No, not unless the firm has built the full advice process around it. An AI agent can give financial guidance: explain products, set out options and their costs, and help customers understand their own accounts. A personal recommendation is regulated advice in most markets, and it needs the firm's permissions, a suitability assessment and a named person accountable for it.
What is the difference between guidance and advice?
Guidance helps a customer understand their options. "A fixed-rate savings account pays a set rate but locks your money for the term; an easy-access account pays less but you can withdraw at any time" is guidance. So is showing a customer how much interest each option would pay on their balance.
Advice tells a customer what they should do, based on their circumstances. "You should put your 20,000 into the two-year fixed account" is advice. It assumes the agent knows enough about the customer's needs, other money, plans and attitude to risk to make that call, and it carries responsibility if the call is wrong. The line is explored in FCA and AI agents giving guidance vs advice.
The same line runs through other professions; compare legal advice. The test is whether a reasonable customer would act on the answer as a recommendation.
Why does advice need a person?
Advice is acted on, and the effects are hard to reverse. A customer who moves a pension, buys an investment or takes out a product on a recommendation may lose money or flexibility that cannot be recovered. If the advice was unsuitable, the remedy is a complaint and compensation, months or years later.
Regulators treat advice accordingly. In most markets, personal recommendations on investments and similar products require authorisation and a suitability process. In the UK, the Senior Managers and Certification Regime makes named individuals accountable, and the FCA Consumer Duty expects good outcomes for retail customers. A firm that lets an agent drift into recommendations without that structure takes on risk it has not managed.
What may the agent do, and what stays with a person?
| Task | The agent | Stays with a person |
|---|---|---|
| Explaining products | Explains features, costs and risks from approved material | Nothing routine |
| Comparing the firm's products | Sets out options side by side, without ranking them for the customer | A recommendation |
| Working with the customer's own figures | Shows what each option would mean for their balance | Telling them which to choose |
| Spotting financial difficulty | Recognises signs and offers help routes | Support decisions |
| Answering "what should I do?" | Explains options and offers an adviser | Advice |
| Advised sales | Gathers the fact-find for an adviser | Suitability and the recommendation |
| Execution-only requests | Carries out the customer's own clear instruction, with warnings | Cases where the customer seems to need advice |
Signs of financial difficulty deserve special attention, because a customer in difficulty needs support, not a product. See how to identify customers in financial difficulty with an AI agent.
A worked example
Take a bank's app agent. A customer has 15,000 in a current account and asks: "What should I do with this money?"
The agent does not recommend anything. It explains the savings accounts the bank offers, with their rates, access rules and terms, and shows how much interest each would pay on 15,000 over a year. It mentions that the bank also offers investment products, which involve risk to capital, and that an adviser can discuss whether any of them suit the customer.
The customer then says they need the money for a house deposit in about eight months. The agent notes that this affects which options might suit them, without saying which. It offers an appointment with an adviser and records the conversation, including the time frame the customer mentioned, so the adviser starts with it.
What controls make it safe?
- Approved product material only. The agent explains from the firm's own approved descriptions.
- A refusal pattern for "should I". Questions asking for a recommendation get options and a route to an adviser. Test this on real customer questions before go-live.
- No ranking for the customer. The agent may sort options by a neutral feature the customer chose, such as rate or access, but does not pick one.
- Limited system access. The agent reads the accounts it needs, through defined actions; see giving an AI agent safe access to your core banking system.
- A full record. Each conversation is filed, so the firm can show that the agent gave guidance, not advice.
The wider limits are set out in AI agents in banking: what they may and may not do.
Questions people ask
Can an agent give regulated advice if the firm is authorised?
A firm could build an advice process that uses an agent, but it would need to meet every requirement that applies to advice, including suitability, records and accountability. That is a large design and governance task, not a configuration change.
Can an agent help an adviser prepare advice?
Yes. It can gather the fact-find, summarise the customer's position and draft a suitability report for the adviser to review. The adviser makes the recommendation and signs it.
What if the customer insists on a recommendation?
The agent explains that it cannot give one and offers an adviser. It should not give a hint or a "most customers choose" answer as a workaround.