By Lynn Räbsamen, CFA | Advisory Board Member, CFA Institute | Author, Artificial Stupelligence
70 percent.
That is the share of financial advisors HSBC plans to cut from its UK wealth business, according to a Financial Times report on October 7. The FT also reported that about half of specialist roles would go. HSBC’s private bank is not affected.
HSBC has not confirmed the figures. It said it is “continuing to evolve to deliver more digitally enabled products and journeys.”
Two years ago, the FT noted, the same business was hiring hundreds of advisors to double its UK wealth assets to £100 billion.
I am mildly skeptical of the number. Not because AI can’t do real work in finance. It can. But finance has announced the end of a profession several times before, and the profession keeps showing up for work.
Headlines like this treat a headcount figure as a verdict on a technology. It is closer to a forecast. And forecasts about the end of professions have a poor record.
Tellers and ATMs
Start with the cash machine.
US ATMs grew from about 100,000 in 1990 to about 400,000 in 2015. By any sensible forecast, the bank teller should have gone the way of the switchboard operator.
It didn’t. As James Bessen’s research shows, teller numbers held steady and even rose slightly. ATMs made each branch cheaper to run, so banks opened more branches. Each one needed fewer tellers. The total held.
What changed was the work. Tellers handled less cash and solved more customer problems.
The obvious prediction was that machines would replace people. It missed the second-order effect. A cheaper branch meant more branches, and every branch still needed someone to handle what the machine couldn’t.
The job changed shape long before it shrank.
Robo-Advisors
Around 2015, the robo-advisor was going to replace the human one. Algorithms would build the portfolio, rebalance it and charge a fraction of the fee.
A decade later, the big banks were the ones leaving. JPMorgan announced the closure of its robo in December 2023. Goldman sold its Marcus Invest accounts to Betterment in April 2024. UBS wound its robo down in 2025, and U.S. Bank shut Automated Investor in October 2025. The reasons cited were thin margins and weak demand.
Yet the robo didn’t disappear. It dissolved. Low-cost model portfolios, digital onboarding and automated rebalancing spread across the industry and became part of how incumbents simply operate.
What the banks shut down was a standalone business. What the industry kept was the capability.
The robo-advisor lost as a product and won as infrastructure.
The recent AI reversals
This wave has produced its corrections faster than the last one.
In 2025, Commonwealth Bank of Australia cut 45 customer service roles after introducing a voice bot. When call volumes rose, it reversed the decision and apologized, admitting it “should have been more thorough.”
Klarna said its AI assistant did the work of 700 agents, and let headcount fall, mainly through attrition. In May 2025 it started hiring human agents again, its CEO conceding that AI had produced “lower quality” support.
Neither is a story about AI failing. Both are stories about a number arriving before the evidence did. The tools did useful work. The error was in the arithmetic: counting the tasks a tool could do, not the work customers still needed done.
The fever and the fiber
Step back and the pattern is older than computing.
Railway mania in the 1840s cost investors fortunes. So did the dot-com telecom boom. But the railroads stayed and carried the economy that followed. The overbuilt fiber, much of it laid and left dark, stayed in the ground and powered what came next.
Hype is priced in the short term. Infrastructure pays off in the long term. Often for someone else.
Apply that to AI in financial services. What lasts will not be a headcount figure. It will be the plumbing. Client and transaction records that everyone can trust. Decisions that can be explained and reconstructed after the fact. Permissions that hold, whoever or whatever is acting. Routine admin and compliance work handed to AI agents that operate inside those limits.
None of that makes a headline. All of it compounds. The fortunes went with the fever. The fiber stayed.
The real question behind the 70%
None of this is an argument for leaving things as they are.
Many advisors are still buried in manual admin and compliance work. AI agents should take it over. Nobody became an advisor to re-key data.
But the 70% raises a narrower question. Was 70% of an advisor’s day really work an agent can take over?
If yes, that says something uncomfortable about how those processes were designed in the first place. A role that is mostly automatable was, for some time, mostly process.
If no, this may become the next rehiring story. We have read a few of those already.
My bet for 2028
So here is where I think the industry lands in two years.
The 70% happens. Just not to the people. Data entry, reconciliations, file notes, suitability paperwork, chasing the missing passport copy: agents do it. The time a professional spends on admin and the time spent with clients swap places. The job inverts, much as the teller’s did.
The audit trail becomes boring. Double-entry bookkeeping was once an innovation. Now nobody thinks about it. By 2028, a complete and explainable record of every decision, human or machine, will be the same: not a feature, just how finance works. Supervisors will read it by machine.
In the UK, the FCA’s own Mills Review has already recommended building an AI-enabled agentic supervisory model. Switzerland’s FINMA is already using one of its own: an AI system that flags issues in inspection documents, with a second model checking the first for hallucinations. The regulators are laying their own fiber.
The shortage nobody is planning for is senior judgment. The first work to be automated is junior work. That work was also the training. Reconciling the spreadsheet, drafting the note, checking the file: tedious, and also how people learned what good looks like.
An industry that automates its apprenticeship will find that it is short of the one thing it cannot automate: people who know which number to check.
By then every firm will have access to similar capability, at a falling price. What clients and regulators will pay for is not the output. It is the person who stands behind it.
What gets built
The question is not how many jobs AI replaces. It is what the industry will have built when the hype burns off.
In two years, the model will be the cheapest thing in the room. The most expensive will be the person who actually knows what it produced.
This article was partially drafted with AI, and reviewed and edited by a human.
For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”.
Subscribe here to be the first to receive my insights.
