If a model gives the advice, the bank still owns every word of it
HSBC is reported to be cutting up to 70 per cent of its UK financial advisers as artificial intelligence takes over. The saving is only real if the duties attached to advice quietly fall away with the advisers.
HSBC is one of Britain's biggest banks. Part of it employs financial advisers — people licensed to tell customers what to do with their savings and pensions, a job that comes with legal duties and a right of complaint if the advice was bad. The Financial Times reports the bank plans to cut up to 70 per cent of those UK advisers as artificial intelligence takes over more of the work.
If the software gives personal recommendations, every duty a human adviser carries attaches to it and the savings are smaller than they look. If it does not, customers are losing their right to complain — and nobody has said which it is.
A bad adviser damages one client book. A bad model damages all of them at once
That is the whole of the argument, and it is why a plan to replace most of a bank’s advisers with software is a regulatory event rather than a human resources one.
The Financial Times, reported by Finextra, says HSBC intends to cut up to 70 per cent of its UK financial advisers as part of changes to its wealth management business driven by rising use of artificial intelligence, and one person describes the cuts as deep, wide and brutal. That is a press report, not a filing: the bank has published no terms, no description of what replaces the advisers, and no statement about the regulatory status of whatever it is. The absence is the interesting part.
Advice is a legal category, not a job title
In Britain, telling an identified customer what to do with their own money — this pension, this amount, this product — is a regulated activity. It triggers a duty to establish that the recommendation suits that person, the Financial Conduct Authority’s Consumer Duty to deliver good outcomes, a paper trail, and two backstops: the Financial Ombudsman Service, which can order compensation when a recommendation was unsuitable, and the Financial Services Compensation Scheme, which pays when the firm that gave it has gone.
Generic information carries none of that. The gap between the two is not a technicality. It is the difference between a customer with a claim and a customer with a grievance.
So when a firm removes most of the people who do the regulated thing, there are only two possibilities. The first is that the machine does the regulated thing instead. Nothing in British regulation says a personal recommendation must be made by a human; it says the firm is responsible for it. On that reading HSBC has not escaped a single duty. It has concentrated them into one system, which it must now supervise, test, document and stand behind — and the cost of doing that properly is not nothing.
The second possibility is that what replaces the advisers stops short of a recommendation: it narrows, explains, educates, nudges, and leaves the customer to press the button. That is cheaper precisely because the duties do not attach. It is also a transfer. The customer who would previously have had a suitability assessment and a route to the ombudsman now has an interface and a disclaimer.
The file is the product
Every large remediation exercise British retail finance has run has turned on records: what was the customer told, when, and on what basis. A human adviser, whatever else is wrong with the model, generates that record as a by-product of doing the job. A generative system does not, unless it is built to.
Reconstructing a recommendation made two years ago means knowing which version of the model answered, what it was instructed to do, what the customer had entered, and what exact words came back. If any of that is overwritten on the next deployment, the firm cannot prove the recommendation was sound and the customer cannot prove it was not. Redress then becomes a negotiation about averages rather than an assessment of cases, which is worse for everyone except the lawyers.
The case for doing it anyway
Here is the strongest version of the other side, and it is strong. Most British adults get no regulated advice at all, because at the price a qualified human must charge it is uneconomic to serve a modest pension pot. Advisers are scarce, ageing and expensive. A system that gives decent, consistent, documented guidance to a million people may do far more aggregate good than excellent advice delivered to forty thousand. Consistency is itself a virtue: software does not have a bad Friday, does not favour the products with the better conference, does not forget to ask about debts.
All true. But none of it argues for ambiguity about which category the service sits in. If the answer is that HSBC is extending genuine advice, with genuine liability, to people who could not previously afford it, that is a public good and the bank should say so loudly, because it is a selling point. The fact that firms in this position tend not to say so loudly is the tell.
What would settle it
One sentence from the bank, and one paragraph from the regulator. Does the output constitute a personal recommendation? Is a customer who follows it and is harmed entitled to take the case to the ombudsman? And can the bank, on demand, produce what the system said and why, as it stood on the day it said it?
Those are answerable now, before anyone is made redundant. They will be very much harder to answer in 2031, when the first cohort of customers who followed an automated recommendation discover it did not suit them, and the advisers who might have remembered the conversation left five years earlier.
The Financial Conduct Authority should require any firm replacing regulated advisers with automated systems to state, before launch, whether the output is a personal recommendation, and to commit in writing to retaining model version, inputs and generated text for the full complaints window. Not as guidance — as a condition of the change.
And HSBC should answer the question itself, in plain English, in the same announcement that confirms the job cuts: when the machine tells a customer what to do with their pension, can that customer take it to the ombudsman? If yes, say so. If no, say that too, and let people decide whether they still want the service.
What would change this view
I am wrong if HSBC states, or the Financial Conduct Authority confirms, that the output of its automated wealth service is a personal recommendation within the ombudsman's jurisdiction and the compensation scheme's cover, and that the bank retains model version, inputs and generated text sufficient to reconstruct any given recommendation years later. Put that in writing and the cuts become an ordinary staffing decision I have no standing to object to.
If instead the service is launched without that statement, and the first published description of it uses words like guidance, support or information, the perimeter shift is the product and the job losses are the by-product.