Will AI Replace Investment Banking Analysts?

Introduction

The honest answer is no, AI is not replacing investment banking analysts wholesale, but it is reshaping the job and shrinking the number of analysts banks need to do it. By 2026, every bulge bracket has rolled generative AI into the analyst workflow: JPMorgan put its in-house LLM Suite in front of roughly 250,000 employees, Goldman Sachs piloted an AI assistant for bankers, and Morgan Stanley deployed an assistant to nearly all its advisor teams. OpenAI reportedly hired more than 100 former bankers from JPMorgan, Morgan Stanley, and Goldman Sachs to train its models to build LBO and restructuring models in Excel to Wall Street conventions. At the same time, banks are running leaner analyst classes, smaller than their pre-2023 levels, even as deal pipelines recover. The role is not disappearing. It is being compressed, and the skills that matter are shifting.

This post separates the hype from what is actually happening. It covers what AI genuinely does in an analyst's day today, what it still cannot do, the headcount and class-size debate with real numbers, the historical analogy to Excel and Bloomberg that puts the panic in context, how the role is changing, the realistic three-to-five-year outlook, what to build now, and how to answer the AI question when an interviewer asks it. If you want to understand the role AI is changing, start with the day in the life of an investment banking analyst and the skills needed for investment banking.

The question "will AI replace analysts?" hides two separate questions. One is whether AI can do the tasks an analyst does. The other is whether banks will employ fewer analysts as a result. The answers are different.

On tasks, AI is already strikingly good at large chunks of the analyst grind: formatting pitch pages, pulling and arranging comparable-company data, summarizing earnings transcripts and filings, and drafting first versions of memos and slides. On headcount, banks are not eliminating the analyst role; they are running leaner teams because each analyst, armed with AI, can produce more. As Fortune put it in a December 2025 investigation, much of the AI "takeover" of finance jobs is more hype than reality, with banks using the technology to augment roles rather than gut them (Fortune).

That framing, augmentation rather than replacement, is the consensus among people actually running these systems, and it is the framing you want in an interview. But it would be naive to pretend nothing is changing. Smaller classes mean more competition for analyst seats, and the analyst who cannot use AI well will be at a real disadvantage against one who can.

To judge the threat honestly, you have to know what the tools genuinely do in 2026, not what a headline imagines. The banks have been specific.

Generative AI

A class of artificial intelligence, typically built on large language models, that produces new content (text, code, spreadsheets, slides) in response to natural-language instructions, rather than just classifying or predicting from existing data. In banking, generative AI drafts memos, builds first-pass models and decks, and summarizes documents, all of which were previously manual analyst tasks.

JPMorgan's LLM Suite, an internal platform that routes requests to models from providers including OpenAI and Anthropic, lets employees generate client-ready presentations, compare financial documents, and analyze earnings transcripts. The bank has said the tool saves users an estimated three to six hours a week and expects up to $2 billion in AI-related benefits, against a technology budget that runs to roughly $18 billion to $20 billion a year. Reporting on JPMorgan's deployment described the system drafting investment-banking decks in around thirty seconds and producing first drafts of confidential M&A memoranda for banker review.

JPMorgan first gave employees an OpenAI-powered assistant in 2024, a deployment CNBC reported as one of the earliest large-scale generative AI rollouts on Wall Street (CNBC). Since then, every major bank has moved, and they have made meaningfully different choices about which AI providers to use and which work to automate first, covered bank by bank below.

The most pointed signal is OpenAI's reported "Mercury" effort, which enlisted over 100 ex-bankers, paid around $150 an hour, to teach models to take a senior banker's instructions, build a properly formatted Excel model for a transaction such as an LBO or a restructuring, iterate on feedback, and deliver something close to client-ready. Fortune covered the project as a sign that the entry-level work is squarely in AI's sights (Fortune). The tasks being targeted are exactly the ones that fill an analyst's first two years.

The Tasks AI Handles Well Now

In concrete terms, here is where AI already removes hours from the analyst week:

These map directly onto the most repetitive parts of the job, the parts most analysts would happily hand off. The pitch book structure breakdown shows how much of a book is standard, repeatable architecture, which is exactly the part AI accelerates.

Master interview fundamentals: Practice 1,000+ technical and behavioral questions, download our iOS app for the drills that prepare you for the work AI cannot do for you in the room.

