The largest section here, and the one where the evidence on entry-level hiring is sharpest and most contested. Finance automates document work well — and document work is most of what a junior analyst was hired to do.
Equity Research Analyst
AI took the data-gathering half of the job; the judgment half, and the relationships, stayed.
Research by Arhan Batur · Sofia Davy · Nadim-Marc El Achkar
Typical employers Goldman Sachs · Morgan Stanley · JPMorgan · Bernstein · Fidelity · BlackRock
What the job is
Equity research analysts cover a set of public companies: building financial models, working through quarterly earnings and filings, and writing notes with buy/sell/hold ratings.
Where AI actually lands in the analyst’s weekFigures are percentages. Share of surveyed sell-side analysts using AI for each task. Data: Christ et al. (2024), as cited in Carrea (2025). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
AI has landed hardest on the data-gathering half of that work. A 2025 survey of finance professionals found AI used most for summarizing text and collecting data, and far less for building forecast models or generating independent recommendations — "AI enhances productivity and efficiency, particularly in data-related tasks, while analysts retain control over judgement-driven responsibilities." Purpose-built platforms now do more than summarize: AlphaSense searches filings and transcripts semantically, Daloopa pulls data from filings straight into a model, and Quartr delivers real-time earnings transcripts. Schroders built its own assistant, Genie, used weekly by thousands of employees. In energy coverage specifically, alternative data has its own stack — Orbital Insight runs imagery from more than 200 satellites to track about six billion barrels of oil storage capacity.
AI enhances productivity and efficiency, particularly in data-related tasks, while analysts retain control over judgement-driven responsibilities.
Carrea (2025), surveying finance professionals on where AI actually gets used
What’s changing
The direction is agreed on; the magnitude is not. A 2026 study of analyst reports found AI-assisted reports linked to more accurate earnings forecasts and stronger market reactions, with investors perceiving the research as "more informative." Goldman Sachs CEO David Solomon has said AI can draft 95% of an IPO filing in minutes — work that once took a six-person team two weeks — and that "the last 5% now matters because the rest is now a commodity." At the far end, Magnetar Capital is reported to be launching an $18 billion fund that skips human analysts entirely, using hundreds of AI agents to "replicate the depth of research and analysis usually provided by fleets of humans" — though that reporting rests on unnamed sources rather than an announcement. Belief about the endpoint splits sharply by seniority: 68.67% of surveyed analysts with 0–5 years of experience thought AI would eventually replace the role entirely.
What a new entrant needs
Expect to spend more time reviewing and correcting AI output than producing it, which means treating it as a first draft to check rather than an answer to trust — the Critical Thinking and Judgement dimension of the AI Readiness framework. The CFA Institute puts it directly: analysts must still verify data, question assumptions, and correct AI errors, because "professional judgment remains essential." Trace every number back to the original filing or transcript; AI can summarize an earnings call convincingly and still get a figure wrong. Treat alternative data as one input among many — as a Point72 executive said of satellite feeds, it is "not magic. It's just another input." The skills that hold value are the ones AI cannot reach: relationships with management teams, sharp questions on earnings calls, and judgment when the data and the story do not line up.
Where the evidence disagrees
Adoption is slowed less by capability than by governance. A 2024 CFA Institute survey found 85% of respondents agreed industry-wide AI standards were needed, and 82% said the absence of those standards was itself what was holding adoption back.
10 sources, located and verified by the contributors
Almost all the research is about sell-side analysts — and this is not that job.
Research by Tim Lin
Typical employers Small newly launched managers · large multi-strategy platforms
What the job is
A long/short equity analyst sources and researches ideas, builds three-statement, DCF, and comparable-company models, and forms buy or short theses on individual stocks. The distinction from sell-side research matters and is usually missed: buy-side analysts build models for internal use by a portfolio manager and are judged on P&L, not on report volume or client servicing.
What analysts think AI will takeFigures are percentages. Sell-side analysts surveyed; the two statements were asked separately. Data: Carrea (2025). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
Adoption skews toward automating the inputs rather than the judgment. A 2024 survey of 190 sell-side analysts found 58% using AI "to a modest extent," concentrated in summarizing text (29%) and collecting data (19%), while independent stock recommendations (3%) and forecast-model building (6%) stayed almost entirely human. Agentic platforms now go further, drafting an earnings review the moment a press release lands.
Certain LLMs not only match, but can exceed human analysts.
Schopf (2025), testing six models against sell-side analysts on SWOT analysis
What’s changing
A 2026 study of investment banks' AI spending found that automation frees analyst time and that such investments "are associated with higher quality and bolder earnings forecasts," alongside expanded coverage of new companies and industries. That reframes the story: AI is not simply cutting headcount, it is changing what the recovered hours buy. Consistent with this, 68.54% of surveyed sell-side analysts believe AI will replace some of their current tasks, but only 26.96% believe it will replace the role. A controlled study of six leading models against human analysts on SWOT analysis found that "certain LLMs not only match, but can exceed human analysts" on specificity and depth — while finding no model could substitute for direct conversations with management.
