients behave, and 32% find them much better informed. A client who has already asked an AI assistant expects more than general information.

What do advisors actually use AI for?

Mostly to handle information. Recommending remains a minority use.

No French statistic measures these uses. The most granular data comes from the US. In June 2025, ISS Market Intelligence surveyed 466 advisors, 48% of whom were already using AI.


AI use

Share of advisors using AI

Summarizing information

58%

Generating written content

56%

General research and brainstorming

55%

Investment research

42%

Processing data

24%

Performance analysis

23%

Investment recommendation or portfolio construction

19%

Source: ISS Market Intelligence, Advisor Pulse, June 2025. Base: US advisors using AI.

The top three are all about handling information. Portfolio construction peaks at 19%, and it covers only part of French wealth advice, which also includes tax and estate planning.

In France, intentions point the same way. In 2025, 79% of advisors expected their AI use to stay focused on admin tasks, and 67% on regulatory monitoring. These are projections, not measurements.

What did the AMF and ACPR find at the firms they inspected?

Advice that was not personalized enough.

On October 12, 2026, the joint AMF-ACPR unit published a review of around sixty inspections carried out from 2020 to 2024. The AMF is France's financial markets authority, and the ACPR supervises banks and insurers.

The inspections covered the sale of financial products and life insurance contracts (assurance-vie) to retail clients.

They targeted firms or topics seen as high-risk. The findings therefore do not describe the market as a whole, and they do not specifically target CGPs.

Four findings bear directly on advice:

  • standardized recommendations, due to insufficient collection of client information;

  • savings capacity often taken from the client's own statement alone;

  • vague or missing justifications for the advice given;

  • in life insurance, tax, estate transfer or the beneficiary clause not always taken into account.

These shortcomings share one trait. None is fixed by writing faster. All of them require knowing the client better, analyzing more and recording the reasoning.

Why is personalized advice so hard to produce?

Because every client requires the same full chain of work, and that chain does not get shorter.

Collect the information. Check it is complete. Understand the goals. Assess the tax and legal effects. Compare scenarios. Check suitability. Write it up. Record the reasoning.

With a few dozen clients, the chain holds. With several hundred, the temptation to use a template grows.

The workload does not end at signing. Since October 24, 2024, an ongoing duty to advise has applied to life insurance in France, with regular updates of client information.

How can AI make advice more individualized?

By strengthening what the inspected files lacked: knowledge of the client, analysis, and a record of the reasoning.

This is our conviction at Apana. AI improves both productivity and the quality of advice when it serves the whole advisory chain, beyond the admin work.

More complete client knowledge

The information often already exists, scattered across meeting notes, a tax notice and an email thread. AI extracts it and organizes it in the client file. The advisor checks it.

Gaps then show up before the recommendation, rather than during an inspection.

More scenarios compared

Take a typical case: a 58-year-old client preparing for retirement who wants to pass wealth on to his children. Life insurance, a PER retirement plan, real estate, gifts, split ownership… Each option calls for its own calculations.

With calculation engines driven in natural language, the advisor compares more of them and justifies the chosen scenario in writing. That is precisely the traceability the inspected files lacked.

Every figure must remain verifiable. A deterministic engine can be checked, while a language model calculating from memory cannot, as we explained in why advisers shouldn't use ChatGPT to build advice.

More time for the human side

Meeting notes, data entry, document preparation: these tasks can be automated. The time saved goes back to listening, analysis and explaining, which is where personalization happens.

Can AI scale up bad practices?

Yes. Fed an incomplete file, it produces an unsuitable recommendation faster, and makes it look better.

Regulators make a related point: using a data aggregator does not exempt a firm from checking that its data is accurate. The same logic applies to AI.

Three operations are not equal. A language model generates text with ease, with no guarantee of accuracy. A calculation requires a deterministic engine and identified rules. The recommendation engages the advisor's responsibility, and the advisor's alone.

Our article on training, fine-tuning and RAG explains what separates a generated answer from a sourced one.

What infrastructure does AI-assisted advice require?

AI plugged into professional tools, with an advisor who validates every step.

At Apana, four applications follow this logic:

  • Copilot captures and structures client interactions (meetings, documents, emails) into wealth data;

  • Advisor provides deterministic calculation engines, simulators and sourced scenarios;

  • Officer brings together CRM, compliance, account aggregation, document management and e-signature;

  • Finance covers portfolio management, allocation models and performance tracking.

Several of these building blocks are available to AI assistants through the Model Context Protocol (MCP), an open standard published by Anthropic in late 2024.

The advisor asks a question in plain language. The assistant queries the client file, runs the calculation in the right engine and returns a sourced result, which the advisor validates.

How should a firm measure its AI maturity?

By what AI brings to the advice, more than by its adoption rate.

The three publications cannot be compared figure by figure, since their populations, countries and methods differ. Read together, they still reveal a gap: adoption is climbing, uses remain administrative, and regulators expect personalization.

Counting the advisors who use AI says little about what it delivers to clients. Other indicators say more.

Completeness of client files comes first. Then the number of scenarios studied before each recommendation, the quality of the written justification and, at the far end, how quickly a profile is updated.

What if the real measure of AI maturity in advisory firms became its measurable contribution to the quality of advice?

FAQ

How many French wealth advisors use AI in 2026?

72%, according to BNP Paribas Cardif's survey conducted with Kantar and published on September 22, 2026. The figure was 56% in 2025 and 23% in 2024. The 2026 release specifies neither the share of regular use nor the tasks involved in that use.

What did the AMF and ACPR find on personalized advice?

Their review of October 12, 2026, based on around sixty inspections carried out from 2020 to 2024, found standardized recommendations at some firms, incomplete collection of client information and insufficient justification of advice. These targeted inspections do not describe the market as a whole.

Can AI make a wealth recommendation in place of the advisor?

No. AI helps collect client information, run calculations and draft documents, but it does not decide. The recommendation engages the advisor's responsibility: the advisor checks the data, reviews the calculations and validates the final advice, in line with the duty to advise.

Sources

  • BNP Paribas Cardif, "Baromètre du marché des CGP 2026 de BNP Paribas Cardif", press release, September 22, 2026.

  • BNP Paribas Cardif, "Baromètre 2025 du marché des conseillers en gestion de patrimoine (CGP)", press release, September 15, 2025.

  • ISS Market Intelligence, "Nearly 9 in 10 Advisors Plan to Use A.I. Within a Year", press release, July 22, 2025, and PLANADVISER, "Advisers Using AI, by the Numbers", February 2, 2026.

  • AMF-ACPR joint unit, "Commercialisation d'instruments financiers et de contrats d'assurance-vie : points d'attention et rappel des autorités", October 12, 2026.

Want to go further? Why can a language model be confidently wrong? Our article on how an LLM works explains it in a few minutes.

One of your client files, structured by AI and checked by you. In a meeting, we compare several scenarios on a real case, with every calculation sourced.

We use cookies to improve your experience. By continuing, you agree to our cookie policy.