A vague sense that "banks are adopting AI" is not enough for an interview. What signals real awareness is knowing how specific firms are deploying it, which providers they chose, and which work they targeted first. The strategies diverge in instructive ways, and a clear-eyed view of the differences is one of the most useful things a 2026 candidate can bring into the room.

One structural split is worth naming up front: the move from chat-style assistants, where an employee asks a question and gets an answer, to agentic AI, where autonomous systems carry out multi-step tasks with limited human prompting. The frontier in 2026 is agents, and Goldman Sachs is the clearest example.

Agentic AI

AI systems that act autonomously to complete multi-step tasks, planning and executing a workflow rather than just answering a single prompt. Unlike a chat assistant that responds to each question, an agent can, for example, gather data, populate a model, reconcile it against a source, and flag exceptions across several steps. Agentic AI is the 2026 frontier in banking and the form most likely to reshape the analyst workflow.

JPMorgan: The In-House Platform at Scale

JPMorgan's approach is breadth. Its internal LLM Suite routes requests to models from multiple providers, including OpenAI and Anthropic, and has been rolled out to roughly 250,000 employees, making it one of the largest enterprise AI deployments anywhere. Employees use it to generate client-ready presentations, compare documents, and analyze earnings transcripts, and the bank has reported time savings of three to six hours per user per week and anticipated benefits of up to $2 billion, against a technology budget of roughly $18 billion. Reporting on the deployment described it drafting investment-banking decks in about thirty seconds and producing first-draft M&A memoranda for banker review. The strategic choice here is to own the platform and plug in whichever model is best for a task.

Goldman Sachs: Anthropic and the Move to Agents

Goldman Sachs has gone furthest toward autonomous agents, and it chose Anthropic's Claude as a core engine. Goldman embedded Anthropic engineers within its teams for roughly six months to co-develop agents built on Claude, initially targeting two areas: accounting for trades and transactions, and client vetting and onboarding. CNBC reported that the agents support operations covering about $2.5 trillion in assets under supervision, with early tests showing roughly 30 percent faster client onboarding and developer-productivity gains north of 20 percent (CNBC). The bank has signaled it expects efficiency gains and slower future headcount growth rather than immediate layoffs, and it plans to expand the agents into areas including pitch-book creation. This is the agentic future arriving in production, and it is the development a sharp candidate should be ready to discuss.

Morgan Stanley: OpenAI Across the Workflow

Morgan Stanley built its strategy on a deep OpenAI partnership, starting in wealth management and extending outward. Its early advisor assistant reached near-universal adoption across advisor teams. It then launched AskResearchGPT, which lets users mine the firm's vast research library and, by the firm's account, cuts the time to answer a typical client query to roughly one-tenth of what it was (Morgan Stanley). A companion tool, AI @ Morgan Stanley Debrief, summarizes client meetings over Zoom, drafts follow-up notes, and files them to the CRM; advisors in the pilot reported it saves roughly thirty minutes per meeting, which across the firm's roughly one million annual advisor calls adds up to enormous aggregate time savings. Morgan Stanley has also extended OpenAI-powered tools from wealth management into its institutional securities (Wall Street) division (CNBC).

Bank of America and the Rest

Bank of America built an internal generative-AI platform for its Global Markets business that lets sales and trading staff search and summarize the firm's research library, alongside its long-running consumer assistant Erica. Across the industry, the common thread by 2026 is that nearly every major bank has a production deployment, increasingly powered by either OpenAI or Anthropic, or routed through an in-house layer that can call both. Anthropic, having launched a dedicated financial-services product in mid-2025, reported its models in production at JPMorgan, Goldman Sachs, Citi, and others, while OpenAI anchors Morgan Stanley and much of JPMorgan's usage. The providers differ, but the pattern is the same: start with document-heavy, repeatable work, prove the time savings, and expand toward agents.

Nothing defines the analyst job more than building models, the LBOs, DCFs, three-statement models, merger models, and comps that underpin every recommendation. So the question that matters most to an aspiring analyst is not whether AI can format a slide, but how far it has come at the modeling itself. The answer in 2026 is striking: surprisingly capable at the first draft, and still nowhere near trustworthy enough to run unsupervised. If you are new to what this work involves, the LBO modeling guide and the three-statement model walkthrough lay out the tasks these tools are now targeting.