What a new entrant needs
The skill separating a useful AI user from a dangerous one is verification: cross-checking historicals and stress-testing assumptions against a second source rather than accepting a first output. The survey evidence carries an uncomfortable implication for anyone starting now — junior analysts are far more likely than senior ones (68.67% vs. 0%) to believe AI will replace the role entirely, and the entry-level tasks most exposed to automation are exactly the ones analysts have traditionally used to learn the job. Judgment about which comps matter and what a guidance beat actually means has to be built deliberately rather than accumulated by doing rote work that no longer exists.
Where the evidence disagrees
Nearly all rigorous published research on this question studies sell-side analysts, not buy-side long/short analysts — and the two are judged on different things. No published study surveying buy-side long/short analysts specifically could be found, which is itself a notable gap given how differently the two roles are incentivized.
4 sources, located and verified by the contributor
The calculation is automatable. Deciding which bet to make is not.
Research by Michelle Ni
Typical employers Mutual fund companies · hedge funds · pension funds · insurers · university endowments
What the job is
A portfolio manager decides how a pool of money is invested on behalf of clients, inside rules the client sets about risk, excluded industries, and liquidity.
85% of the respondents believe there is need to set industry-wide standards.
CFA Institute (2024), surveying 200 investment-industry employers — the absence of those standards is itself what respondents said was slowing adoption
How AI is used today
Two very different kinds of AI show up, and they are worth keeping separate. The first is number-crunching: machine-learning models trained on large amounts of historical financial data to help predict performance and rank thousands of possible investments faster than a person could. BlackRock's Aladdin is the well-known example, pairing that prediction with ongoing risk monitoring across a firm's entire book. The second is conversational AI, used to summarize earnings calls and filings, draft first-pass investment write-ups, and read the tone of news coverage.
What’s changing
The clearest shift is in who does the first pass of research. Work that used to fill a junior analyst's day — reading an annual report, summarizing an earnings call, building a comparables table — is increasingly drafted by a tool and then checked by a person. The more uncertain question is whether AI moves from assisting research to proposing portfolio decisions a manager merely approves. Nobody in the industry agrees on the pace, and a 2024 CFA Institute survey of 200 employers found genuine interest mixed with genuine anxiety and broad agreement that the rules for responsible use are not yet in place. Worth remembering: earlier shifts in investing technology, like the rise of computer-driven quantitative strategies, took more than a decade to become normal practice.
What a new entrant needs
Check everything, against the original source, before it reaches a client or influences a real decision — as basic a skill now as building a spreadsheet model. Understand roughly how the models work, not just how to use them: knowing what a reasonable attribution number or benchmark comparison looks like is what lets you catch a tool getting it wrong. Know which calls need a human — anything touching a client directly, or carrying legal or ethical exposure, should go through a person even if AI drafts it. And expect a wider toolkit than Excel and a Bloomberg terminal, including basic Python to work with model output.
Where the evidence disagrees
The calculation-heavy parts of the job — running a factor model, computing a benchmark comparison, breaking down attribution — are the most automatable. Deciding which factor bet to make, or recognizing when an attribution number looks wrong, is a judgment call rather than a calculation, and none of the available sources settle whether AI gets there soon.
3 sources, located and verified by the contributor
Richer pitch materials, produced faster, with measurably worse forecasts.
Research by Jacob Rosenzweig
Typical employers Goldman Sachs · JPMorgan · Morgan Stanley · Evercore · Moelis
What the job is
Entry-level analysts support senior bankers on M&A and capital-raising deals: building DCF, LBO, and comparables models, assembling pitch books, and working live transactions on tight deadlines.
One study, one platform: richer reports and worse forecastsFigures are percentages. Change in analyst reports after FactSet launched its generative AI platform in 2023. Data: Xue, Zhang & Zhu (2025). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
Much of that work is repetitive and document-heavy, which is exactly what generative AI handles well. Big banks are past piloting — JPMorgan's internal LLM Suite is in front of roughly 250,000 employees, and Goldman Sachs has been piloting an assistant built for its bankers. The data platforms analysts already live in have added generative features: FactSet's Mercury, launched December 2023, lets analysts query data in plain English and auto-builds pitch decks from it. The effect shows up in the work product — reports written with AI-platform help carry about 40% more distinct information sources and 25% more analytical methods.
What’s changing
Goldman's own economists estimate AI could automate about a quarter of current U.S. work hours, with roughly two-thirds of occupations exposed to some degree — an economy-wide number, but modeling and formatting sit squarely in the most-exposed bucket. At the junior level specifically, entry-level Wall Street hiring has reportedly been pulled back by as much as two-thirds at some firms even as deal volume recovers. Bank leadership pushes back hard on that framing: Solomon argues adoption raises the bar for what each banker produces rather than shrinking headcount, saying the tools increase "their productivity." The honest read is that everyone agrees the daily tasks are changing fast and nobody agrees what it means for headcount — and both sides are pointing at real current evidence.
