Only 11 percent of AI applications at wealth managers run productively. Why?

Eleven percent. That is the share of AI applications at independent wealth managers in Switzerland that run productively. 66 percent remain experiments with public tools, 23 percent make it to a pilot and no further. That is what the study «AI Use Cases and Crypto Assets in External Asset Management» by Lucerne University of Applied Sciences and Arts (HSLU) and Finup shows, published in July 2026 (to the study). My thesis: the technology is not the problem. What is missing is a data foundation, accountability and a measurable goal.

What exactly does the study say?

HSLU and the Zurich-based WealthTech company Finup surveyed more than 40 independent wealth managers in a structured format between April and June 2026. The result in three sentences: 49 percent of firms already use AI in selected processes. Yet of the 65 reported applications, only 11 percent are productive implementations. 85 percent name data protection as the biggest hurdle, and only 27 percent consider their own data ready for AI.

The distribution at a glance:

• 66 percent experiments: Use of public tools, with no connection to the firm's own processes and data.
• 23 percent pilots: First standalone applications, not yet in regular operation.
• 11 percent productive: AI works reliably in daily business, with a clear process and clear ownership.

Of 65 reported AI applications, only 11 percent run productively.

Why do so many projects stall? Four reasons from practice

First: the data is not ready. Only 27 percent of respondents consider their data fit for AI. That matches what I see in first conversations: client data is spread across banking portals, spreadsheets and inboxes. An agent can only work as well as the data allows. Firms that clean up first win twice: the data becomes usable, and the firm learns what it actually has.

Second: data protection becomes a stop sign. 85 percent name data protection as the biggest hurdle. The question «where does my data live?» is legitimate; in practice it comes up first, time and again. But it is solvable. Protected Swiss cloud environments, hosted for example on Microsoft Azure in the Zurich North region, keep client data in the country and separated from public models. Firms that treat the question as an endpoint instead of a requirement stay stuck in experiments.

Third: nobody is accountable. Many pilots run on the side, carried by one person's curiosity. As soon as that person is absent or daily business calls, the pilot falls asleep. Productive applications have a named owner who looks at the numbers every week.

Fourth: the pilot has no measurable goal. «Let's try this out» is not a target. Without a number, such as minutes per protocol or reports per month, there is no reason to continue after eight weeks and no reason to stop. So the pilot keeps running without ever becoming productive.

What do the firms behind the productive 11 percent do differently? My assessment

What follows is my assessment from projects, not a statement of the study. The firms that go productive do three things differently. They start with a single, narrowly bounded process instead of a broad programme. They deliberately choose a process with low data sensitivity, such as protocols or research, rather than the most complex case in the firm. And they run the agent in assisted mode: the machine creates drafts, the human approves, every action is logged. That is how trust is built through measured quality, not promises.

From my practice: From my projects with additiv I know what makes the step from pilot to production at large institutions. It is rarely the bigger model. It is unambiguous process ownership and operations planned like any other production system: with monitoring, with approvals, with clear rules for failure cases. What is a programme there, a small firm can adopt as a mindset.

Three questions to check yourself

1. Can you name the process of your current or planned AI pilot in one sentence?
2. Do you know how many hours it is supposed to save you per month?
3. Is one person accountable for it by name?

If you answer no to any of these questions, you have probably found the reason your project belongs to the 89 percent that do not run productively.

Conclusion: the number is an invitation

The 11 percent are not bad news but a map. They show that experimenting is widespread across the industry and that the bottleneck lies elsewhere: in the data foundation, in accountability and in the goal. Firms that start now with a well-chosen pilot will be among the productive ones in a year, while the majority is still experimenting. Which use cases are best suited to start with, and what the first 30 days look like, is in the pillar article AI for independent wealth managers: 7 use cases that already work today. For a conversation about your situation, reach us via AI for wealth managers.

FAQ

How representative is the study?

More than 40 independent wealth managers were surveyed in a structured format between April and June 2026. It is one of the first structured surveys of its kind for Switzerland and is deliberately exploratory. The sample is small enough not to overinterpret individual percentage points, and large enough to take the order of magnitude seriously.

What does «productive» mean in the study?

The study distinguishes experiments with public tools, pilot projects and productive implementations. Productive means the application is embedded in regular operations and runs reliably, not just in a trial.

Where can I find the study?

The study «AI Use Cases and Crypto Assets in External Asset Management» by HSLU and Finup (July 2026) is linked via Finup's research page.