The Tools Building Models Today

A distinct category of software now aims squarely at the modeling workflow, and knowing the names is part of sounding current:

As Axios reported in late 2025, bankers increasingly trust AI to help build models, distill diligence, and generate client-ready materials with far less manual effort than even a year earlier (Axios).

Where the Human Still Has to Drive

Building the mechanics of a model is the part AI now does well; deciding what goes into it is not. The assumptions are where modeling value actually lives: the revenue growth rate, the margin trajectory, the exit multiple, the leverage the credit markets will bear, the synergies that are genuine versus aspirational. Those are judgment calls that require understanding the business, the sector, and the deal, and an AI that confidently fills them with plausible-but-wrong numbers produces a model that looks finished and is quietly dangerous. Someone still has to set the drivers, stress-test the outputs, catch the circular reference or the broken link, and put their name behind the result in front of a deal team and a client.

This is why the modeling skill that matters is shifting. Raw speed at assembling a model from a template is being commoditized; the durable skill is judgment about the inputs and the instinct to know immediately when an output is wrong. The what interviewers look for in Excel modeling tests discussion already reflects this: increasingly the test is whether you understand the model, not whether you can build it fastest. An analyst who can direct these tools and rigorously check their output is worth far more than one whose only edge was being able to build the model by hand.

The reason analysts are being augmented rather than replaced is that a large part of the job is not document production at all. It is judgment, relationships, and accountability, and current AI does none of those reliably.

Automation Versus Augmentation

Automation means a task is performed end to end by a machine with no human in the loop. Augmentation means a machine assists a human who remains responsible for the output. Most AI adoption in investment banking is augmentation: the AI drafts, the banker decides, edits, and owns the result. The distinction explains why productivity rises without the role disappearing.

The Human Core of the Job

Several things stay stubbornly human. Client relationships are the core of banking, and clients pay for the trust, discretion, and counsel of a person, not a chatbot. Judgment under ambiguity, deciding which comparables are truly comparable, whether a projection is credible, or how to position a weak quarter, requires context and taste that models lack. Accountability is non-negotiable: when a model goes to a client or a board, a human has to stand behind every number, and no bank will let an unchecked AI output carry that liability. Negotiation and deal dynamics depend on reading people and improvising, which is not a document-generation problem. And the mentorship and apprenticeship structure of banking, where juniors learn by doing the grunt work under senior review, is itself a reason banks are cautious about removing the bottom rung entirely.

The Skill-Formation Problem

There is also a quieter point about skill formation. If AI does all the grunt work, how does a future managing director ever learn to sense when a model is wrong? Banks know that the tedious early years build the intuition senior bankers rely on, and several are wrestling with how to preserve that learning curve while still capturing AI's efficiency. That tension is a brake on full automation, not just a technical limit.

The clearest way to see the boundary is to lay the tasks side by side.

The pattern is consistent: AI is strong on production and weak on judgment, relationships, and ownership. The analyst job of 2026 is shifting toward the right-hand column.

Beyond the task-level boundary, there are institutional limits that slow full automation in banking far more than in other industries, and they are worth understanding because they explain why the human role persists even as the tools improve.

Model Risk

The risk of loss or error arising from decisions based on incorrect or misused model outputs. In banking it is a regulated discipline: firms must validate, monitor, and govern the models they rely on. Generative AI introduces a new and difficult form of model risk because its outputs are probabilistic and can be plausibly but confidently wrong.

The first limit is hallucination. A general-purpose language model can produce fluent, confident, and entirely fabricated figures. In most contexts that is an annoyance; in a regulated bank it can be a compliance event, because a wrong number in a regulatory filing, a fairness opinion, or a client deck carries legal consequences. Regulators have begun pressing banks for model risk frameworks specifically tailored to AI, and the techniques for reliably detecting and preventing financial hallucinations are still immature.

The second limit is confidentiality. Banks handle highly sensitive, market-moving client information, and feeding that data into a public AI tool risks leaking it to external servers. This is exactly why JPMorgan, Goldman Sachs, and others initially banned employee use of consumer ChatGPT and then built their own in-house, walled platforms like JPMorgan's LLM Suite, where data stays inside the bank's controlled environment. The need for that infrastructure is a real constraint on how fast and how broadly AI can be deployed against the most sensitive deal work.