What a new entrant needs
Verification skill, not prompting skill, is the differentiator. The FactSet study found that richer AI-assisted analyst reports came with 59% higher forecast errors, because analysts had more information to synthesize rather than less — getting a good answer out of a tool and being able to check it are two different skills, and the second is what protects your credibility. Be fluent in whatever your specific desk already uses rather than AI tools in general. And develop judgment about where AI stops being useful: client relationships, negotiation, and deal judgment remain human work, which the framework names as Human Creativity.
Where the evidence disagrees
Goldman Sachs' own messaging contradicts itself. The CEO says AI adoption will not mean fewer people, while the bank's OneGS 3.0 initiative reportedly asked partners to identify roles AI could make redundant, alongside more than 1,000 actual cuts. The academic evidence on quality is mixed in the same way: AI adoption made analyst reports richer and faster to produce, and also less accurate, because more information is not the same as better synthesis.
5 sources, located and verified by the contributor
Screening a deal fell from a day to an hour — and screening is what juniors learned on.
Research by Emilio Dibildox
Typical employers Bain Capital · Vista Equity Partners · lower-middle-market buyout firms
What the job is
A PE associate screens incoming deals by reading the confidential information memorandum a seller circulates, builds the LBO model, coordinates diligence, drafts the investment committee memo, and monitors portfolio companies after closing. The job is reading a mountain of documents fast and turning them into a numbers-backed argument.
[Generative AI can] bring down the screening time per company from a day to an hour.
Bain & Company (2024), on generative AI at the top of the deal funnel
How AI is used today
The clearest use of AI is at the top of the deal funnel: Bain's 2024 Global Private Equity Report describes a fund that looks at roughly ten deals to find one worth pursuing, spending up to a full day on each first look, and estimates generative AI can "bring down the screening time per company from a day to an hour." Bain also describes tools that turn thousands of online customer reviews into summaries within minutes, and a diligence team that used a GPT-4 API to build rival prototypes in days — showing a target's proprietary AI product could be matched off the shelf, which let the fund make a fast call. In a Bain and StepStone survey of 103 GP professionals, respondents most often named due diligence and deal sourcing as where generative AI delivers the highest internal return.
What’s changing
The most exposed tasks are the ones juniors have always learned on: first-pass screening, data room summaries, comparable-company pulls, and first drafts of memos. Evidence that AI can replace an associate's core judgment is thinner than the headlines suggest — a widely publicized 2024 paper reporting that GPT-4 beat financial analysts at predicting earnings was withdrawn by its own authors in February 2025 after a co-author "identified inconsistencies in the data and analyses." Stanford's revised Canaries study finds employment of 22-to-25-year-olds in AI-exposed occupations "now stands 19% below where it would be" had it kept pace with less-exposed peers, mostly through reduced hiring — but the authors call this descriptive rather than causal, note some divergence predates generative AI, and study exposed occupations broadly rather than PE.
What a new entrant needs
Fundamentals first: you cannot check what you do not understand, and the associate still owns every number that reaches investment committee. Verification is now a core skill, because one hallucinated figure in a model can misprice a deal — in practice that means tracing every AI-extracted figure back to the audited financials or data-room document it came from. Every deal needs an AI lens: Bain argues deal teams should ask of every target whether generative AI helps or threatens the business, as routinely as legal or commercial diligence. And people skills are the moat — in the lower middle market, targets are often founder-owned with thin data and messy books, and earning a founder's trust or spotting what the memorandum leaves out happens in conversation, not in a data room.
Where the evidence disagrees
GPs see strong returns from AI in their own deal work, yet 39% "don't expect AI to have any material financial impact" on their portfolio companies in 2026. Vendor claims need separating from evidence: Hebbia says its software takes analysts from a raw data room to "a comprehensive first-pass memo in hours," but that is marketing, and no independent study verifies it.
5 sources, located and verified by the contributor
Early days, high stakes: a misread fee moves real money to the wrong place.
Research by Karina Advani
Typical employers Bain Capital · KKR · Blackstone · mid-sized PE firms
What the job is
This role manages the firm's own finances rather than deciding which companies to buy: tracking money from investors, calculating the fees the firm earns, preparing quarterly investor reports, and managing the firm's budget. Because these numbers determine how money moves between the firm and its investors, accuracy is the whole job.
What private equity firms say is stopping themFigures are percentages. Barriers to AI integration named by surveyed PE firms. Data: S&P Global Market Intelligence (2026). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
AI use is early. A February 2026 S&P Global survey of private equity firms found integration "remains in early stages," with due diligence — deal work rather than internal finance — the most advanced area at 31% somewhat or fully integrated. The same survey found the biggest barriers were lack of expertise (49%), data privacy (43%), and model accuracy (38%). For finance teams specifically the common uses are pulling data out of investor notices and company reports, drafting routine report text, and summarizing long legal agreements.
They were using AI in certain ways when, in fact, they were not.
