OpenAI launched a financial-services edition of ChatGPT co-built with Morgan Stanley and Evercore, targeting the company research, financial analysis and pitch-deck work that defines junior investment banking roles. The design-partner structure implies compliance architecture was co-developed rather than retrofitted, opening the sell-side research and advisory stack to a general-purpose model. Whether peers follow, and whether entry-level analyst headcount responds, is the number to watch.
OpenAI released a financial-services edition of ChatGPT built with Morgan Stanley and Evercore, aiming at the company research, financial analysis and pitch-deck drafting that occupy the first two years of an investment banking analyst's career. The product is a customized build of OpenAI's enterprise tier, and the two banks served as design partners rather than customers of a finished tool.
"Financial services is one of the highest-value enterprise segments, and the workflows are document-heavy and rules-bound, which is exactly where models perform reliably," OpenAI said in its launch materials. The company did not disclose pricing, seat counts or benchmark scores for the financial-services edition.
The economics are the point. Entry-level investment banking analysts at large U.S. banks earn base salaries of roughly $110,000 to $125,000, with all-in compensation including bonus typically landing between $180,000 and $220,000, according to figures published by Wall Street Oasis and comparable to levels reported by Bloomberg. A single analyst class at a bulge-bracket bank can run 100 to 300 people. Morgan Stanley (MS) and Evercore (EVR) are the first two named institutions to put a general-purpose model inside that workflow at scale, and both have an incentive to say so: Evercore's advisory model depends on revenue per managing director, and Morgan Stanley has spent three years arguing that technology spend is a margin story, not a cost story.
The competitive frame matters more than the product. Microsoft has pushed Copilot into Excel and Outlook, where most of the actual analyst work happens, and Bloomberg has spent years embedding data and analytics directly into the terminal that banks already pay roughly $30,000 per seat for annually. OpenAI's advantage is model quality; its disadvantage is that it owns neither the spreadsheet nor the terminal. A pitch deck assembled in ChatGPT still has to be reconciled against a Bloomberg terminal, a FactSet screen and a bank's own compliance archive.
That last constraint is the real gate. Financial-services deployments require data residency guarantees, audit trails, and model outputs that can be reconstructed for regulators. OpenAI has signed enterprise agreements with banks before, but a design-partner build with two named firms is a different commitment — it implies the compliance architecture was co-developed, not retrofitted. If that architecture holds, the addressable market is the entire sell-side research and advisory stack, not just the drafting layer.
For investors, the read-through splits three ways. Morgan Stanley and Evercore get a productivity claim they can put in front of clients and shareholders before peers have one; Evercore's headcount is roughly 2,500 employees against a market capitalization near $10 billion, so a 10 percent reduction in analyst hours is a visible margin line rather than a rounding error. Enterprise software vendors selling into banks — Microsoft, Bloomberg, and to a lesser degree Salesforce and Palantir — face a narrower wedge for any product whose main value was assembling documents. And the junior-banker labor market, already compressed by a multi-year dealmaking slump, gets a second source of downward pressure on entry-level hiring that has nothing to do with deal volumes.
The counterargument is that banks have automated analyst work before. Excel, PowerPoint templates, and offshore research centers in India and the Philippines each promised to cut junior headcount and instead absorbed the savings into more output per banker. Whether generative models break that pattern depends on whether they reduce hours or raise the volume of work a single analyst is expected to produce. The early evidence from design-partner deployments is not yet public.
What to watch: whether a third and fourth bank names itself as an OpenAI financial-services customer, and whether any of them discloses analyst class sizes for the 2027 hiring cycle. Morgan Stanley's next quarterly disclosure and Evercore's next earnings call are the first scheduled opportunities for either to quantify the effect. Until a bank puts a number on hours saved or seats cut, the productivity claim remains a marketing position rather than a financial one.
This article is for informational purposes only and does not constitute investment advice.
