Agentic AI in Wealth Management: What the Technology Can Do and What It Must Not Decide

Agentic AI refers to AI systems that plan and execute multi-step tasks autonomously, without requiring human input at every stage. In wealth management, this means the system researches, filters and summarises. What it does not do, and is not permitted to do, is make substantive decisions on investment strategy or mandate management. This distinction is not merely technical, it is regulatory.

The term has circulated through industry discussion for months, and more than a few providers are already marketing their products as complete operational infrastructure for financial advisors built on agentic systems. For Relationship Managers who work daily with real mandates and real liability, a closer look is worthwhile: what can this technology actually deliver, which decisions must remain with the human, and which questions should be answered before any deployment?

Key Points at a Glance

• Definition: Agentic AI = AI systems that break down objectives into sub-steps and execute them autonomously, sometimes drawing on external data sources or interfaces.
• Market dynamics: Several specialist providers have been positioning agentic systems specifically for wealth management and financial advisory since 2024/2025 (source: various industry reports, including EY, 2025 Global Wealth and Asset Management Outlook).
• Regulatory framework in Switzerland: Investment decisions must be attributable to a supervised and responsible person (FinSA, FINMA regulation for wealth managers).
• Time potential: According to McKinsey estimates, Relationship Managers spend around 60 to 70 per cent of their time on non-revenue-generating, often administrative activities; generative AI could shift a portion of that time (approximately 20 to 30 per cent) towards value-adding client work (source: McKinsey, insights on US Wealth Management, 2024).
• Core insight: Technology reduces the burden on the process. It does not replace judgement.

What Does “Agentic” Actually Mean?

Agentic AI differs from conventional automation in one decisive respect: it can interpret unstructured tasks, plan sub-steps on its own, and find alternative routes when it encounters obstacles. A conventional automation system follows fixed rules: if condition X is met, execute step Y. An agentic system receives an objective, such as “produce a summary of the most relevant market developments of the past week for this client type”, and independently plans which sources to use, in which order to proceed, and how to structure the output.

This flexibility is the value it adds. It is also the reason why agentic systems are less predictable than simple rule-based automation. And that is precisely why clear boundaries, audit trails, and human control points are not optional extras but a fundamental requirement for responsible deployment.

Where Does Agentic AI Deliver Concrete Time Savings for Relationship Managers?

Existing practical examples and pilot projects from the financial sector show that agentic systems can deliver genuine time savings in clearly defined process steps. The most common areas of application:

Information preparation: The system scans defined sources, identifies topic-relevant updates, and produces a structured summary for the Relationship Manager. Instead of 45 minutes of research in the morning, the RM receives a prepared reading list with relevance assessments.

Document processing: Incoming documents, such as annual accounts, custody statements, or legal documents, are automatically captured, categorised, and condensed to the key figures. The RM sees what matters, not the raw text.

Meeting preparation: Based on available client data and past interactions, the system generates a structured meeting proposal covering open points, portfolio developments, and potential topics. The RM decides which points to take up.

Compliance pre-checks: The system verifies whether required documents for a planned transaction or new mandate are in place, and flags missing documents before the human review.

In all these cases, the core task stays with the Relationship Manager: the substantive assessment, the client conversation, the decision.

What Agentic AI Must Not Decide

Clarity matters more here than enthusiasm. Under FinSA and the FINMA supervisory rules for wealth managers, responsibility for investment decisions, due diligence, and mandate transparency lies with the supervised institution and its staff. No AI system is recognised under regulation as a decision-making subject.

In practice, this means:

• Investment decisions require a responsible person who was demonstrably involved.
• Mandate transparency towards the client must be humanly accountable, even when systems have contributed.
• Documentation of decision bases may be AI-supported, but completeness and accuracy remain a human responsibility.
• Error control is mandatory: agentic systems produce errors, so-called hallucinations, factually incorrect outputs with no correction mechanism. Those who pass on AI output without checking it bear responsibility for the consequences.

This regulatory framework is not a brake, but a sensible safeguard: for the client, for the Relationship Manager, and for trust in the industry.

Which Questions Should Be Answered Before Deployment?

For Relationship Managers and leaders in wealth management firms considering the use of agentic systems, the following questions are central:

Data protection and data processing: Where is client data processed? Does it leave the defined regulatory perimeter? Which third-party providers are involved, and are they qualified under revFADP and GDPR?

Traceability: Does the system provide an audit trail for its outputs? Can it subsequently be established which sources a summary was based on?

Quality control: Is there a documented approval process before AI output reaches client contact? Who is responsible?

System boundaries: Is it clearly defined which tasks the system takes on and which it does not? Are there technical or procedural safeguards preventing the system from advancing into areas that have not been approved?

The answers to these questions should be documented before any system goes into production.

Frequently Asked Questions on Agentic AI in Wealth Management

What is Agentic AI in wealth management?

Agentic AI refers to AI systems that plan and execute multi-step tasks autonomously, without requiring human input at every stage. In wealth management, this means the system researches, filters and summarises information, or carries out defined process steps. The substantive decision over investment strategy and mandate remains with the responsible Relationship Manager.

Is Agentic AI permitted to make investment decisions?

No. Under FinSA and the FINMA supervisory rules for wealth managers, responsibility for investment decisions and due diligence lies with the supervised institution and its staff. AI can prepare information and automate process steps, but it is not a recognised under regulation as a decision-making subject. Every investment decision must be attributable to a responsible person.

Where does Agentic AI deliver concrete time savings for Relationship Managers?

Typical use cases include the automated preparation of client documents, pre-selection of relevant market updates, summarising documents, and drafting meeting preparation based on available data points. According to McKinsey estimates, Relationship Managers spend around 60 to 70 per cent of their time on non-revenue-generating, often administrative activities; generative AI could shift a portion of that time towards value-adding client work.

How does Agentic AI differ from conventional automation?

Conventional automation follows fixed rules (if X, then Y). Agentic AI can interpret unstructured tasks, plan sub-steps on its own, and find alternative routes when it encounters obstacles. This makes it more flexible, but also less predictable. That is precisely why every productive deployment requires clear boundaries, audit trails, and human oversight at defined control points.

What risks does the use of Agentic AI carry in a financial context?

The greatest risks are hallucinations (factually incorrect outputs), lack of traceability in decision steps, data protection questions when processing client data, and uncontrolled transfer of information to external systems. Careful selection of systems, a documented approval process, and regular quality checks are therefore essential.

---

This article does not constitute investment or legal advice and is not a substitute for individual legal or regulatory review.