U.S. Securities and Exchange Commission (2024), on two investment advisers it fined for overstating their own use of AI to clients
What’s changing
Over five years the role likely shifts from producing numbers to checking and explaining them. Routine data entry and standard reporting are the easiest to hand over, and early evidence suggests entry-level work is already affected: Stanford researchers found employment for 22-to-25-year-olds in highly AI-exposed occupations about 19% lower than if it had kept pace with similar workers in less-exposed jobs, while experienced workers show no comparable gap. Regulators are also watching how firms describe their AI use — in 2024 the SEC fined two investment advisers for claiming they used AI in ways they did not, finding "they were using AI in certain ways when, in fact, they were not."
What a new entrant needs
Check everything: AI can pull the wrong number or misread a fee, and here that means real money moves to the wrong place — the Critical Thinking and Judgement dimension, with PE firms themselves naming accuracy among their biggest AI concerns. Keep private information private; investor and deal details should never go into public tools, and anything a firm says publicly about its AI use has to be true. And recognize that the routine tasks entry-level analysts used to learn on are the ones being absorbed, which makes real experience through co-op matter more rather than less.
Where the evidence disagrees
The Stanford authors caution that "these are descriptive patterns, not causal estimates of the effect of AI" — other factors could explain the employment trends, and tools are changing fast enough that any five-year prediction is a guess.
3 sources, located and verified by the contributor
A seed-stage VC analyst helps decide which very young startups a fund backs: sitting in on founder pitch calls, researching markets and competitors, running early diligence, drafting investment memos, and supporting portfolio companies after investment. Sourcing is still mostly human, and the best survey data contradicts vendor marketing on this point. In a survey of roughly 900 VCs, most deal flow came through networks — professional contacts, referrals from other investors, referrals from portfolio companies — with only about 10% arriving inbound from founders.
Investors only outperformed the algorithm when they had extensive investment experience and managed to suppress their cognitive biases.
Blohm et al. (2022), comparing 255 angel investors with a machine-learning model
How AI is used today
Early diligence and memo drafting is where AI does the most work: pulling from PitchBook, Koyfin, and company materials, then using AI to understand them and draft memos while the analyst directs and checks. Research supports that division of labor — when a large language model scored 138 startup business plans against the same rubric as a panel of VC and angel judges, "LLM-evaluations of business plans are positively correlated with those of experienced investors" (r = 0.52), though the model agreed least with investors on innovation and product quality, which are central to seed investing.
What’s changing
AI will likely become standard for first-pass screening and diligence groundwork while relationship-based sourcing and final judgment stay with people. Confidence should be low, and past predictions are a warning: in 2021 Gartner predicted that "by 2025, more than 75% of venture capital and early-stage investor executive reviews will be informed using artificial intelligence and data analytics." That deadline has passed with no reliable data showing whether it came true. A second shift is that pitch materials may be losing value as signals — in the same study that tested AI screening, investors rated AI-written business plans slightly higher than plans written by founders themselves. If anyone can produce a polished deck, the deck says less about the founder.
What a new entrant needs
Direct and verify AI output rather than just generating it; the analyst's value shifts from producing materials to guiding the tool and catching its mistakes, which means checking every market figure in an AI-drafted memo before a partner sees it. Build a network early, since most seed deals still arrive through relationships. Judge founders, not decks — VCs rank the founding team as the most important factor, and team quality is exactly what AI-polished materials obscure. Understand the technology you are investing in: Carta reports AI companies received 54% of startup investment dollars on its platform in early 2026, up from about 40% in 2025, so analysts need enough fluency to tell a defensible AI product from a thin wrapper. And watch for bias in screening tools trained on historical funding data, which reproduce patterns in who got funded before.
Where the evidence disagrees
The evidence disagrees on what human judgment is worth. One study comparing a machine-learning model with 255 angel investors found that "investors only outperformed the algorithm when they had extensive investment experience and managed to suppress their cognitive biases."
5 sources, located and verified by the contributor
A financial advisor helps individuals and families manage investments, retirement planning, taxes, and long-term goals — work that depends on both financial analysis and client relationships.
Clients serviced per support hireThe firms are comparable, but the gain cannot be attributed to AI alone rather than technology generally. Data: Kitces (2025). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
AI supports the workflow rather than the advice, and the best-documented uses sit on either side of the client conversation rather than inside it. Morgan Stanley introduced AI @ Morgan Stanley Debrief in June 2024 to produce meeting notes with client consent, draft follow-up emails, and file notes in Salesforce — with the release specifying that an advisor can "edit and send at their discretion." A companion Assistant retrieves internal knowledge, and the firm says around "98% of Financial Advisor teams had adopted" it. FINRA's 2026 report found summarizing documents and extracting information the most common generative AI use among member firms. FINRA had already described firms using virtual assistants for basic account questions in 2020 — a reminder that AI entered financial services well before the current wave. These are deployed tools, not predictions.
Because at the end of the day, the Financial Advisor’s service, advice, and relationships with clients — the human touch — remain fundamental.
