OpenAI has quietly launched a new initiative called Project Mercury, assembling a team of more than 100 former investment bankers to train its AI systems to automate the manual, time-consuming tasks typically assigned to junior bankers.
According to internal documents obtained by Bloomberg, the contractors — many of whom previously worked at Morgan Stanley, JPMorgan Chase, and Goldman Sachs — are building and refining financial models for transactions such as restructurings and initial public offerings (IPOs).
The project aims to teach OpenAI’s models how to handle complex financial data, perform valuation modeling, and streamline the repetitive workflows that dominate early-stage investment banking careers.
Inside “Project Mercury”
Codenamed Project Mercury, the effort pays contractors $150 per hour to feed and annotate financial models in Excel, following standard industry templates. The goal is to help OpenAI expand the practical, real-world use of AI across sectors like finance, consulting, and technology.
Despite achieving a $500 billion valuation earlier this month, making it the world’s most valuable private company, OpenAI has yet to turn a profit. Project Mercury represents one of its most ambitious pushes into enterprise-grade AI applications designed to create sustainable revenue streams.
“This is OpenAI’s first major foray into domain-specific automation for high-value industries,” a source familiar with the project told Bloomberg. “Finance is the perfect test bed — it’s data-heavy, rules-driven, and demands precision.”
How to Join Project Mercury
Applications for Project Mercury are largely automated. Candidates complete a 20-minute AI interview, followed by tests assessing financial modeling skills and statement analysis. Those who advance must submit one financial model per week for validation.
Notably, the project is not listed on OpenAI’s careers page, but it has already drawn interest from Wall Street veterans and MBA candidates at Harvard University and MIT.
Automation Meets Tradition
Project Mercury targets the “grunt work” of investment banking — repetitive modeling, PowerPoint edits, and report generation — which can consume up to 100 hours per week for junior analysts.
By offloading these tasks to AI, OpenAI envisions a hybrid workflow where bankers focus more on analysis, client strategy, and deal execution.
However, experts warn that full automation could disrupt traditional training pipelines.
“Reading the documents, analyzing them — there’s a process you need to learn,” said Jeanne Branthover, Global Head of Financial Services at DHR Global. “Skipping that step could be detrimental to young bankers’ long-term growth.”
While automation may reduce burnout, it also risks depriving early-career bankers of critical hands-on learning experiences that shape their professional development and judgment.
Industry-Wide Adoption of AI
Project Mercury reflects a broader trend in finance. Leading banks and consulting firms are already deploying generative AI tools to improve productivity:
- Citigroup rolled out Stylus, an AI platform now used by 140,000 employees across eight countries to summarize, compare, and analyze financial documents.
- McKinsey & Company reports that 75% of its 43,000 employees use Lilli, an in-house AI tool that generates presentations, research, and client proposals.
These initiatives suggest that AI is moving from pilot programs to mainstream financial operations, reshaping not only workflows but also the skills and training expected of the next generation of finance professionals.
As OpenAI scales Project Mercury, the implications reach far beyond Wall Street. By teaching AI to model, summarize, and communicate with financial accuracy, the company is effectively training a new kind of digital analyst — one that could soon become standard across the global economy.
