Artificial intelligence in wealth management: what it can and cannot do

The common thread is trust under uncertainty. Clients are not looking for louder technology but for a calmer overview of a financial life that has grown more complex. Where AI delivers that, it is welcome. Where it poses as an oracle, it meets a limit that is not technical but human.

The AI divide runs along wealth

Perhaps the most telling figure in the study is not how many clients reject AI, but how unevenly openness is distributed. Those who manage more wealth, and therefore more complexity, are markedly more receptive.

The doubling among wealthier clients is no accident. Anyone running a branching portfolio across several accounts, regions and asset classes feels the value of a tool that draws those parts together and explains them in seconds. The real demand, then, is not “more automation” but “more clarity”. This is exactly where artificial intelligence is strong, and exactly where it is harmless, because it decides nothing.

Where artificial intelligence really helps

The most durable value of AI in wealth management lies not where the imagination runs largest, but where the quiet work sits. In its 2026 trends, Oliver Wyman estimates that advisers today spend only about a quarter of their time on value-adding, client-facing work. The rest drains away into preparation, documentation and the hunt for figures that already exist somewhere. AI assistants, the firm argues, can roughly double an adviser's capacity without lowering quality, because they take over that routine.

Deloitte puts numbers beside this in its 2026 prediction. Depending on how far adoption has matured, it estimates an adviser productivity uplift of around 32 percent in early stages, over 57 percent in more advanced ones, and up to 103 percent in fully AI-native models. The lever is explicitly the relief from recurring background work, not the replacement of judgement. These figures come largely from the US market and are projections, not a measured present; what transfers is the pattern, not the exact number.

For the private investor running their own portfolio, the same holds in miniature. Nobody has a personal adviser reviewing the portfolio every week. That is precisely the gap a tool fills when it explains, summarises and puts things in context: not advice, but a far better overview than a spreadsheet alone.

Where it is overrated

Honesty belongs in the picture. The same Deloitte analysis notes that while around 73 percent of advisory firms use AI in some form, only about 6 percent use agentic, independently acting tools and only around 5 percent have cross-system integration. Between announcement and practice lies a wide gap.

The reason is not a lack of ambition but a hard principle: language models are trained to sound plausible, not to compute. A freely worded number can be convincing and still wrong. In a financial product that is no trifle; it is the difference between a useful tool and a dangerous one. So the decisive design question is not how eloquent an AI sounds, but where every number it utters comes from, and whether a human decides in the end.

How Investboard uses AI

Investboard answers that question with a clean separation. The numbers are produced deterministically, that is, by traceable calculation from your own data. The artificial intelligence does not compute; it interprets. It sees only the individual user's portfolio, runs in the EU on AWS Bedrock with a model from Anthropic (Claude), and no user data flows into any training.

The list follows exactly the shifts McKinsey describes. The wish for more frequent, more personal contact becomes the weekly briefing that arrives every week without an appointment and without a surcharge. The need for more clarity becomes the widget explanation that translates a single metric into plain language. And the openness of wealthier, more complex clients meets an assistant that draws a branching portfolio together in a single question instead of scattering it across tabs.

The limit is important, and it runs the same everywhere. The assistant can read but not act. The nine tools available to it retrieve data only; none can place an order. Investboard is a platform for strategy and analysis, not execution, and the AI changes nothing about that.

Explanation, not decision

What separates a financial product from a chat window is the guardrails behind it. At Investboard they are not an accessory but the actual substance of how AI is used.

Every answer passes a check in both directions. On the way in, attempts at manipulation and personal data such as an IBAN or tax ID are detected and filtered out. On the way out, the system reconciles every number in the answer against the underlying data and flags anything that deviates by more than one percent: the structural bar against invented figures. Language that sounds like a concrete buy or sell recommendation is detected and marked with the clear note that this is not investment advice, as Article 50 of the EU AI Act also requires by way of disclosure.

There is also a self-restraint that is easy to miss. During onboarding, when a user sets their investment strategy, the AI tutor is deliberately explanatory only: it may clarify terms but recommend no template and name no percentage. That decision stays with the human. This is the heart of the stance: every interpretation the AI offers needs an active decision by the user to take effect. The machine never proposes what to do; it makes visible what is.