Morgan Stanley (2024), announcing its own AI assistant
What’s changing
Advisors will likely spend less time assembling research and routine client updates: AI preparing a first draft of a portfolio review, flagging changes in holdings, and answering basic questions. Kitces found that in 2022 firms with one support hire serviced 86 clients, and by 2024 comparable firms handled 111 — though that gain cannot be attributed to AI alone rather than technology generally. The more interesting shift is on the client side: a Betterment survey found roughly 75% of people with capital invested with an advisor use AI occasionally for financial questions, most commonly to understand concepts, research investments, and brainstorm before a major decision. As that report puts it, "AI can help clients gather information, but advisors can help them understand it and apply it to their lives" — which pushes advisors toward interpretation rather than information delivery.
What a new entrant needs
Know which tools your firm allows before entering any client information — many permit only their own internal system. Check the work: AI can summarize a fund report or draft a portfolio update, but the advisor still verifies the fees and the numbers, because a mistake in front of a client is expensive. Finance fundamentals stay load-bearing precisely because they are what let you catch a bad answer — a working grasp of risk, diversification, and the assumptions under a projection is how you notice an implausible result. One practice worth building early: use AI to summarize a public company filing, then work backwards through the filing to find every correction you had to make. Using AI also creates a supervisory duty, not just a personal one. FINRA's rules are "intended to be technology neutral," so a generated client email still needs real review, and knowing when to stop and ask a colleague is part of competence.
Where the evidence disagrees
The sources agree AI makes information work faster but measure different things — FINRA lists actual uses and risks at member firms while the SEC lists what examiners will look at, and neither gives a reliable number for how many advisor jobs AI will replace. FINRA's list also covers securities firms generally rather than personal financial advisors specifically; one contributor's verification narrowed this further, establishing that FINRA's infographic describes member-firm use cases and not advisor-specific adoption rates. Morgan Stanley presents efficiency benefits while FINRA emphasizes accuracy and supervision risks — different vantage points rather than contradictory findings. The limit worth holding onto: a company announcement demonstrates implementation, but it cannot establish industry-wide productivity gains or job losses, and task automation is not proof that a role disappears.
8 sources, located and verified by the contributors
The work is document-heavy and repetitive — the profile AI absorbs fastest.
Research by Dominic Rusu
Typical employers Large institutions with wealth divisions · independent registered investment advisers · boutique private-wealth firms
What the job is
A wealth management analyst works behind the advisor rather than in front of the client: building portfolio models, preparing client-review materials, and doing the investment and market research advisors translate into recommendations. The core of the day is repetitive, document-heavy analytical work — performance summaries, meeting prep, planning projections, research synthesis — which is precisely what current tools do best. At large firms deployment is explicit and firm-wide.
250,000Employees given access to JPMorgan’s internal LLM Suite, about half of whom use it daily
CNBC (2025)
How AI is used today
JPMorgan routes staff through LLM Suite, rolled out to roughly 250,000 employees with about half using it daily, for generating client-ready presentations, analyzing transcripts, comparing documents, and synthesizing data. Consumer tools are banned internally over data-privacy concerns, so everything routes through the controlled environment. Analysts also draw on BlackRock's Aladdin Wealth, research platforms like AlphaSense and Bloomberg, and wealth-specific offerings. Day to day that means AI-drafted client-review packets in minutes rather than hours, scenario modeling and retirement projections, automated portfolio monitoring, and drafted meeting notes.
The operative rule is boring and absolute: client data never goes into a consumer-tier tool. That is why JPMorgan bans ChatGPT and Gemini internally and routes everything through a platform it controls.
What’s changing
The reliable pattern is task automation driving role recomposition, where the analyst's value shifts from producing first drafts to verifying, editing, and challenging machine output. Whether that recomposition also means compression — fewer analysts, or a hollowed-out entry rung — is genuinely uncertain, and much of the optimistic framing comes from parties with a stake in it. The structural risk worth naming: if AI absorbs the junior tasks that once trained future seniors, firms may thin their own talent pipeline even as each quarter looks more efficient. Confidence in any specific five-year headcount forecast should be low.
What a new entrant needs
Almost nothing from the pre-AI skill set is subtracted. Modeling, domain fundamentals, and statement literacy still matter and matter more, because you cannot catch a wrong output without knowing what a right one looks like. What is added is fluency with enterprise AI tools, the judgment to know when a purpose-built tool is required over a general model, and the discipline never to put client data into consumer-tier tools. The framework's Level-2 outcome of systematically evaluating AI-generated content against multiple verification sources is the literal job here rather than an academic exercise.
4 sources, located and verified by the contributor
CNBC. (2025, September 30). Here's JPMorgan Chase's blueprint to become the world's first fully AI-powered megabank.
Johnson, L. (2026, January 7). JPMorgan eschews proxy advisers for internal AI tool. ESG Dive.
An investment operations analyst works in the middle or back office, making sure investment activity is processed accurately after a trade: trade capture and settlement, cash and position reconciliations, corporate actions, pricing and NAV support, reference data, and resolving exceptions.
