Skip to content
AI Automation

63,000 Banking Jobs, $47bn Of Quarterly Profit: The Year AI Stopped Being An Experiment And Became A Workforce Plan

The numbers finally landed in 2026, and they are not ambiguous. At least sixteen banking and financial services firms have announced a combined 63,000 job cuts this year - an average of about 5.5% of each firm's workforce - with Citigroup pursuing roughly 20,000 reductions in an explicitly AI-led restructuring. In the same period, the six largest US banks reported $47bn of first-quarter profit, up 18%, while cutting around 15,000 roles. Junior analyst intakes at some global investment banks are down by as much as two thirds. This is the clearest evidence yet of what AI automation actually does to a financial institution - and the lesson for everyone else is not the one the headlines suggest.

AlchmAI Editorial13 min read

63,000

Job cuts announced across at least 16 banking and financial services firms in 2026 - roughly 5.5% of workforce on average

$47bn

First-quarter profit across the six largest US banks, up 18%, in the same period as around 15,000 role reductions

~20,000

Reductions Citigroup is pursuing as part of an AI-led restructuring of operations, legal document review and internal workflows

54%

Of US banking jobs that a Citigroup report identifies as potentially automatable

For three years the conversation about AI and financial services jobs was conducted almost entirely in the conditional tense. It could displace this. It might automate that. Analysts published ranges wide enough to accommodate any future, and executives said carefully-worded things about augmentation. In 2026 the tense changed. At least sixteen banking and financial services firms have now announced a combined 63,000 job cuts this year, averaging around 5.5% of each firm's workforce, and in a growing number of cases the restructuring is explicitly attributed to AI rather than to cycle, cost discipline or strategy. Citigroup is pursuing roughly 20,000 reductions in a programme tied directly to automation across operations, legal document review and internal workflows. Morgan Stanley announced around 2,500 in March. Goldman Sachs has cut a cumulative 3,000 across the year.

What makes 2026 different is not the scale - banking has cut more than this in a bad year before - but the coincidence of direction. The six largest US banks collectively reported $47bn of first-quarter profit, up 18%, while eliminating around 15,000 roles. Cuts during a downturn are a familiar story with a familiar ending: the roles come back when revenue does. Cuts during an 18% profit expansion are a different thing entirely. They are a statement about what the work now requires. We build AI and workflow automation for financial institutions, so we see the projects behind these numbers before they become press releases, and the picture from inside is both less dramatic and more interesting than the headline.

Where The Work Actually Went

The publicly-described use cases behind these programmes are consistent across firms, and consistently unglamorous. Automated review of legal documents. Account opening approvals. Credit assessments. Customer-call handling. None of these are the parts of banking anyone puts on a recruitment poster, and all of them share four properties that make them exceptionally well suited to automation: they are high volume, rules-heavy, text-shaped, and they already have a definition of done that a human checks against. That combination is the single best predictor we know of whether an AI automation project in financial services will work.

  • Document-heavy operations. Reviewing contracts, extracting terms, comparing a signed agreement to a standard form. The output is checkable in seconds by someone who knows what they are looking at, which is exactly the condition under which AI review is safe.
  • Onboarding and KYC. Gathering, validating and chasing documents is coordination work rather than judgement work. The judgement - is this client acceptable - stays with a human, and should.
  • Reconciliation and exception handling. Two systems disagree; something has to work out why. Enormous volume, narrow decisions, immediate feedback on whether the answer was right.
  • Client and internal reporting. Assembling the same pack every month from the same six sources, with commentary that varies less than anyone admits.
  • First-line support and internal helpdesk. Answering the same policy question for the thousandth time, with a citation back to the policy.

Notice what is absent from that list. Nothing about deciding whether to lend. Nothing about pricing risk, structuring a deal, or managing a relationship. The automation is landing on the throughput layer, not the judgement layer, and the firms doing this well are explicit about keeping that line clear - partly because it works better, and partly because regulators in both the UK and EU now expect demonstrable human oversight on consequential decisions.

The Junior Analyst Question Is The Real Story

The most consequential number in this year's reporting is not the 63,000. It is that major global investment banks are reportedly reducing incoming junior analyst cohorts by as much as two thirds, as AI takes over pitch-book generation and financial modelling. That is not a cost cut. It is a change to how the industry reproduces itself.