The most concrete evidence of AI's impact is in hiring, and here the data is genuinely mixed. Several banks are running noticeably leaner analyst classes than before 2023 even with deal activity recovering, hiring fewer juniors per deal while expecting higher output per head, a signal that they expect to need fewer juniors per unit of work. A Citigroup analysis found that a majority of financial-services jobs have high potential for automation, and Goldman Sachs has been reported to be planning sizable layoffs tied partly to AI productivity gains.

But the counter-evidence is just as real. Fortune's reporting concluded that the finance "job takeover" is, so far, more smoke than fire, with banks reorganizing around AI rather than firing en masse, and with hours on the desk stubbornly refusing to fall because saved time gets reinvested into more output and higher expectations. A first-year analyst who can supervise AI to produce what once took three analysts is a productivity story, and productivity stories usually mean leaner teams over time rather than sudden mass unemployment.

For a candidate, the practical implication is competitive, not existential: the seats are fewer and the bar is higher, so demonstrating you can work with these tools is becoming part of being hireable. The investment banking exit opportunities landscape is also shifting, since the same AI pressures apply to the private equity and hedge fund roles analysts traditionally exit into.

It is worth zooming out, because the analyst role most candidates worry about is actually one of the more protected seats relative to other functions in a bank. AI's impact is uneven across the firm, and understanding that distribution clarifies the real picture.

In sales and trading, large parts of execution were automated long before generative AI, and the new tools mostly extend an existing trend toward systematic, technology-driven markets. In equity research, the report-drafting and data-summarization core of the job overlaps heavily with what language models do well, making it one of the more directly exposed front-office roles. In compliance, know-your-customer, operations, and middle-office functions, the work is rules-based and document-heavy, which is precisely what automates most cleanly. The Citigroup finding that a majority of financial-services jobs have high automation potential is driven largely by these high-volume, process-driven roles, not by deal-team advisory work.

Advisory analysts sit in a comparatively defensible spot because so much of the job is judgment, client interaction, and bespoke problem-solving that resists automation. That does not make the role immune, but it does mean the popular framing, that the deal-team analyst is the first domino to fall, gets the gradient backward. The roles most exposed to near-term automation are the standardized, high-volume ones, while the bespoke, relationship-driven advisory seat is more insulated. For a sense of how these functions differ in the first place, the how investment banks make money breakdown maps the revenue lines and the roles attached to each.

There is a structural reason banks are cautious about cutting junior ranks too aggressively, and it is one of the more interesting arguments in the whole debate. The analyst and associate years are how banks manufacture their future vice presidents and managing directors. The tedious work of building models, checking numbers, and assembling decks is not just output; it is the apprenticeship through which juniors develop the financial intuition that senior bankers rely on when they glance at a model and instantly sense something is off.

If AI does all of that entry-level work, the obvious short-term win is efficiency. But the long-term question banks are grappling with is uncomfortable: where does the next generation of senior judgment come from if no one spends years building the foundational skills that judgment rests on? You cannot supervise and sanity-check an AI-built model if you never learned, the hard way, how such a model is built and where it tends to break.

This is also why the learning curve is likely to steepen rather than vanish. Juniors will be expected to develop judgment faster, because they will spend less time on rote production and more time on interpretation and verification from earlier in their tenure. The apprenticeship is being compressed, not eliminated.

It helps to remember that banking has absorbed technology shocks before without the analyst disappearing. Before spreadsheets, junior bankers built financial projections by hand on paper, and pitch books were assembled physically. The arrival of Lotus 1-2-3 and then Excel did not eliminate the analyst; it changed what the analyst did, eliminating the manual arithmetic and freeing time for more analysis, while raising the complexity of what clients expected. The Bloomberg terminal did the same for data: it removed the drudgery of hunting for prices and let analysts do more with the time saved.

The lesson is not that AI is harmless. It is that automation in banking historically raises the floor on what is expected rather than removing the people. When models build themselves, the differentiator becomes the quality of the questions, the assumptions, and the judgment layered on top, which is precisely the part juniors will need to develop faster than before.

Get the complete guide: Download our comprehensive 160-page PDF, access the IB Interview Guide for the technical and behavioral foundation that makes you the analyst who directs the tools instead of competing with them.

If the role is being redefined rather than removed, what does the redefinition look like in practice? The center of gravity is moving from producing to directing and verifying.

In an AI-augmented workflow, the analyst increasingly acts as the editor and quality controller of machine-generated work rather than the originator of every page from scratch. That changes the daily rhythm.

