A wealth client opens the app after a rough week in the market and types: "How exposed am I to this morning's tariff news?" A scripted chatbot deflects to a static disclosure page. A relationship manager, if the client can reach one before markets close, needs time to pull the numbers together. Coach mode, the guidance side of Conversational Banking, can answer the exposure question in the moment. What it should never do next is tell that client to sell.
That line, between explaining a position and recommending a trade, is the actual product design problem in wealth. Too cautious, and Coach mode is just a slower way to read a statement. Too permissive, and it's giving unlicensed investment advice through a chat window.
What Coach mode actually does for wealth clients
Conversational Banking runs in two modes. Assist executes a task: the customer states what they need, the system confirms it's possible, and it acts within policy. Coach handles guidance, questions with no single correct action, like "Am I on track with my savings?", where the system has to reason across the customer's actual financial history and answer specifically rather than return a canned line. See Conversational banking in 2026: the complete guide for how the two modes work together, and Digital wealth management platform: how leading banks unify advisors, clients, and operations for the fuller platform Coach mode sits inside, where AI is already expected to work "within guardrails, not outside them."
In wealth, Coach carries more weight than it does in retail. "Am I on track for retirement" and "how exposed am I to this morning's tariff news" both sound like simple questions. Neither has a safe, canned answer, and neither is a transaction. That's exactly the terrain Coach mode has to hold.
What conversational AI can say to a wealth client
Jules Bordat, Backbase's Principal Wealth Expert & GTM Lead, has made the same point on Banking Reinvented and in his own input to a Finextra report on AI in wealth management: high-net-worth clients are comfortable instructing and querying through conversation, for explaining a portfolio, pulling context together, or the kind of check-in that used to require booking an advisor's time. What they actually want to talk about, in his framing, goes well beyond the investment recommendation itself: the family situation, what a geopolitical event means for estate planning, whether it's time to reconsider how assets are structured. Coach mode can hold that conversation. It can explain exposure, summarize a position, surface a pattern in spending or drawdown, and set up the next real conversation with the human relationship manager.
What conversational AI can't say: the regulatory line
The boundary here isn't a UX choice, it's regulatory. In the EU, the AI Act sets a universal floor, transparency, explainability, data rights, that applies across every industry. In the UK, sector regulators like the FCA apply existing principles, suitability, best execution, fiduciary duty, directly, because those obligations already sit with them. Bordat's framing of the right regulatory posture is direct: regulate the outcome, not the mechanism. The suitability of the advice, the explainability of the reasoning, and the person or body accountable for it are what get scrutinized, not which model produced the words.
Practically, that rules out Coach mode ever generating a specific buy, sell, or reallocation recommendation on its own. It can explain what a tariff shock did to a portfolio's exposure. It cannot tell the client to rebalance because of it, that determination carries a suitability and fiduciary obligation attached to a licensed human, and no model output changes who is accountable for it. Voice agents in banking: use cases, governance, and what production looks like covers how a Decision Token draws the same line for any AI-initiated action, not just conversation.
Why the advisor stays the accountable point
Bordat frames the model as a supplier rather than a colleague, a distinction that keeps the advisor, not the AI, accountable for the advice itself. His research points to 93% of advisors wanting the final say on any output that reaches a client, which is less about distrust of the technology and more about where liability actually sits.
The architecture behind that accountability is the same one that makes fiduciary accountability structural rather than a policy promise: a registered identity, a defined permission boundary, and a Decision Token on every action. Why agentic AI stalls without a unified execution layer covers that mechanism in depth, how a Decision Token gives compliance an audit trail instead of a policy statement, in a wealth-specific architecture context. This piece is about what that accountability means inside the client conversation itself; that one is about how it holds up structurally underneath it. Know your agent: a governance framework for AI in banks covers how the same discipline extends across a distributed agent network, and AI agents vs. AI tools for financial advisors: what's the difference? covers where that leaves the advisor's own tooling.
Why the boundary builds trust, not limits it
The trust data cuts against the assumption that clients want the AI to go further. Clients get comfortable with AI that assists their decisions and answers questions in real time. They get nervous the moment it looks like the AI is making decisions about their money. That's the actual design target: reassurance that comes from a visible boundary, not a system pretending it doesn't have one. The same logic holds for advisors and staff. The trust gap closes when AI visibly takes work off their plate, not when it operates unseen in the background.
What to ask before deploying Coach mode in wealth
Does the system draw a hard line between explaining a position and recommending an action, or does that line depend on a prompt going well? Is every Coach mode output logged with the context, the reasoning, and the human decision that followed, so it holds up if a regulator asks later? Can an advisor see, override, or correct what Coach mode told their client before it becomes the record of what was communicated?
None of these questions are about how natural the conversation sounds. That's deliberate. A fluent answer that crosses the suitability line is a worse outcome than a stilted one that doesn't.
Frequently asked questions
Can conversational AI give investment advice to wealth clients?
No. It can explain a portfolio, summarize exposure, and answer questions grounded in the client's actual data. Recommending a specific buy, sell, or reallocation carries a suitability and fiduciary obligation that stays with a licensed human advisor.
What's the difference between Assist mode and Coach mode for wealth clients?
Assist executes a defined task, like checking a balance or moving money. Coach handles open-ended guidance questions, like portfolio exposure or progress toward a goal, where the system has to reason across the client's history rather than return a fixed answer.
Who is liable if Coach mode gives a client inaccurate information?
The advisor and the institution remain accountable, not the model. That's why every Coach mode interaction should be logged with a traceable decision record and why advisors, not the AI, retain final say over anything that reaches a client.
How do regulators expect banks to govern AI in wealth management?
The EU AI Act sets universal requirements around transparency and explainability. UK regulators like the FCA apply existing sector principles, suitability, best execution, fiduciary duty, directly. Both approaches converge on the same standard: regulate the outcome and who's accountable for it, not the underlying model.