The traditional analyst programme was never an efficient way to produce pitch books. It was an apprenticeship disguised as a production line: you learned the business by building the model badly, having it torn apart, and building it again, and after two years of that you had judgement you could not have acquired any other way. Remove the production line and you remove the apprenticeship with it. The work still gets done - better and faster, by all accounts - but the people who were supposed to become the next generation of managing directors are not getting the repetitions.

“Every institution cutting its junior intake is running an unhedged bet that it can manufacture senior judgement some other way, and almost none of them have said how.”

This is the part of the story we would watch if we sat on a bank board. The productivity gain is real and available now; the cost arrives in 2033 when the bench is thin. A few firms are responding sensibly - restructuring analyst programmes around reviewing and stress-testing AI output rather than producing it, which turns out to be a genuinely good way to learn, because catching a plausible-but-wrong model requires understanding it. That is a design choice, though, not something that happens by itself.

What This Means If You Are Not A Global Bank

Most firms reading this are not Citigroup, and the temptation is to treat these numbers as spectator sport. That would be a mistake, because the economics that produced them scale down more cleanly than almost anything else in enterprise technology. A bank automating document review across 20,000 people is doing the same thing a forty-person asset manager does when it automates its monthly client reporting. The difference is the zeroes, not the method.

There is also a competitive dimension that smaller firms consistently underrate. When the largest institutions cut their cost-to-serve by double digits, they do not simply pocket it. They compete with it - on price, on turnaround time, on the minimum client size they are willing to take. A wealth manager who takes three days to produce a suitability report is competing against one who produces it in twenty minutes and can therefore profitably serve clients half the size. That pressure arrives regardless of whether you have an AI strategy.

The Honest Caveats

Two things are worth holding onto against the momentum of the narrative. First, attribution is genuinely messy. Banks have many reasons to cut staff and exactly one that currently reads as forward-looking to shareholders, and 'AI-led restructuring' is a considerably better headline than 'we over-hired in 2022'. Some meaningful share of the 63,000 would have happened anyway, and honest analysts on both sides of this debate acknowledge it. The direction is real; the precise causal split is not knowable from the outside.

Second, the projections deserve more scepticism than they usually get. A Citigroup report identifying 54% of US banking jobs as potentially automatable is a statement about task composition, not a forecast of employment. Technically automatable and economically worth automating are different sets, and the gap between them is filled with integration cost, regulatory constraint, model risk, and the enduring fact that a great deal of banking is relationship work that clients will pay to receive from a person. Forecasts that pre-tax profits could be 12-17% higher in 2027 than today are plausible directionally and should be read as a range of scenarios rather than a plan.

What We Would Do With This If It Were Our Firm

  1. 01Audit the connective tissue first. List every point where a human moves data between two systems, or checks that one system agrees with another. That list is your automation backlog, and it is almost certainly longer than your strategy deck suggests.
  2. 02Pick the process with the highest volume and the clearest definition of done, and automate all of it. Partial automation of a high-volume process reliably creates more exception work than it removes.
  3. 03Design the human checkpoint deliberately. Not an approval queue with a 99.6% approve rate - a reviewer who can see the evidence, has a real ability to reject, and whose rejections are captured and used to improve the system.
  4. 04Protect the apprenticeship explicitly. If juniors no longer build the model, they must review and break it, and someone senior has to own that as a training objective rather than assuming it happens.
  5. 05Measure cost-to-serve, not headcount. The prize is serving more clients, smaller clients, or faster clients at the same cost. Firms that measure only headcount reduction get the cut and miss the growth, which is the version of this story that ends badly.

The Bottom Line

2026 is the year the AI-and-jobs argument in financial services stopped being speculative, and the evidence is more specific than the headline number implies. Sixty-three thousand announced cuts alongside $47bn of first-quarter profit at the six largest US banks and an 18% rise tells you that firms are not trimming in a downturn - they are re-specifying what the work requires. But the automation is landing on the connective tissue between systems rather than on judgement, which means the opportunity for everyone else is not to brace for impact but to find their own version of the same win: the high-volume, rules-heavy, checkable process that quietly consumes a third of somebody's week. The firms getting real value here are not the ones that bought the most AI. They are the ones that automated one complete process, kept a human genuinely in the loop, measured what changed, and then did it again. That is the work we do as an AI automation and workflow automation agency in London, and on this year's numbers, the cost of waiting has stopped being theoretical.

References & Further Reading

AI AutomationAI Automation LondonWorkflow Automation Agencybanking jobsFintech AI Agency LondonAgentic AIClient-Facing & Internal Tools
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information

Keep reading

All articles