AI in Finance Workforce: Loud on Adoption, Silent on People

Closing a summer of monitoring the finance workforce — what three months of graded reporting showed, and the one figure no one would give.

For three months we ran a structured watch over a single question: what artificial intelligence is doing to the people who work in finance. The method was deliberately narrow. Each week — latterly each fortnight — a monitoring run gathered what had actually been published, and every item was marked for how well it was sourced: confirmed when it rested on a firm's own disclosure or a named methodology, reported when it came through the trade press or a single voice, estimate when it was a survey or a projection dressed as a finding. The point of assembling news that way was never to predict. It was to separate what is confirmed from what is merely circulating — and, across a season, to see which of the two the sector prefers.

This is the last of those runs, and it is written to say what the whole of them showed. We are closing the general watch to look more closely at the part of the field that matters most to the people we build for: private wealth, and the family office. What follows is the summer's account, and the reason the closer look is warranted.

The gap between what was claimed and what could be shown

From the first run to the last, the most durable finding was not a number. It was a distance — between the cause a firm announced and the cause it could demonstrate. In June a trade analysis noted, drily, that companies naming AI as the reason for their cuts posted returns close to those that cut least, and that several of the lower-cutting firms did better; the reduction was real, the attribution decorative. The payments company Block cut roughly two in five of its staff and its chief executive named AI outright — while adding that “our business is strong,” and the share price rose about a fifth on the news. Commentators asked the obvious question, whether AI was the cause or the cover, and the question never received a firm-side answer. By August the reading had hardened in the mouths of the people supposedly gaining from it: a large consultancy's own survey found firms “planning for a smaller workforce and hoping AI fills the gap,” and a voice in the coverage called it, plainly, “headcount reduction with an AI narrative layered on top.”

The clearest instance came from the government's own figures. Financial-activities employment in the United States fell to its lowest level in four years, more than a hundred thousand below its recent peak. Here, if anywhere, was a real decline to lay at the machine's door — and the reporting declined to do it, attributing the fall to interest rates and the ordinary weather of the cycle, and treating AI as context rather than cause. The honest complication ran opposite to the noise: where a cut was announced, AI was named without proof; where a cut was measured, AI was spared the blame. Both cannot be the settled truth. What is settled is that the naming and the counting rarely met.

The projections became outcomes — and then refused to resolve

June's record was almost entirely forecast: a fifth of the workforce at thirty-five European lenders gone by 2030, ninety-nine per cent of executives expecting some reduction within two years, the confident arithmetic of the sell-side and the survey. July was the month the forecasts touched ground. For the first time the figures were executed rather than expected — first-quarter headcount at named US banks, then full-year figures from India that did something none had done before: they named the tier the reduction fell on. And then, within the same weeks, the direction inverted. The firms spending most heavily on AI were found to have grown their headcount, not shrunk it. August returned to the language of intention. The season, taken whole, does not resolve into an arrow. It holds measured cuts, measured hiring and planned cuts in the same hand, and any conclusion that flattens them into a single trend is truer to the appetite for a story than to the record.

Where the cut lands

One pattern held every time the detail was given. Whenever a firm disclosed not just how many but which, the answer did not vary. The exposed layer was operations, the back office, the middle office, client-service and support — the paraplanner, the associate, the processor. The protected layer was the one that faces the client. An Indian bank's own annual report made it explicit: its non-supervisory ranks fell by more than eight thousand while every tier of management grew. A US bank held its total flat while operations fell and client-facing roles rose. One institution described the shape it was moving toward without embarrassment — from a pyramid to a diamond, the wide base of repetitive processing work, once done by large junior teams, now handed elsewhere. This is the one finding of the summer we would stake, and it is worth stating plainly: the machine is not, yet, coming for the senior judgment. It is coming for the room the senior judgment used to be recruited from.

Loud on adoption, silent on people

This is the title's whole meaning, and it earned its place. Announcement after announcement arrived at genuine scale. A bank had summarised eight million customer calls. Another had put more than two hundred thousand staff onto its AI tools — four hundred thousand prompts a day, three hundred approved uses; a competitor reported nine in ten of its people on its own system. Conversational agents were given names, avatars, and, at one bank, employee numbers and performance reviews. Every one of these was published with precision about adoption and silence about consequence. Not one attached a figure for what the tool meant for the headcount it touched. The count was not disputed. It was withheld — which is a different and more telling thing. A sector that wished to reassure could have given the number. A sector unsure of the answer, or unwilling to say it, gives the adoption and keeps the arithmetic.