What “adoption” means depends on what you countFigures are percentages. Data: Alpha FMC (2026); Citi Securities Services (2025). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
Because much of the work is structured and repetitive while errors carry financial or regulatory consequences, it is a natural automation target that still needs human judgment. Adoption is real but uneven. Citi reported that 57% of surveyed firms were piloting generative AI specifically for post-trade, while 67% of institutional investors reported using it for post-trade reconciliation and reporting and for clearing and settlement. Current uses include settlement processing, investment-versus-custodian reconciliation, exception-based processing, continuous monitoring, and Copilot-style assistance. The clearest gains are in clean, high-volume data workflows.
What’s changing
The job moves away from manually processing every item toward supervising automated workflows, investigating exceptions, validating outputs, and maintaining controls. That shift is being accelerated by something that is not AI at all: the SEC's T+1 rules compressed the U.S. settlement cycle and require allocations, confirmations, and affirmations "as soon as technologically practicable and no later than the end of trade date," which raises the value of straight-through processing regardless of AI. Over five years, EY reports 68% of wealth and asset managers expect substantial workforce transformation, particularly in middle- and back-office roles — while 97% currently report minimal headcount change.
What a new entrant needs
The fundamentals still apply: settlement mechanics, reconciliations, portfolio data, controls, and enough market knowledge to recognize when something looks wrong. On top of that, be comfortable using AI productivity tools, working with structured data, and reviewing AI output rather than accepting it. This maps to Critical Thinking and Judgement, which emphasizes verifying AI-driven insights, checking source reliability, and combining output with field-specific expertise. In practice: AI can identify or prioritize a break, but the analyst remains responsible for understanding the transaction, checking the evidence, and deciding whether the proposed resolution makes sense.
Where the evidence disagrees
Adoption depends entirely on how it is measured. Citi reports 57% piloting generative AI in post-trade; Alpha FMC finds only 7% of firms have scaled AI beyond pilots, even though 91% are using or planning to adopt it — and "piloting" is not production. BLS projects an 8% decline for brokerage clerks from 2025 to 2035, but that occupation is only a proxy for investment operations and BLS attributes the decline to productivity-enhancing technology generally rather than AI specifically. The evidence is stronger for changing tasks than for predicting job elimination.
5 sources, located and verified by the contributor
Faster underwriting, and nobody yet knows what the AI collateral is worth.
Research by Anton Schuster
Typical employers Alternative asset managers with credit platforms · bank commercial finance groups · specialty non-bank lenders
What the job is
Asset-based lending analysts underwrite and then monitor loans secured by a borrower's assets rather than its cash flow, so every deal reduces to one question: what would this collateral actually fetch if the loan went bad?
The only balance sheet number I am sure about is my loan balance.
A veteran lender, after the First Brands collapse
How AI is used today
In a technology, AI, and energy-infrastructure practice, that collateral increasingly means a fleet of GPUs, a data-center lease, or a long-term power contract rather than a warehouse of inventory. AI's clearest impact is on speed, not judgment. PwC's 2026 survey of more than 120 credit portfolio managers found AI used most heavily in underwriting — summarizing data rooms, analyzing covenants and cash flow, screening for fraud — while noting that "fully replacing human judgment in credit decisions remains far off." Asset-based lending has its own collateral-side tools: Crediflow AI scans a borrowing base for duplicate invoices and inventory values that do not hold up, and AssetWriter drafts the field-exam report while leaving the examiner's conclusions untouched. As one consultant put it, effective monitoring runs on "a blend of AI and HI" — artificial intelligence paired with human judgment built from real lending experience.
What’s changing
Five years out, the job likely means fewer analysts covering more collateral. A 2026 Stanford study tracking millions of workers through payroll records found no broad economy-wide job losses, but employment for young workers in the most AI-exposed occupations sitting 19% below where their less-exposed peers' trajectory would predict. A harder question follows for this sector: what will the collateral itself be worth? GPUs get pledged as security the way real estate does, yet as one University of Chicago analysis puts it, "nobody yet knows what a used one is worth." That same modeled stress scenario works out to $60–140 billion in realized credit losses falling mostly on private credit funds rather than banks — illustrative rather than predictive. The task mix is shifting faster than anyone's confidence about the assets underneath it.
What a new entrant needs
The most useful thing a new analyst brings is not fluency with any particular tool. It is a settled habit of distrust toward anything unverified, whether it comes from a borrower's spreadsheet or a model's output. First Brands' 2025 bankruptcy revealed more than $3 billion in off-balance-sheet obligations apparently never disclosed to many of its own lenders, several of whom had taken comfort from a clean audit opinion. One veteran lender's reaction afterward: "the only balance sheet number I am sure about is my loan balance." That instinct is what Critical Thinking and Judgement means — evaluating output against multiple independent sources instead of trusting it because it reads smoothly.
