OpenAI trains AI with ex-bankers to automate Wall Street grunt work

Junior investment bankers are notorious for working grueling 80-hour weeks. Most of that time isn't spent cutting glamorous deals, but rather wrestling with Microsoft Excel models and fielding endless "pls fix" emails from senior partners.

OpenAI wants to automate all of that away.

According to a report from Bloomberg, the AI powerhouse is running a secretive internal effort dubbed Project Mercury. The company has brought in more than 100 former investment bankers to teach its artificial intelligence how to build complex financial models from scratch.

Wall Street veterans at $150 an hour

The contractor roster includes former employees from Goldman Sachs, JPMorgan Chase, Morgan Stanley, KKR, Evercore, and Brookfield, alongside current MBA candidates from Harvard and MIT. They are recruited and managed through third-party vendors, earning $150 an hour on flexible schedules.

Even the hiring pipeline is largely automated. Applicants go through a 20-minute screening with an AI chatbot that grills them on their resumes, followed by an accounting knowledge quiz and a hands-on modeling test.

Once on board, contractors submit roughly one model per week. Their task is to write straightforward prompts, execute models in Excel for deals like restructurings and initial public offerings, and adhere strictly to industry formatting norms, down to margin sizes and italicized percentage signs. Reviewers check the work, and the polished output is fed directly into OpenAI’s training pipeline.

An OpenAI spokesperson acknowledged that the firm works with third-party domain experts across multiple industries to improve and evaluate model capabilities.

Why it matters

This project shines a light on OpenAI's race to justify its staggering $500 billion valuation. Despite being the world's most valuable startup, OpenAI has yet to turn a profit. To get deep-pocketed enterprise clients in finance, consulting, and law to pay premium enterprise rates, OpenAI cannot rely on generic chatbots. It needs agents that understand the unspoken rules of specialized industries.

For finance professionals, it cuts both ways. Automating the drudgery sounds great on paper, but junior banking grunt work has always served as the apprentice forge where young analysts learn how balance sheets actually tick. If AI takes over the spreadsheets, the traditional pipeline of Wall Street talent could break down.

From an industry standpoint, OpenAI’s playbook shows where the AI data race is heading. Web-scraped text is exhausted; the new frontier is paying high-tier white-collar professionals to distill their craft into bespoke training sets. Still, we think getting Wall Street to actually trust an AI agent on a live multi-billion-dollar merger is a massive hurdle where a single hallucinated cell could spell disaster.