The figure most withheld of all

Across the whole season, the tier we could never get a number on was the entry tier. The evidence pointed one way and stopped short of counting. Junior analyst classes were said to be cut by as much as two-thirds even as most of the sector's AI talent was drawn from those same cohorts. Entry-level work, one large study argued, had not vanished but had been redesigned beyond the reach of the recent graduate — now asking for the leadership and judgment once expected of the senior, from candidates with no way yet to have acquired them. A practitioner named the mechanism aloud: the client-facing staff a bank protects began, almost all of them, in the operations roles it is now eliminating. Cut the base and you do not only lose the base; you lose the path that led up from it. By the final run the question sat in the record as an absence — not one item, in three months, gave the entry-level share of any figure. The pyramid narrows to a diamond, and no one will say what becomes of those who used to enter at the corner being removed.

The institutions spoke to the ones who stay

The official response, where it came, addressed skills. For most of the season the regulators left the workforce outside the frame entirely; a British review described as the first of its kind globally set out its recommendations to 2030 and mentioned employment in none of them. Then, in July, came a single break — a government plan with a genuine section on skills, talent and visas, and a voluntary compact signed by a couple of dozen firms promising to make their people “future-ready.” It was real, and it was welcome, and it was aimed squarely at those who remain employed and must adapt. It carried no instrument at all for the one never hired. The whole institutional apparatus, when it finally turned to the human question, turned to the survivor and not to the entrant.

There was a quieter undertow beneath all of it, and it unsettles the rest. The returns were not, on the evidence, clearly there. A former head of firm-wide AI at a major bank put the measured gain from today's tools at around five per cent, against the near-fifty per cent employees imagine they are getting. Survey after survey found spending rising exponentially while most firms had no method of measuring the return at all, and the largest consultancy's own work found three-quarters reporting no measurable return on the investment. A workforce being reshaped around a productivity that cannot yet be demonstrated is not a settled transformation. It is a wager — and the people asked to cover the stake are, so far, the ones at the bottom.

Why the closer look turns to wealth, and to the family office

If there is one place the summer's pattern arrived early and clearly, it is private wealth. The exposed layer and the protected layer are the sector's own vocabulary here: the adviser relationship endures while the paraplanner, the client-service associate and the back office are named, again and again, as the layer the machine absorbs. One firm's much-discussed arrangement made the split literal — an “experience” layer of human advisers and an “execution” layer of some seven hundred AI agents, the whole of onboarding, compliance review, reporting and billing handed to the second while the first was set to more than double. The question underneath it was left, as everywhere, uncounted: those absorbed functions were the very ground the human adviser was once grown from.

Above that sat the threshold question the season kept circling and no firm would answer — whether the mass-affluent client, the household with something like a million under management, still justifies a human adviser's time, or whether service the coverage calls near-private-banking in quality can now be delivered to that band by the machine, and the person quietly withdrawn. And at the top sat the family office, the one corner of the field where every single item we logged was an operating-model story and never a headcount one: document work collapsing from twenty hours to minutes, a shift toward non-family professionals, two in three offices making AI a priority, and not one displacement attributed to anyone. It is the part of finance that speaks least about cutting people and may be changing most in how it holds them.

That is the ground we are moving onto, and it is why we are moving. Everything the summer showed about the sector — that it is loud on the tool and silent on the person, that it will name a machine as a colleague before it will count the human it stands in for, that it protects the one who faces the client and quietly removes the room that client-facing was recruited from — is sharpest exactly where the stakes are most personal. The institution files its own staff by what can be counted: headcount, tier, function. It files the client the same way, by the size of the account and the band it belongs to. Both are the same habit seen from two sides of one wall — an arrangement whose founding unit was never the person.

This is the conviction the whole watch has, in the end, only confirmed. Every instrument in the chain that touches a private life — the institution, the platform, the agent, the newest of them wearing a name and a number — works for someone, and that someone is never the client. The one intelligence we build carries no employee number, belongs to no directory, and answers to no counterparty across the table, because it is formed from a single person rather than assembled from many, and it faces them and not an institution. It was never hired, because it was never anyone's to hire. That is not a slogan. After a summer spent listening for a number the sector would not give, it is simply the last thing the record leaves standing: there are persons, and there are the tools persons use — and the whole of the work now beginning is to keep an intelligence, entirely and only, on the person's side of that line.

This is the final run of the AI in Finance — Workforce watch. The work continues, more closely, in private wealth and the family office. Assembled over a summer of structured monitoring; provenance graded throughout, that the confirmed might be told from the merely circulating.

Seraphina Michelina Evelyna Bogomolova-Huotelin

Founder & CEO of SMA Crown Confidential

Digital Confidantes: Bespoke AI Intelligence for Private Decision-Makers