Where the evidence disagrees
The same PwC survey found 54% of credit managers already leaning on AI for underwriting but only 16% treating AI-enabled monitoring as a current priority — even though monitoring is exactly where First Brands and Tricolor unravelled. And the Stanford gap shrinks by roughly half once education differences are controlled for; the authors describe their own results as descriptive rather than causal.
5 sources, located and verified by the contributor
Firms say AI will augment auditors; the only measured job losses were junior ones.
Research by Blake Rappaport
Typical employers BDO · Deloitte · EY · Grant Thornton · KPMG · PwC
What the job is
An audit associate performs the fieldwork behind an external audit: reviewing financial documents for accuracy, testing internal controls, documenting processes, and supporting senior staff on engagements.
Premature reliance on GenAI may undermine the foundational skill development essential for junior auditors.
An experienced auditor, quoted in Li & Goel (2026)
How AI is used today
PCAOB staff found that as of July 2024, generative AI use at the large firms "focused primarily on administrative and research activities" such as first drafts of memoranda and researching the firm's own accounting guidance — not substantive audit testing. The Big Four built internal tools on large language models, fine-tuned with firm-specific data and governance controls; notably, none of the available sources names a commercial AI audit platform, and some firms did not permit AI in audit or attest procedures at all, citing data privacy and output reliability. By an April 2026 report, PCAOB staff recorded audit committee chairs describing auditors using data analytics, proprietary platforms, and AI tools to support work such as "testing and reviewing journal entries." A 2026 study based on interviews with 37 Big Four professionals found AI increasingly used for drafting workpapers, researching guidance, and analyzing data — alongside concerns that it threatens audit quality through unverified reliance, reasoning that is hard to inspect, client-confidentiality exposure, and software that cannot exercise professional skepticism.
What’s changing
A 2022 archival study of the 36 largest U.S. audit firms measured AI investment by tracking the share of each firm's workforce holding AI skills across more than 310,000 résumés. A one-standard-deviation increase over three years was associated with improved audit quality and lower client fees — and with accounting staff reductions of 3.6% after three years and 7.1% after four. Separated by seniority, there was no significant displacement at mid-tier or senior levels: the effect was concentrated entirely among junior staff, at 5.7% after three years and 11.8% after four. The significant limitation is that the sample runs 2010–2019, before the current wave of tools. Against that, a 2024 PCAOB update reported that large firms told staff they expected AI to augment rather than replace auditors.
What a new entrant needs
A few experienced auditors in the interview study feared that "premature reliance on GenAI may undermine the foundational skill development essential for junior auditors" — an associate who has never reviewed underlying documents may never learn what an error looks like. Despite that, the firms told the PCAOB that "an engagement team member who uses a GenAI-enabled tool is still responsible for the results and documentation of the work." The same study found juniors using AI to draft communications and get workpapers started while experienced auditors used it in a "value-added capacity," reducing errors during review — juniors prioritizing immediate task relief, seniors prioritizing audit quality.
Where the evidence disagrees
A survey of entry-level audit postings at all six firms found that only KPMG explicitly names AI as a candidate qualification. Deloitte describes associates using "audit tools and technologies that use artificial intelligence," which describes the job rather than the skill set; PwC's AI mention appears to be template language repeated across unrelated postings. None of the six postings asks for human judgment or critical thinking as a means of identifying flawed AI output — a striking gap given that the firms describe exactly that as the associate's responsibility. The survey is one entry-level posting per firm, reviewed in September 2026 and not retained, and a single posting can cover many locations and industries — so read it as a snapshot rather than a systematic sample.
4 sources, located and verified by the contributor
Forecasting under regulation, where most published AI numbers do not transfer.
Research by Madeleine Attar
Typical employers Pfizer · Novartis · small and mid-size biotech firms
What the job is
An FP&A analyst at a biotech company handles budgeting and forecasting across all clinical trial stages, compares monthly and quarterly budgets, and coordinates with accounting and business development.
[Identify] AI-generated financial insights rather than simply process data.
Wylie (2025), on what entry-level finance hiring now expects graduates to be able to identify
How AI is used today
Within smaller firms, AI is increasingly used in cash runway modelling — projecting how long a company can operate before running out of funding — and applied to clinical trial accruals and biotech accounting, historically among the industry's most error-prone areas, by closing gaps left by manual spreadsheet work.
What’s changing
The general expectation is that FP&A shifts away from manual reporting, spreadsheets, and routine variance analysis as AI absorbs those tasks. How confident anyone can be is genuinely debatable: reports show anywhere from 20% to 40% growth in AI use over five years, but some of those figures come from unverified sources or third-party vendors with no real standing. Biotech adds another layer of uncertainty, operating under far more regulation than general financial firms because these companies work with clinically tested medicines that affect patient outcomes — and most growth predictions found were industry-wide rather than biotech-specific.
What a new entrant needs
AI literacy and strategic thinking matter most. One honours thesis on entry-level finance hiring found the expectation is that new employees can identify "AI-generated financial insights rather than simply process data" — which maps onto Critical Thinking and Judgement, where the job is evaluating output for accuracy and gaps before acting rather than just reading it.