The skills that rise in value in this workflow are not the mechanical ones AI now handles. They are the ability to ask the right question, to spot when an output is wrong, to understand the underlying finance deeply enough to sanity-check it, and to communicate and build relationships. The skills that fall in value are pure speed at formatting and data entry, because those are exactly what the machine does best. The what interviewers look for in Excel modeling tests discussion is instructive here: increasingly, the test is whether you understand the model, not whether you can build it fastest, because the building is becoming automated.

Forecasting precisely is a fool's errand, but the direction of travel is reasonably clear. Over the next several years, expect continued, steady compression of junior headcount through smaller incoming classes rather than dramatic purges of existing analysts. Expect AI tools to become a standard, assumed part of the toolkit, the way Excel and Bloomberg are now, with fluency expected rather than impressive. Expect the analyst role to skew more analytical and verification-heavy and less production-heavy, and expect the learning curve to steepen, because juniors will be asked to exercise judgment earlier now that the rote work that used to build that judgment is automated.

McKinsey estimates generative AI could add the equivalent of $200 billion to $340 billion annually in value to the banking industry, mostly through productivity, which is roughly 9 to 15 percent of operating profits (McKinsey). That scale of value capture guarantees banks will keep pushing adoption. But the same research, and surveys like Accenture's finding that only a minority of banks have moved beyond pilots into production, show that the rollout is gradual and uneven. The transition is a multi-year reorganization, not an overnight event.

For a student or early-career candidate, the strategy is straightforward. Build deep financial intuition, because the ability to know when a number is wrong is the scarce, durable skill. Get genuinely fluent with AI tools, because supervising them well is becoming part of the job description, and increasingly worth pointing to in a recruiting conversation: being able to describe how you used an AI tool to accelerate a real piece of analysis, while still owning and checking the output, signals exactly the mindset banks now want. Invest in communication and relationship skills, since those sit firmly in the human column. And understand the underlying finance, the accounting, valuation, and deal mechanics, well enough that you are the one directing the machine rather than the one the machine makes redundant. Understanding how investment banks make money and why the work commands the pay it does, covered in why investment banking bonuses are so high, grounds all of this in the economics that will decide which roles banks protect.

This topic now comes up directly in interviews, and how you handle it signals whether you think clearly. Interviewers ask some version of "how will AI change banking?" or "are you worried AI will take your job?" to see if you are thoughtful, realistic, and not either naive or panicked.

The strong answer is balanced and specific. Acknowledge that AI is genuinely automating large parts of the production work, name concrete examples like deck drafting and document summarization, and then make the augmentation point: that the role is shifting toward judgment, verification, client work, and ownership, which AI cannot do. Show that you see it as a tool that raises the bar rather than a threat that ends the career, and signal that you intend to be the analyst who uses it well.

Cite Specifics to Stand Out

What separates a good answer from a forgettable one is concrete knowledge. Most candidates will say something vague about AI "changing the industry." You can do far better by referencing what banks are actually doing: that JPMorgan has rolled an in-house platform to a quarter-million employees, that Goldman Sachs has built autonomous agents with Anthropic that already support operations across trillions in assets and is extending them toward pitch-book creation, or that Morgan Stanley uses OpenAI tools that cut research-query time dramatically. Naming the shift from chat assistants toward agentic AI, and noting that the document-heavy junior work is the first thing being automated, signals that you have genuinely thought about the topic rather than reciting a talking point. You do not need to memorize every figure; one or two specific, accurate examples are enough to set your answer apart.

The skills needed for investment banking post pairs well with this, because the skills that survive AI are the ones interviewers were always really screening for.

Will AI replace investment banking analysts? Not in the way the scary headlines suggest. The more accurate picture is a profession being reorganized around powerful new tools: fewer analysts, each more productive, doing less mechanical production and more judgment, verification, and client-facing work. The grunt tasks that defined the first two years are being automated fast, but the parts of the job that actually create value for clients, and the accountability that no bank will hand to a machine, remain human.

For anyone entering the field, that is a manageable reality rather than a reason to flee. The seats are fewer and the bar is higher, so the move is to build the durable skills, deep financial understanding, sharp judgment, strong communication, and genuine fluency with AI, that make you the person directing the technology rather than the person it displaces. The analysts who thrive in the next decade will not be the ones who resisted AI or the ones it replaced. They will be the ones who learned to use it better than anyone else in the room.