Where the evidence disagrees
One market report cites an 18% compound annual growth rate, but a closer read shows that figure describes market size — dollars spent on AI tools — not the rate of adoption or usage growth. Conflating the two is exactly the error this kind of research has to guard against.
3 sources, located and verified by the contributor
An entry-level finance role at a media company running a streaming service: keeping the P&L accurate, supporting month-end close, reconciling cost and cash reports, working with accounts payable on vendor invoices, and building models that answer questions like how many new subscribers a sports package needs to pay for itself.
What finance teams actually hand to AIFigures are percentages. Share of surveyed finance functions using AI for each task. Data: Gartner (2024). Chart drawn for this page; the underlying figures are the contributor’s verified sources.
How AI is used today
Finance teams are adopting fast — in a Gartner survey of 121 finance leaders, 58% said their finance function was using AI in 2024, up 21 points year over year. The top uses line up almost exactly with this workload: automating repetitive processes like invoice handling (44%) and flagging errors and outliers in large datasets (39%), which is essentially what a reconciliation is. The tools are ones analysts already open daily — Copilot is built into Power BI, where it can write DAX formulas, summarize a report page, and draft a new page from a plain-English request. In practice much of the real work still happens by hand with AI speeding specific steps, because every number has to check out to the cent.
58%Of finance functions were using AI in 2024, up 21 points in a single year
Gartner (2024)
What’s changing
The routine, checkable parts of the job shrink: tying out two reports, fixing broken formula references, writing first-draft variance explanations, matching invoices to purchase orders. The analyst's time moves toward checking AI's work and toward real business questions — pricing, churn, whether a content rights deal is worth the money. The hiring implication is the uncomfortable part. A Stanford study using ADP payroll data found employment for 22-to-25-year-olds in the most AI-exposed jobs now 19% below where it would be had it kept pace with less-exposed jobs, while older workers in those same jobs show no gap — driven mostly by reduced hiring rather than firing, and concentrated where AI automates tasks instead of helping workers do them. The World Economic Forum's survey of 1,000-plus employers points the same way, with 40% expecting to cut staff where AI can automate tasks.
What a new entrant needs
Know the finance logic, because you cannot check output you do not understand — if Copilot says two reports tie out, someone still has to prove every total matches to the cent, and in a P&L with large dollar figures a small mistake is a big mistake. Learn Power BI and clean data structure, since Copilot works on top of the semantic model and the person who understands how accounts are mapped is the one who makes it useful. Understand the streaming business specifically: AI can do the math in a subscriber model, but knowing why you compare a rights fee to net revenue after the app-store cut, or why churn should be modelled by cohort, is judgment. Never send an AI number upward until you have tied it out yourself, because AI can make mistakes that look right.
Where the evidence disagrees
Not very confident, on the author's own reading. The Stanford authors say their results are descriptive rather than proof that AI caused the decline, their headline number has moved with each version of the paper (13%, then 16%, now 19%), and they note the gap looks smaller in national survey data.
4 sources, located and verified by the contributor
Someone has to be accountable for a mispriced claim, and an AI cannot be.
Research by Jayden Tom
Typical employers Insurance companies
What the job is
Actuaries take data and build models that evaluate the probability of risk for a specific claim, then use those models to set a suitable premium.
They apply judgment in uncertain situations, shaping the business questions that matter most and influence decisions that drive growth, profitability and resilience.
EY (2026), on what remains the actuary’s work
How AI is used today
AI is being used to streamline and format the data that actuaries once spent extra time on — documentation assembly, standardizing reports, data preparation. The Society of Actuaries' July 2025 AI bulletin highlighted that 77% of US insurers now use AI in claims processing and underwriting, a 15% increase from early 2024. What it has not done is replace the judgment: as EY frames it, actuaries "apply judgment in uncertain situations, shaping the business questions that matter most and influence decisions that drive growth, profitability and resilience."
What’s changing
AI will stay in the actuarial field without replacing the actuary, partly for a reason specific to the profession: there has to be accountability when a claim is improperly produced, and that accountability is regulated such that a human has to be behind it. An AI cannot be held accountable, so there must always be at least one actuary who understands the work the AI does. The Bureau of Labor Statistics projects 22% actuarial employment growth through 2034.
What a new entrant needs
Junior employees used to do the repeatable tasks — data management, documentation — so the baseline now includes knowing how to use AI to do those more efficiently. One practitioner quoted by the American Academy of Actuaries "sees a shift in the role actuaries will play, predicting the possibility of a reduction in the need for traditional and repetitive skills, along with an expansion of the roles of savvy actuaries." That may lower the barrier to entry in some ways, such as not needing to hand-build every model, while raising it in others.
Where the evidence disagrees
Much of the available coverage overlaps heavily: the same underlying information appears across multiple industry sources with different attribution. That could mean the finding is well corroborated, or it could mean actuarial topics simply have thin independent coverage.
4 sources, located and verified by the contributor